Digital twin-based decision support system for the

prospective and the retrospective analysis of an

operating room under uncertainties
Leah Rifi

To cite this version:

Leah Rifi. Digital twin-based decision support system for the prospective and the retrospective analysis
of an operating room under uncertainties. Other [[Link]]. Ecole des Mines d’Albi-Carmaux, 2023.
English. �NNT : 2023EMAC0020�. �tel-04584289�

HAL Id: tel-04584289

[Link]
Submitted on 23 May 2024

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
 THÈSE
En vue de l’obtention du

DOCTORAT DE L’UNIVERSITÉ DE TOULOUSE

délivré par
IMT – École Nationale Supérieure des Mines d’Albi-Carmaux

présentée et soutenue par

Léah RIFI
le 15 décembre 2023

Titre :

Digital twin-based decision support system for the

prospective and the retrospective analysis of an
operating room under uncertainties

École doctorale et discipline ou spécialité :

EDSYS : Génie Industriel

Unité de recherche :
Centre Génie Industriel, IMT Mines Albi

Directeur et Directrice de thèse :

Franck FONTANILI, Maître Assistant HDR, IMT Mines Albi
Maria DI MASCOLO, Directrice de Recherche CNRS, Laboratoire G-SCOP, Grenoble

Autres membres du jury :

Evren SAHIN, Professeure, CentraleSupélec-Université Paris-Saclay, Rapporteure
Vincent AUGUSTO, Directeur de Recherche, Mines Saint-Etienne, Rapporteur
Sondès CHAABANE, Maîtresse de conférences, Université Polytechnique Hauts-De-France, Examinatrice
Vincent CHEUTET, Professeur, INSA Lyon, Examinateur
Virginie GOEPP, Professeure, INSA Strasbourg, Présidente
Cléa MARTINEZ, Maître-Assistante, IMT Mines Albi, Encadrante
Virginie FORTINEAU, Cheffe de projet excellence opérationnelle, GIE Vivalto Santé, Invitée
ABSTRACT

With healthcare demand rising worldwide, hospital services are increasingly needed. Hospitals’
performance is tightly linked to their surgical suite performance. Indeed, the surgical suite is an
important revenue and expense center with over 40% of the hospital’s budget dedicated to it
(Macario et al. 1997) and 60% of the patient coming into the hospital for surgical intervention
(Fugener et al. 2017). This makes it necessary for surgical suites to be efficient.
However, running a profitable surgical suite is quite hard and requires a methodological
approach due to the complexity of its functioning: the diversity of patient pathways, the
multiplicity of professions, the tight link with upstream and downstream wards, the
synchronization of several resources and logistic flows (drug and medical devices), etc. On the
other hand, durations variability and disruptions inherent in medical care like emergency cases
are the main factors and events that degrade the scheduled execution and involve the staff
making decisions frequently to preserve the surgical suite activity in an optimal way. Therefore,
OR planning and scheduling activities are of increasing interest to the scientific community.
In this PhD thesis, we focus on offline operational and online operational levels (Hans and
Vanberkel 2012). This leads us to the following research questions: (1) How can we assess the
robustness and the resilience of the schedule before its execution (prospective way)? (2) How
can we replay the schedule to have feedback and assess the decisions made during its execution
(retrospective way)?
The contribution of this manuscript is threefold: (1) we propose a digital twin-based decision
support system for the prospective and retrospective simulation and analysis of the operating
room schedule execution, (2) we describe a standardized methodology to conceive, build and
implement this tool in any surgical suite, (3) This methodology is applied to an operating room
inspired by the Private Hospital of La Baie (Vivalto Santé group, France), in order to have a
proof of concept allowing to simulate an operating program prospectively and retrospectively.
Keywords: Operating room, Digital twin, Operating room management, Decision support
system, Modeling and simulation, Uncertainties.

i
RESUME

Avec l'augmentation de la demande de soins dans le monde, les services hospitaliers sont de
plus en plus sollicités. Leur performance est étroitement liée à la performance de leur bloc
opératoire. En effet, le bloc opératoire est un important centre de revenus et de dépenses
puisqu'il représente 40% du budget de l'hôpital (Macario et al. 1997), et que 60% des patients
viennent à l'hôpital pour une intervention chirurgicale (Fugener et al. 2017). Il est donc
nécessaire que les blocs opératoires soient efficients.
Cependant, cela est rendu difficile par la complexité de leur organisation due à la diversité des
parcours patients, la multiplicité des métiers, les liens étroits avec les services amont et aval, la
synchronisation de plusieurs ressources et flux logistiques (personnels, médicaments et
dispositifs médicaux), etc. D'autre part, la variabilité des durées et les perturbations inhérentes
à la pratique médicale, comme les cas d'urgence, sont les principaux facteurs et événements qui
dégradent le programme opératoire et impliquent que le personnel prenne de fréquentes
décisions pour maintenir l'activité du bloc opératoire de manière optimale. Par conséquent, les
activités de planification et d'ordonnancement du bloc opératoire intéressent de plus en plus la
communauté scientifique.
Dans cette thèse de doctorat, nous nous concentrons sur les niveaux opérationnels hors ligne
et en ligne (Hans et Vanberkel 2012). Ceci nous amène aux questions de recherche suivantes :
(1) Comment évaluer la robustesse et la résilience du programme opératoire avant son exécution
(dimension prospective) ? (2) Comment rejouer le programme opératoire pour avoir un retour
d'expérience et évaluer les décisions prises lors de son exécution (dimension rétrospective) ?
La contribution de ce manuscrit est triple : (1) Nous proposons un système d'aide à la décision
basé sur un jumeau numérique pour la simulation et l'analyse prospectives et rétrospectives de
l'exécution du programme opératoire. (2) Nous décrivons une méthodologie standardisée pour
concevoir, construire et mettre en œuvre cet outil dans n'importe quel bloc opératoire. (3) Cette
méthodologie est appliquée à un bloc opératoire inspiré de l'Hôpital Privé de La Baie (groupe
Vivalto Santé), afin de disposer d'une preuve de concept permettant de simuler un programme
opératoire de façon prospective et rétrospective.
Mots clés : Bloc opératoire, Jumeau numérique, Régulation, Outil d’aide à la décision,
Modélisation et simulation, Incertitudes

iii
RESUME LONG EN FRANÇAIS

Avec l'augmentation de la demande de soins dans le monde, les hôpitaux se doivent d’être plus
performants. Cela est étroitement lié à la performance de leurs blocs opératoires. En effet, le
bloc opératoire est centre de revenus et de dépenses important : 40 à 50 % du budget de l'hôpital
lui est consacré, ce qui représente 30 % des coûts globaux des soins de santé (Macario et al.,
1997 ; Kaye et al., 2020). Deux études mentionnent que le coût horaire d'un bloc opératoire en
2014 est compris entre 2 000 € et 2 500 € (Mercier et Naro 2014 ; Childers et Maggard-Gibbons,
2018). En France, le coût moyen du parcours d’un patient opéré varie de 1 316 € (chirurgie
courte ou ambulatoire) à 16 653 € (chirurgie lourde). En ce qui concerne le flux de patients, en
2021, en France, pour les services de médecine, de chirurgie et d'obstétrique (MCO pour «
Médecine, Chirurgie, Obstétrique »), 38 % des patients hospitalisés (12 millions) ont été admis
pour une intervention chirurgicale (4,6 millions). 2,8 millions (61%) des patients opérés le sont
en ambulatoire1. Pour toutes ces raisons, il est nécessaire que les blocs opératoires soient
efficaces et rentables.
Dans ce manuscrit, nous faisons la différence entre le bloc opératoire (BO), qui est l'ensemble
du service hospitalier dédié à la chirurgie, et la salle d'opération (SO) ou salle de bloc, qui est
une pièce à l'intérieur de du bloc opératoire où l'intervention chirurgicale est pratiquée. Par
exemple, aux États-Unis, un bloc opératoire compte en moyenne 6 à 7 salles d'opération2.
Tout au long de cette thèse, nous avons eu l’opportunité de travailler avec trois hôpitaux français
différents : l'Hôpital Privé de La Baie (HPB, GIE Vivalto Santé, Avranches, France), le Centre
Hospitalier d'Albi (CHA, Albi, France) et le Centre Hospitalier Intercommunal de Créteil (CHIC,
Créteil, France). Afin de nous assurer que nos travaux de recherche répondent au besoin du
système de soins et de ses acteurs, nous avons formé un comité d'experts avec des directeurs
d’hôpitaux, des [Link], des anesthésistes et des ingénieurs. Grâce à eux, nous nous
sommes assurés de la pertinence de notre question de recherche, et avons défini un périmètre
d'étude en termes de KPI, d’incertitudes, et de décisions opérationnelles.

1 [Link]
2 [Link]

v
Résumé long en français

1. Partie 1 - Contexte et problématiques

Dans cette première partie, nous décrivons le contexte et les problématiques en lien avec notre
projet de recherche, tant du point de vue métier, que du point de vue de la communauté
scientifique.

1.1. Chapitre 1. Contexte de l’étude : le bloc opératoire et ses salles opératoires

Ce Chapitre est divisé en 5 sections. Tout d’abord, nous proposons un aperçu du

fonctionnement d'un bloc opératoire (section 1). Puis, nous décrivons les différents
types de parcours patient possibles pour un patient admis en chirurgie (section 2). Nous
donnons ensuite une présentation approfondie des différentes ressources requises
(section 3) et des indicateurs de performance utilisés (section 4) dans les blocs
opératoires. Enfin, nous concluons ce chapitre par une discussion autour des
problématiques sélectionnées pour cette thèse (section 5). Ci-dessous un court résumé
de chacune des sections.

Veuillez noter que dans ce premier chapitre, nous structurons et synthétisons les connaissances
que nous avons recueillies sur le fonctionnement général des blocs opératoires de nos trois
partenaires. Ces connaissances sont basées sur (1) des observations sur place, (2) des entretiens
avec le personnel et (3) l'analyse de la base de données des logiciels du bloc opératoire. Les
lecteurs experts de l'organisation des blocs opératoires peuvent ne pas lire ce chapitre.
Section 1. Le bloc opératoire est un environnement complexe qui interagit constamment avec
les services internes ou externes de l'hôpital. D’une part, en interne, l'organisation du bloc
opératoire s'articule autour de trois types de processus : les processus décisionnels (planifier et
gérer l'activité), les processus opérationnels (prodiguer des soins aux patients) et les processus
supports (permettre la bonne exécution du parcours patient). D’autre part, le bloc opératoire
est fortement lié à des services externes. Ces services peuvent être (1) à l'intérieur ou à l'extérieur
de l'hôpital, et (2) médicaux ou non médicaux. La performance du bloc opératoire dépend donc
fortement de la façon dont le flux de patients, des informations et des services sont gérés. Dans
notre étude nous nous concentrons sur les flux internes, c’est-à-dire sur les processus qui se
déroulent entre l’arrivée et la sortie du patient au bloc opératoire.
Section 2. L'activité principale du bloc opératoire est de fournir des soins aux patients. Ce
faisant, le service, qui accueille une variété de patients, propose plusieurs parcours différents.
On distingue notamment le type d’admission (en ambulatoire ou hospitalisation), le type de
programmation (électif ou non-électif) et le niveau d’urgence (non urgent, semi-urgent, urgent).
Les patients en ambulatoire sont admis, subissent une intervention chirurgicale et quittent
l'hôpital le jour même, tandis que les patients hospitalisés restent au moins une nuit à l'hôpital.
Les patients en ambulatoire et hospitalisés diffèrent en termes de type de chirurgie, de date
d'admission, de nombre de nuits d'hospitalisation et de parcours du patient. Dans notre étude,
nous prenons en compte à la fois les patients hospitalisés et les patients en ambulatoire ; nous
les modélisons de la même manière car ils suivent les mêmes parcours patients et sont opérés
dans les mêmes blocs opératoires.
La chirurgie des patients électifs n’est pas urgente : ils sont programmés des semaines ou des
mois à l'avance. Les patients non-électifs arrivent à l'improviste et doivent être traités
vi
immédiatement (patient urgent) ou dans les jours qui suivent (patient semi-urgent). Dans notre
étude, nous considérons les patients électifs/non-électifs ainsi que les patients non urgents,
semi-urgents et urgents. Nous les modélisons de différentes manières.
Section 3. Les ressources requises au bon fonctionnement du bloc opératoire peuvent être
matérielles ou humaines. Les ressources matérielles peuvent être achetées ou louées. Elles
englobent les moyens de transport, l'équipement de protection individuelle, les instruments
chirurgicaux (réutilisable après stérilisation), les fournitures chirurgicales (non réutilisables), le
matériel d'anesthésie, les médicaments, le matériel d'imagerie et de visualisation, les systèmes
d'information et de communication et les ressources d'infrastructure.
Pour ce qui est des ressources humaines, le personnel est médical, paramédical, technique et
administratif. Le personnel médical et paramédical prodigue des soins directs au patient, tandis
que l'équipe administrative veille au bon déroulement de la planification et à l'exécution du
planning opératoire. Le personnel technique, bien qu'il ne rentre généralement pas dans les salles
opératoires, contribue au fonctionnement du service. Le bloc opératoire est un service
multidisciplinaire où les individus travaillent ensemble vers le même objectif qui est de fournir
des soins de qualité aux patients mais avec des organisations de travail différentes.
Section 4. La performance d’un bloc opératoire est définie à partir de trois critères : (1) le
nombre d’heures supplémentaires (sa diminution reflète de meilleures conditions de travail pour
le personnel), (2) les temps d'attente du patient (des temps d'attente plus courts satisferont les
patients, et (3) le taux d’utilisation des salles d'opération (des taux d'utilisation élevés assurent
une efficacité organisationnelle rentable). Les taux de débordement, d’enchainement, de
surutilisation, et de sous-utilisation sont des indicateurs qui viennent en complément du taux
d’utilisation.
D’autre part, l'organisation d'un bloc opératoire est dite robuste si elle peut maintenir son niveau
de performance sans s’adapter, malgré des perturbations aléatoires (exemple : les incertitudes
sur les durées opératoires). L'organisation d'un bloc opératoire est résiliente si après qu’une
perturbation aléatoire telle l’arrivée d’urgence lui ait faite baisser en performance, elle peut
revenir au niveau de performance antérieur.
Section 5 – Synthèse. Le bon fonctionnement du bloc opératoire nécessite de synchroniser les
ressources à l'intérieur et à l’extérieur du bloc avec l'exécution du parcours du patient. Par
conséquent, l'organisation d'un bloc opératoire performant repose sur la communication et la
coordination des ressources (1) en son sein, et (2) entre le bloc opératoire et le monde extérieur.
Ce point est particulièrement complexe pour les raisons que nous présentons maintenant.
Tout d'abord, le personnel du bloc opératoire est composé de ressources humaines variées
(personnel médical, personnel paramédical, personnel technique, personnel administratif) qui
ont toutes besoin d'un accès rapide à des ressources matérielles adéquates (fournitures, locaux,
stockages...). D’autre part, en raison des interactions requises entre le bloc opératoire et les
unités extérieures, les dysfonctionnements apparaissant dans l’un ont un impact sur l’autre, et
vice-versa. Tout retard, absence ou erreur peut entraîner une perturbation et entraver l'exécution
du planning opératoire. En effet, si la bonne ressource n'est pas disponible au bon endroit et au
bon moment, le parcours patient s'arrêtera. Cela vaut autant pour un.e [Link] que

vii
Résumé long en français
pour un.e infirmier.ère. Dans ce contexte, le personnel doit avoir une bonne communication,
une bonne collaboration, une bonne coordination et le respect des processus établis.

1.2. Chapitre 2. Revue de littérature et travaux similaires

Ce Chapitre est divisé en 4 sections. Dans un premier temps, nous abordons le problème
de planification et d’ordonnancement des bloc opératoires (section 1). Ensuite, nous
discutons des méthodes de management prédictive et réactive des perturbations au BO
(section 2). Ensuite, nous proposons une courte bibliographie sur l’utilisation du jumeau
numérique et de la simulation à évènement discrets dans le domaine de la santé – et plus
particulièrement au bloc opératoire (section 3). Enfin, nous clôturons le chapitre avec
une synthèse (section 4). Ci-dessous un court résumé de chacune des sections.

Section 1. Il existe quatre niveaux de décision en matière de planification et d'ordonnancement

au bloc opératoire. Premièrement, le niveau stratégique (long-terme) traite de la planification
et de l'allocation de la capacité et des problèmes de combinaison de cas. Deuxièmement, le
niveau tactique (moyen-terme) s'attaque au problème du calendrier de la chirurgie principale
(MSSP). Troisièmement, le niveau opérationnel hors ligne à court terme lié au problème de
planification de la chirurgie (qui peut être divisé en problèmes de planification avancée et
d'allocation). Quatrièmement, le niveau opérationnel en ligne en temps réel consiste dans le
problème d'exécution du calendrier.
Section 2. La gestion des perturbations au niveau opérationnel se divise en deux enjeux
principaux : (1) la gestion prédictive des perturbations : les perturbations sont anticipées et
traitées avant qu'elles se produisent ; (2) la gestion réactive des perturbations : les perturbations
sont traitées après qu’elles se sont produites.
Section 3. La simulation à évènements discrets (SED) est une technique de modélisation qui
permet de modéliser le comportement d'un système sous la forme d'une séquence d'événements
dans le temps. Elle est notamment utilisée dans les blocs opératoires pour évaluer la
performance de l’organisation, ou pour soutenir les processus de planification et de
programmation. Même si la littérature scientifique regorge d'études sur le jumeau numérique
(JN), il n'en existe toujours pas de définition unique. On retrouve cependant certaines
caractéristiques communes d’une définition à l’autre : un JN doit (1) être la réplique virtuelle
d'un jumeau physique désigné, (2) être capable de simuler le comportement du jumeau physique
en temps réel ou presque, et (3) permettre une communication bidirectionnelle avec le jumeau
physique. Bien que moins développés que dans l'industrie, les jumeaux numériques sont
également utilisés dans les soins de santé. On retrouve des JN de personnes, d’organes et de
parcours patient.
Section 4 – Synthèse. Après avoir étudié les problématiques du terrain et les problématiques
scientifiques liées au bloc opératoire, nous avons décidé de focaliser notre étude sur la gestion
des perturbations aléatoires au niveau opérationnel. Notre question de recherche générale est la
suivante : « Comment améliorer et maintenir la performance d’un bloc opératoire soumis à des
incertitudes ? ». Nous proposons 5 questions de recherche plus spécifiques :
Q1. Comment anticiper les déviations avant l’exécution du programme ?
Q2. Comment évaluer la qualité du management prédictif des déviations ?

viii
 Q3. Comment évaluer la qualité du management réactif des déviations ?
Q4. Comment identifier la cause racine des manques de performance ?
Q5. Comment entrainer les régulateurs au management des déviations ?
Dans le cadre de notre étude, nous étudions la performance du bloc opératoire via le taux
d’utilisation et le taux de débordement des salles opératoires, ainsi que le temps d’attente des
patients au bloc opératoire. Nous prenons en compte 3 types d’incertitudes : l’arrivée de patients
non-électifs, la variabilité des durées d’activité et la disponibilité des ressources. Nous nous
concentrons sur un type de décision : la programmation des arrivées de patients non électifs.
Sur la base de notre analyse documentaire et de nos observations sur le terrain, nous avons
décidé de proposer un système d'aide à la décision hors ligne basé sur un jumeau numérique
utilisant la simulation à évènements discrets. Cette outil devrait nous permettre d'améliorer les
décisions prises au niveau opérationnel en ligne via une analyse prospective et rétrospective du
programme opératoire.

ix
Résumé long en français

2. Partie 2 – Proposition d’une méthode et d’un outil

Dans la Partie 1, nous avons présenté le contexte et la problématique de notre recherche. Dans
cette Partie 2, nous décrivons notre proposition de solution.

2.1. Chapitre 3. Proposition de solution

Ce Chapitre est divisé en 5 sections. Nous décrivons une méthodologie d’analyse

prospective du programme opératoire prévisionnel (section 1), une méthodologie
d’analyse rétrospective du programme effectivement réalisé (section 2), une
méthodologie pour la formation du régulateur de BO à la gestion des perturbations
(section 3), une méthode standardisée pour construire un jumeau numérique de BO
(section 4). Nous concluons avec une synthèse (section 5). Ci-dessous (voici) un court
résumé de chacune des sections.

Dans ce résumé de chapitre, nous utilisons les termes suivants :

- programme prévisionnel ou programme opératoire prévisionnel (POP) : programme
théorique disponible la veille de la journée opératoire. Il contient des interventions non
urgentes et semi-urgentes.
- programme réalisé ou programme effectivement réalisé (PER) : programme tel qu’il
s’est vraiment déroulé lors de la journée opératoire. Il contient des interventions non
urgentes, semi-urgentes et urgentes.
Le régulateur se sert du POP comme guide au court de la journée opératoire bien qu’il sache
que le PER n’en sera pas identique ; les perturbations aléatoires (incertitudes sur les durées,
arrivées d’urgences) creusent en effet un écart entre les deux programmes.
Section 1. La méthodologie d'analyse prospective du programme opératoire prévisionnel a
pour objectifs d'anticiper les perturbations avant l'exécution du POP (Q1), et d'évaluer la qualité
de la gestion prédictive des perturbations (Q2). Cette méthodologie correspond à notre
contribution #1. Les étapes de l’analyse prospective sont les suivantes : évaluer pour le POP :
(1) la faisabilité, (2) la performance, (3) la robustesse, (4) la résilience et (5) la meilleure stratégie
de programmation des cas non-électifs.
Section 2. La méthodologie d'analyse rétrospective du planning effectivement réalisé a pour
objectifs d'évaluer la qualité de la gestion réactive des perturbations (Q3), et d'identifier si le
manque de performance provient de décisions hors ligne ou en ligne (Q4). Au cours de cette
analyse, nous souhaitons notamment fournir un retour d'informations sur le déroulement du
POP qui a conduit à l'obtention du PER. Cette méthodologie correspond à notre contribution
#2. Les étapes de l’analyse rétrospective sont les suivantes : évaluer pour le PER : (1) la
performance, (2) l’impact de la stratégie de programmation des cas non-électifs effectivement
choisie, et (3) les causes racines des écarts à la performance.
Section 3. Dans le Chapitre II, nous avons décrit un outil d'aide à la décision capable de simuler
l'exécution du programme opératoire dans un environnement stochastique (variabilité de la
durée et arrivées supplémentaires de cas non-électifs) tout en suivant une stratégie de gestion
des perturbations spécifiques. Nous proposons d'utiliser ce même environnement virtuel pour
permettre au régulateur de tester ses propres stratégies afin de le former. Cela nous permet de

x
répondre à notre question de recherche « Comment former les régulateurs à la gestion des
perturbations ? » (Q5). Cette méthodologie correspond à notre contribution #3.
Section 4. Une fois ces trois méthodologies présentées, nous proposons une méthode
standardisée en 5 étapes pour construire le jumeau numérique d'un bloc opératoire à l'aide de la
simulation d'événements discrets. Les étapes sont les suivantes : (1) recueillir les données, (2)
traiter et analyser les données, (3) construire un modèle déterministe, (4) implémenter des
incertitudes, (5) développer les fonctionnalités de l’outil d’aide à la décision. Cet outil modélise
l'exécution d'un programme prévisionnel ou exécuté, dans un environnement déterministe ou
stochastique, tout en respectant les ressources et les contraintes du patient, en appliquant des
stratégies de programmation d’interventions non-électives, et en calculant des KPIs (section 4).

2.2. Chapitre 4. Modélisation et simulation du déroulement du programme

opératoire
Dans ce Chapitre divisé en 3 sections, nous décrivons comment nous modélisons et simulons
l’exécution du programme opératoire, qu’il soit prévisionnel ou réalisé. D’abord, nous
proposons différentes manière de modéliser le bloc opératoire en fonction des données
disponibles (section 1). Ensuite, nous proposons une application de cette méthode sur le cas
d’un de nos hôpitaux partenaires (section 2). Nous concluons avec une synthèse (section 3). Ci-
dessous voici un court résumé de chacune des sections.
Section 1. Les processus du bloc opératoire (PBO) sont complexes et nécessitent des ressources
humaines et matérielles variées à différents moments et lieux. Puisqu’il est courant de ne pas
avoir accès à tous les jalons temporels inclus dans le processus (exemples : heure d’arrivée et de
départ du patient au bloc, heure d’entrée et de sortie de salle du patient, etc.), nous proposons
de définir deux modèles de PBO : (1) un modèle agrégé basé sur les données horodatées
disponibles dans le logiciel de bloc, et (2) un modèle détaillé basé sur les données horodatées
disponibles dans le logiciel de bloc et sur nos observations sur site. Le modèle agrégé comprend
naturellement moins de jalons et d’activité(s) que le modèle détaillé. Nous modélisons
l'exécution du programme prévisionnel à l'aide du processus détaillé et nous
modélisons le programme réalisé à l'aide du processus agrégé.
D’autre part, pour modéliser de manière exhaustive les processus du bloc opératoire, nous avons
besoin d'informations sur les ressources utilisées : quelle ressource est nécessaire ? Quand ? Où
? Combien de temps ? Cependant, comme pour les horodatages, les informations relatives aux
ressources ne sont pas toujours disponibles dans la base de données de la salle d'opération. Par
conséquent, nous proposons de définir deux types de contraintes sur les ressources à respecter
lors de la modélisation des PBO : (1) les contraintes strictes sont les contraintes
(éventuellement incomplètes) imposées par les valeurs des jalons temporels de la base de
données sur les ressources matérielles et humaines, et (2) les contraintes flexibles qui sont les
contraintes imposées de la base de données sur les salles opératoires seulement. Nous utilisons
des contraintes flexibles pour modéliser le déroulement du programme réalisé, et des
contraintes strictes pour modéliser l’exécution du prévisionnel.
Section 2. Nous proposons une application concrète des modèles et contraintes théoriques
proposés dans la section 1.

xi
Résumé long en français
Section 3 - Synthèse. Nous modélisons l'exécution du programme prévisionnel à l'aide du
processus détaillé et nous modélisons le programme réalisé à l'aide du processus agrégé. Nous
utilisons des contraintes flexibles pour modéliser le déroulement du programme réalisé, et des
contraintes strictes pour modéliser l’exécution du prévisionnel.

2.3. Chapitre 5. Calculer les durées

Ce Chapitre est divisé en 5 sections. Dans un premier temps nous proposons et

appliquons une méthode pour corriger les données d’horodatage initialement
disponibles dans la base de données (section 1). Ensuite, nous calculons les durées de
chaque activité du parcours patient à partir des jalons corrigés (section 2). Puis, nous
utilisons ces durées historiques afin de calculer un maximum de jalons initialement non
relevés dans la base de données (section 3). Enfin, nous combinons toutes ces données
temporelles du programme réalisé pour calculer les jalons et les durées du programme
prévisionnel (section 4). Nous concluons avec une synthèse (section 5). Ci-dessous voici
un court résumé de chacune des sections.

Pour information, les données d’horodatage (ou jalons temporels) initialement disponibles dans
la base de données correspondent au programme réalisé. En effet, les jalons prévisionnels ne
sont pas préservés dans le logiciel de bloc. Des exemple de jalons sont : heure d’arrivée du
patient au bloc opératoire, heure d’incision, heure de suture, etc.
Section 1. Nous proposons et appliquons une méthode pour corriger deux types d’erreur dans
les données d’horodatage initialement disponibles : les incohérences sur (1) les jalons au sein
d’un même parcours patient (exemple : le patient se fait inciser avant d’entrer en salle), et (2)
l’utilisation des salles opératoires (exemple : deux patients sont en salle au même moment).
Section 2. Nous utilisons les jalons restant dans la base données pour calculer les durées des
activités du programme réalisé. Ces durées permettront : (1) de simuler l'exécution du
programme réalisé dans notre DT-DSS, (2) de calculer les durées prévisionnelles discrètes, et
(3) de calculer les durées prévisionnelles stochastiques.
Section 3. Nous combinons les jalons temporels corrigés (section 1) et les durées discrètes
calculées (section 2) pour en déduire les jalons manquant de la base de données tout en
respectant les contraintes sur l’utilisation des blocs opératoires et la cohérence des jalons d’un
même parcours patient. Pour cela, nous implémentons les étapes suivantes :
- Corriger les jalons aux extrémités du parcours patient : entrée au bloc, sortie du bloc.
- Corriger chaque parcours patient de manière indépendante : incision, suture et entrée
en salle de réveil.
- Corriger les parcours des patients en se basant sur l’utilisation des salles opératoires :
entrée en salle, sortie de salle.
Les étapes sont toutes répétées jusqu'à ce que le nombre de jalons corrigés cesse d'augmenter.
En effet, puisque les horodatages sont dépendants les uns des autres, en corriger un peut
permettre d’en corriger un autre.
Section 4. Les sections précédentes nous ont permis de corriger, de compléter l’horodatage du
programme réalisé. Nous allons maintenant utiliser ces données historiques pour déterminer

xii
l’horodatage du programme prévisionnel (jalons et durées), ainsi que pour calculer des durées
prévisionnelles stochastiques. Cette tâche est particulièrement complexe à cause de la non-
standardisation des noms de chirurgie dans notre base de données. En effet, cette information
est habituellement utilisée avec l’ID du/de la [Link] pour estimer les durées
prévisionnelles. Nous proposons et appliquons différentes méthodes de calculs en fonction des
données disponibles dans la base de données.
Section 5 – Synthèse. Dans cette section, nous avons proposé et illustré une méthodologie
permettant de corriger et de compléter les horodatages et les valeurs de durée des plannings
réalisés et prévisionnels extraits de la base de données réelles pour alimenter le jumeau
numérique.

2.4. Chapitre 6. Modéliser et programmer les interventions non-électives

Ce Chapitre est divisé en 3 sections, Nous décrivons la modélisation des arrivées de cas
non-électifs (section 1) et la modélisation de leur programmation (section 2). Nous
concluons avec une synthèse (section 3). Ci-dessous voici un court résumé de chacune
de ces sections.

Section 1. Nous souhaitons modéliser l’arrivée des cas non-électifs lors de la simulation de
l’exécution du programme prévisionnel, et lors de la simulation du programme réalisé. Nous
appelons « programme initial », le programme tel qu’il est avant le lancement de la simulation.
Notez que :
- Un programme prévisionnel initial contient des interventions non-urgentes
et potentiellement des interventions non-électives semi-urgentes.
- Un programme réalisé initial contient des interventions non-urgentes ainsi
que potentiellement des interventions semi-urgentes et urgentes.
Premièrement, dans le cadre de l’analyse prospective, nous souhaitons modéliser l’arrivées des
cas non-électifs au cours de l’exécution du programme prévisionnel à deux dates distinctes : (1)
lors de la réunion de programmation : dans ce cas, le programme prévisionnel initial ne contient
que des interventions non urgentes), et (2) la veille de la journée opératoire : dans ce cas, le
programme prévisionnel initial contient des interventions non urgentes et potentiellement semi-
urgentes.
Deuxièmement, afin de réaliser notre analyse rétrospective, nous souhaitons reproduire
l’arrivée des cas-électifs telle qu’elle était dans la réalité. Il n’y a pas besoin de modéliser
d’interventions non-électives additionnelles.
Enfin, pour l’outil d’entrainement des régulateurs, nous souhaitons proposer un environnement
virtuel pour permette de (1) simuler un programme prévisionnel ou réalisé, (2) d’inclure ou pas
des cas non-électifs du programme initial, et (3) de rajouter ou pas des cas non-électifs
additionnels.
D’une part, nous modélisons les cas non-électifs additionnels par (1) une heure d'entrée dans la
salle d'opération, (2) la liste des chirurgiens capables d'effectuer la chirurgie, (3) la liste des salles
d'opération dans lesquelles le cas peut être programmé, (4) un type d'anesthésie, (5) le niveau
d'urgence du cas et (6) la durée des activités du parcours (du) patient. Ainsi, avant d'être

xiii
Résumé long en français
programmée, une intervention non-élective additionnelle n'a pas de chirurgien, d'anesthésiste
ou de salle d'opération attitrés. D’autre part, nous modélisons les scénarios d’arrivées de cas
non-électifs additionnels par (1) le niveau d'urgence des cas, (2) le nombre de cas entrants et (3)
la fenêtre d'heures d'arrivée des cas au bloc opératoire.
Section 2. Nous proposons 6 stratégies de programmation pour les interventions non-électives
: (1) maintenir la programmation initiale (si on simule un programme réalisé), (2) first in first
out, (3) best fit, (4) worst fit, (5) programmation manuelle et (6) file d'attente.
Section 3 – Synthèse. Notre DT-DSS permet de modéliser et de programmer des interventions
non-électives, qu’elles soient urgentes ou semi-urgentes.

xiv
3. Partie 3 – Preuve de concept
Dans la Partie 1, nous avons présenté le contexte et la problématique de notre recherche. Dans
la Partie 2, nous avons décrit notre proposition de solution. Dans cette troisième et dernière
partie, nous fournissons une preuve de concept basée sur un cas d'étude.

3.1. Chapitre 7. Présentation du cas d’étude

Le Chapitre 7 est divisé en 3 sections. Tout d'abord, nous décrivons les établissements
de santé de nos partenaires (section 1). Ensuite, nous discutons des bases de données
fournies par chacun d’eux, nous expliquons comment nous avons structuré et traité ces
données, et nous présentons une méthode pour sélectionner un cas d'étude pertinent
(section 2). Puis, nous mettons en œuvre cette méthode et présentons le cas d'étude que
nous avons retenu pour notre recherche (section 3). Nous concluons avec une synthèse
(section 4).

Section 1. Nous présentons nos trois partenaires hospitaliers : l'Hôpital Privé de La Baie (GIE
Vivalto Santé, Avranches, France), le Centre Hospitalier d'Albi (Albi, France) et le Centre Hospitalier
Intercommunal de Créteil (Créteil, France). La combinaison de ces trois sites nous permet d’avoir
accès à des hôpitaux : des secteurs publics et privé, régionaux et universitaires, avec différents
volumes d’activité, des services de chirurgie ambulatoire, une maternité, une unité de soins
intensifs, des services d’urgence, différentes tailles de bloc opératoires, différentes populations
de patients dans les services d’urgences.
Section 2. Nous proposons une méthode pour choisir un cas d’étude permettant de : (1) simuler
le déroulement de programmes prévisionnels et réalisés pour une journée entière, (2) tester
la robustesse d’un programme en simulant son exécution dans un environnement où les durées
sont stochastiques, (3) tester la résilience d’un programme en simulant son exécution dans un
environnement où il y a des arrivées stochastiques de cas non électifs, et (4) être représentatif
d'une journée opératoire normale.
Section 3. Nous présentons pour le programme prévisionnel et le programme réalisé du cas
d'étude : le planning des vacations, le diagramme de Gantt de l'activité des salles opératoires, la
description des interventions non-electives, et le calcul des KPI.
Section 4 - Synthèse. Dans ce chapitre, nous avons extrait et présenté notre cas d’étude.

3.2. Chapitre 8. Application des analyses prospective et rétrospective.

Le Chapitre 8 est divisé en 4 sections. Dans un premier temps, nous présentons les
expérimentations à réaliser pour appliquer les analyses prospective. (section 1). Ensuite,
nous décrivons les résultats obtenus pour l’analyse prospective (section 2) et pour
l’analyse rétrospective (section 3). Enfin, nous proposons une synthèse de chapitre qui
fait le lien entre les deux analyses et qui explicite les actions concrètes qui pourraient être
prise après l’application de notre démarche d’aide à la décision (section 4). Ci-dessous
un court résumé de chacune des sections.

xv
Résumé long en français
Section 1. Nous présentons les différentes expérimentations nécessaires à l’application des
analyses prospective et rétrospective. Pour ce faire nous décrivons la configuration des
paramètres du modèles.
Sections 2 et 3. Nous appliquons les étapes 1-5 de l’analyse prospective et l’étape 2 de l’analyse
rétrospective sur notre cas d’étude.
Section 4 – Synthèse. Nous présentons des actions concrètes à appliquer après
l’implémentation de nos deux analyses.

xvi
4. Conclusion
La conclusion est divisée en 4 sections. Dans un premier temps, nous rappelons la
problématique adressée au cours de la thèse (section 1). Puis, nous mettons en évidence les
contributions réalisées (section 2). Enfin, nous discutons des limites et des améliorations
possibles (section 3), avant de terminer en évoquant de futures pistes de recherche (section 4).
Ci-dessous voici un court résumé de chacune des sections.
Section 1. Comme nous l'avons démontré dans la partie 1 de cette thèse de doctorat, le bloc
opératoire est un environnement intrinsèquement complexe et soumis à des incertitudes. Cela
est dû aux nombreux moyens humains et techniques impliqués et devant être synchronisés pour
chaque cas d'opération. Cette complexité implique de maîtriser la problématique de planification
et d'ordonnancement des cas chirurgicaux. Le problème de planification et d'ordonnancement
est divisé en 4 niveaux de décision : (1) stratégique (problème de planification de la capacité),
(2) tactique (problème de planification de la chirurgie principale), (3) opérationnel hors ligne
(problème de planification de la chirurgie) et (4) opérationnel en ligne (problème de gestion de
l'exécution du programme opératoire). Dans le Chapitre II, nous montrons qu'il existe peu de
travaux de recherche pour traiter les problèmes survenant au niveau opérationnel par rapport à
l'ensemble des contributions aux 2 niveaux supérieurs. En effet, des observations et des
entretiens dans 5 blocs opératoires hospitaliers nous ont permis de mettre en évidence qu'il
n'existe pas d'outil d'aide à la décision aidant le personnel du bloc opératoire à chaque fois qu'une
perturbation survient. C'est le point de départ de cette thèse qui se concentre sur le niveau
opérationnel.
À la fin du niveau opérationnel hors ligne, un planning prévisionnel est créé. Ce planning est
utilisé par le responsable de la salle d'opération comme guide lors du niveau opérationnel en
ligne (exécution du planning), bien que l'on sache dès le début de la journée qu'il ne sera pas
entièrement respecté. En effet, une partie de la complexité du bloc opératoire réside dans le
caractère stochastique de son activité. Nous nous concentrons sur deux types d'incertitudes : (1)
la variabilité des durées et (2) les arrivées non électives. Ces incertitudes peuvent perturber
l'exécution du planning. Par conséquent, les niveaux opérationnels sont fortement liés à la
problématique de (la) gestion des perturbations.
Dans le chapitre II, nous montrons que la gestion prédictive des perturbations peut avoir lieu à la fois
au niveau hors ligne et en ligne ; son objectif est d'anticiper et de prendre en compte les
perturbations avant même qu'elles ne se produisent. D'autre part, la gestion réactive des
perturbations est liée uniquement au niveau en ligne, au cours duquel le responsable de la salle
d'opération ne fournit un remède à une perturbation qu'après qu'elle se soit produite. La gestion
des perturbations vise à maintenir, voire à améliorer, la performance du programme prévisionnel
tout au long de son exécution. Dans ce travail, nous évaluons la performance de la salle
d'opération en fonction de trois indicateurs clés : (1) l'utilisation de la salle d'opération, (2) les
heures supplémentaires du personnel et (3) le temps d'attente moyen des patients.
Cela nous amène à notre question générale de recherche : « Comment favoriser et maintenir la
performance de l'organisation d'un bloc opératoire dans des conditions d'incertitude ? ». Nous
nous concentrons sur la question de la gestion des perturbations au niveau opérationnel.

xvii
Résumé long en français
Section 2. Dans cette section, nous revenons sur toutes les contributions que ce manuscrit
apporte, à savoir :
(1) Une méthodologie prospective pour évaluer la qualité de la gestion prévisionnelle des
perturbations.
(2) Une méthodologie rétrospective pour évaluer la qualité de la gestion réactive.
(3) Une méthodologie pour former le régulateur à la gestion des perturbations dans un
environnement virtuel.
(4) Une méthodologie pour concevoir et construire un système d'aide à la décision d'un
bloc opératoire basé sur un jumeau numérique.
(5) Une proposition de solution pour modéliser l'exécution du programme prévisionnel et
du programme réalisé
(6) Une méthodologie pour nettoyer et compléter les horodatages du parcours du patient
dans une base de données de bloc opératoire du monde réel.
(7) Un prototype d'un système d'aide à la décision basé sur un jumeau numérique appliqué
à une étude de cas réelle.
Section 3. Dans cette section, nous établissons un parallèle entre les limites de notre étude et
les améliorations possibles que nous pourrions mettre en œuvre dans les recherches futures.
Premièrement, il aurait été intéressant d’avoir accès une base de données plus riche et plus
fiable. Deuxièmement, la modélisation du bloc opératoire pourrait être améliorée en termes
de ressources (humaines et matérielles), d’incertitudes, et de périmètre spatial (étendre l’étude
au-delà du bloc opératoire afin d’englober des départements voisins qui ont un impact sur le
BO). Troisièmement, des développements intéressants pour l’outil d’aide à la
décision pourraient inclure le rajout (1) d’indicateurs de performance d’utilisation de
ressources, (2) de stratégies de gestion des perturbations au cours de l’exécution du programme,
(3) de stratégies de gestion de perturbation avant l’exécution du programme, et (4) de la prise en
compte du point de vue d’experts terrain testant notre OAD. Enfin, une dernière piste
d’amélioration pourrait porter sur les cas d’étude. D’une part, il serait intéressant d'avoir un
panel de différents plannings d'une journée pour avoir une vision plus exhaustive des
informations que notre DT-DSS peut fournir en fonction de la situation. D’autre part, nous
utilisons un cas d'étude d'une journée. Or, les gestionnaires de salle d'opération doivent en réalité
prendre des décisions en matière de planification et de rééchelonnement ne se limitant pas à
l’horizon d'une journée. Par exemple, un cas électif peut être reporté au lendemain ou à la
semaine suivante. Il en va de même pour les cas semi-urgents. Une amélioration importante
serait d'examiner des cas d'étude de plusieurs jours (par exemple, un cas d'étude d'une semaine)
et d'élaborer des décisions telles que la programmation ou le report d'un cas sur un horizon
d'une semaine.
Section 4. Au cours de ce projet de recherche, nous avons développé un jumeau numérique de
bloc opératoire basé sur la simulation d'événements discrets et nous l'avons utilisé comme
système d'aide à la décision hors ligne pour améliorer la gestion prédictive et réactive des
perturbations. Dans cette section, nous discutons des perspectives de recherches futures : la
génération automatique de jumeau numérique de bloc opératoire et le passage en mode « en
ligne » de notre outil.

xviii
ACKNOWLEDGEMENTS

Je souhaite tout d’abord remercier chacun de mes encadrants. Merci à Franck Fontanili, que
je connais maintenant depuis plus de 6 ans (déjà !), pour m’avoir encouragée à me lancer dans
cette aventure farfelue qu’est la thèse. Merci à Maria Di Mascolo de m’avoir fait confiance et
de m’avoir ouvert les portes du G-SCOP. Merci à Cléa Martinez d’avoir été une encadrante
incroyable bien que je fusse sa première doctorante. Merci à Canan Pehlivan d’avoir rejoint le
train en route et d’avoir très largement participé à la relecture et correction de ce manuscrit. Je
vous suis sincèrement reconnaissante pour toute l’aide que vous m’avez apportée au cours de
ces trois ans de thèse et j’espère que nous pourrons continuer de travailler ensemble par la suite.
Ensuite, j’adresse mes remerciements à tous les membres de mon jury, à commencer par mes
rapporteurs Evren Sahin et Vincent Augusto pour leur précieux retour sur mon (bien long)
manuscrit. Merci à Virginie Goepp d’avoir présidé mon jury. Merci à Sondès Chaabane,
Vincent Cheutet, et Virginie Fortineau pour votre bienveillance durant nos échanges et la
pertinence de la discussion que nous avons pu avoir durant la soutenance.
Je remercie le personnel des trois établissements hospitaliers qui nous ont ouverts leur portes et
qui ont rendus cette thèse possible. Merci à Vincent Gervaise et à Stéphanie Durel Pinson
(Hôpital Privé de la Baie, groupe Vivalto Santé), qui ont notamment permis de financer ce
projet. Merci au Dr Pascal Cariven et à Anne Rouzaud (Centre Hospitalier d’Albi). Merci au
Dr Jane Poincenot (Centre Hospitalier Intercommunal de Créteil). Nombre de ces personnes
font partie du comité de pilotage qui a permis de s’assurer que notre travail de recherche était
pertinent. Merci également à Sophie Kerambellec qui en faisait partie. Merci à tous les
membres du personnel hospitalier que j’ai retenu en otage durant de longs interviews et que j’ai
inondé de questions durant mes observations sur le terrain. Ce travail de recherche est certes,
pour vous, mais il est aussi grâce à vous.
Merci infiniment à Clair Augsburger et Allister Wilson, l’équipe de Support Technique
Flexsim, pour avoir répondu au cours de ces trois années à mes 96 mails de détresse (j’ai compté)
et avoir notamment corrigé l’erreur malheureuse qui transformait mes patients en camions ;
notre prototype n’aurait jamais fonctionné sans vous.
Merci à tous les personnels de services information et administratif de l’IMT Mines Albi qui
nous facilitent toutes nos démarches. Merci aux copains (dans l’ordre alphabétique) : Abdallah,
Araceli, Aurélie, Audrey, Clara, Emilie, Hanae, Jiayao, Ghassan, Gui, Guillaume,
Marine, Marlène, Nafe, Robin, Rodolphe, Tianyuan, Wassim, Yohann, Ziqing. C’est un
plaisir de venir au travail et d’y retrouver des gens qui accepteront de garder mon chat quand je
suis en voyage plus que de simples collègues.
Je termine avec un petit mot pour ma famille et mes amis proches (toujours dans l’ordre
alphabétique) : Houba, Layane, Lisa, Lucile, Maman, Maya, Nathalie, Noé, Papa, Tante
Rania. Merci de vous être intéressés à mon travail et de vous être déplacés pour assister à ma
soutenance (sauf les Libanais, mais je vous pardonne).
Samer, merci pour tout, je t’aime.

xix
List of abbreviations

LIST OF ABBREVIATIONS

KPI Key Performance Indicator

OR Operating Room
DT Digital Twin
DES Discrete Event Simulation
LRA Locoregional Anesthesia
OIP Ophthalmologic Induction Preparation
PACU Post Anesthesia Care Unit
MSSP Master Surgery Scheduling Problem
SSP Surgery Scheduling Problem
DT-DSS Digital Twin Decision Support System

xx
SUMMARY

Abstract ................................................................................................................................. i

Résumé................................................................................................................................. iii

Résumé long en français ..................................................................................................... v

1. Partie 1 - Contexte et problématiques .......................................................................... vi
2. Partie 2 – Proposition d’une méthode et d’un outil ....................................................... x
3. Partie 3 – Preuve de concept ....................................................................................... xv
4. Conclusion ................................................................................................................. xvii

Acknowledgements ........................................................................................................... xix

List of abbreviations .......................................................................................................... xx

Summary ........................................................................................................................... xxi

INTRODUCTION ........................................................................................... 1

PART 1 - CONTEXT AND PROBLEMATICS ........................................ 11

Chapter I. The context of the surgical suite / operating room ...................................... 13
1. Overview of the working of a surgical suite................................................................ 14
2. Patient pathways involving the surgical suite.............................................................. 17
3. Surgical suite resources ............................................................................................... 30
4. Performance evaluation in the OR............................................................................... 37
5. Chapter synthesis: selected on-site problematics ........................................................ 39

Chapter II. Background and related works. ................................................................... 41

1. Planning and scheduling problem................................................................................ 42
2. The predictive and reactive disruption management problem..................................... 47
3. A digital twin for the surgical suite based on discrete event simulation ..................... 54
4. Chapter synthesis ......................................................................................................... 58

PART 2 - PROPOSED METHODOLOGY AND TOOL ......................... 59

Chapter III. Solution proposal ........................................................................................ 60
1. Prospective analysis of the provisional schedule ........................................................ 62
2. Retrospective analysis of the performed schedule ...................................................... 64

xxi
Summary
3. A training environment for the OR manager .............................................................. 65
4. Proposition of a standardized method to build a surgical suite digital twin ............... 66
5. Chapter synthesis......................................................................................................... 68

Chapter IV. Modeling and simulation of the schedule execution ................................. 69

1. Proposal of different modelling approaches depending to available data ................... 71
2. Application to our study case ...................................................................................... 74
3. Additional modeling hypotheses applied to our simulation tool................................. 85
4. Chapter synthesis......................................................................................................... 89

Chapter V. Computing durations .................................................................................... 91

1. Correcting the initial timestamps ................................................................................ 93
2. Computing durations ................................................................................................... 98
3. Computing missing timestamps ................................................................................ 101
4. Compute timestamps and durations for the provisional schedule ............................. 111
5. Chapter synthesis....................................................................................................... 120

Chapter VI. Modeling and scheduling non-elective cases ........................................... 121

1. Solution proposal....................................................................................................... 124
2. Modeling non-elective scheduling scenarios ............................................................ 128
3. Chapter synthesis....................................................................................................... 132

PART 3 - PROOF OF CONCEPT ............................................................ 133

Chapter VII. Presentation of the study case ................................................................. 135
1. Presentation of our partners and their facilities ......................................................... 136
2. Proposed method for choosing a study case.............................................................. 138
3. Description of the study case .................................................................................... 140
4. Chapter synthesis....................................................................................................... 147

Chapter VIII. Prospective analysis and retrospective analysis................................... 149

1. Experimentations design ........................................................................................... 150
2. Prospective analysis experimentations ...................................................................... 153
3. Retrospective analysis - step#2. Test other scheduling strategies on the performed
schedule non-elective cases ........................................................................................... 165
4. Chapter synthesis....................................................................................................... 167

CONCLUSION ............................................................................................ 169

xxii
 Summary
1. Reminder of the problematic ..................................................................................... 170
2. Contributions ............................................................................................................. 171
3. Limits and possible improvements ............................................................................ 176
4. Future research perspectives ...................................................................................... 180

Table of contents .............................................................................................................. 181

References......................................................................................................................... 187

List of figures ................................................................................................................... 198

List of tables ..................................................................................................................... 203

Appendices ....................................................................................................................... 206

1. Appendix #1 – description of the study case ............................................................. 207
2. Appendix #2 – visual illustration of our surgical suite digital twin .......................... 210
3. Appendix #3 – description of publications ................................................................ 216
4. Appendix #4 – description of the dt-dss .................................................................... 220

Outil d’aide à la décision à base de jumeau numérique pour l’analyse prospective et

rétrospective d’un programme de bloc opératoire soumis à des incertitudes ........... 233

Digital Twin-Based Decision Support System for the Prospective and the
Retrospective Analysis of an Operating Room under Uncertainties .......................... 234

xxiii
 Introduction

INTRODUCTION

With healthcare demand rising worldwide, medical services are increasingly needed. Hospitals’
performance is tightly linked to their surgical suite’s performance. Indeed, the surgical suite is an
important revenue and expense center with 40%-50% of the hospital’s budget dedicated to it,
which amounts to 30% of overall healthcare costs (Macario et al. 1997; Kaye et al. 2020). Two
studies mention a surgical suite hourly cost in 2014 between 2,000€ and 2,500€ (Mercier and Naro
2014; Childers and Maggard-Gibbons 2018). In France, the average cost of a surgical patient
pathway ranges from 1,316€ (short or ambulatory surgery) to 16,653€ (heavy surgery). Concerning
patient flow, in 2021, in France, for medicine, surgery and obstetric services (French acronym:
MCO for “Médecine, Chirurgie, Obstétrique”), 38% of hospitalized patients (12 million) were
admitted for a surgery (4.6 million). 2,8 million (61%) of surgery patients are outpatients 3. For all
these reasons, it is necessary for surgical suites to be efficient and profitable.
In this manuscript, we make the difference between the surgical suite (suite), which is the entire
hospital service dedicated to surgery, and the operating room (OR), which is a room within the
surgical suite where the surgery is performed. For example, in the US, a surgical suite has an average
of 6 to 7 ORs4.
In Figure 1, we illustrate a patient pathway for a surgery (blue boxes) and the surgical suite within
the hospital as well as its interactions with external services. We represent intra-hospital non-clinical
units (green boxes), intra-hospital diagnostic units that can delay patient access to the surgical suite
(orange boxes), and units outside hospitals (grey boxes). The full arrows represent the usual patient
pathway: the patient is admitted in the hospital, goes to the ward, enters the surgical suite, and
comes back to the ward before being discharged. The dashed arrows represent a similar process,
with the only difference being that the patient stops by the intensive care unit after the surgical
suite. This research work focuses on the surgical suite.

3 [Link]
4 [Link]

1
Introduction

Figure 1 –Interactions between the surgical suite and the other services.

Running a surgical suite is quite complex. First, patient pathways change depending on the patient
admission type: some patients come and go during the same day (outpatients, ambulatory care),
while others stay at least one-night (inpatients, conventional care). Second, patient pathway can
also differ based on the patient emergency level: patients either need immediate and acute care
(non-elective) or can wait for up to several months before their surgery (elective). Third, the surgical
suite is the workplace of a multiplicity of medical and paramedical professions such as
surgeons, anesthesiologists, and registered nurses. Although they all aim at providing care to the
patient, they can have different – sometimes even conflicting – work organization. Fourth and
finally, as displayed in Figure 1, the surgical suite must maintain a close relationship with upstream
wards that supply patients (inpatient units, outpatient units, emergency services, intense care unit),
downstream wards that retrieve patients (inpatient units, outpatient units, intense care unit),
intra-hospital non-clinical units that directly affect the efficiency of the surgical suite (pharmacy,
sterilization, procurement services, informatics services, technical services, stretcher-bearers,
laundry services…), intra-hospital diagnostic units that can delay patient access to the surgical suite
(technical platform, analysis laboratory…), and companies outside the hospital (external laboratory,
temping agency…) (S. Zhu et al. 2019). In Chapter I of this thesis, we will discuss in more details
the relationship between the surgical suite and outside units because it can affect the organization
and the scheduling of the surgeries.
The complexity of the surgical suite’s organization is worsened by the uncertainties inherent to
medical practice that degrade the activity. They both impact and come from the patient, human
resources, and material resources. They can lead to duration variability such as cases lasting longer
or shorter than expected, or they can lead to unexpected events such as no-shows, cancellations,
postponements, and emergency arrivals.

2
 Introduction

Within this context, the objective of this research work is to promote a performant, robust
and resilient surgical suite organization - we define these key words in Table 1 and review
them in more depth in Chapter I and Chapter II. This goal contributes directly to surgical
suite efficiency and profitability, and indirectly to high quality medical care and patient
safety.
Table 1 – Proposed definitions

Key words Proposed definitions

We define performance based on three key performance indicators (KPIs):

1. staff working conditions (overtime),
Performant 2. patient satisfaction (patient waiting time),
3. organization efficiency (OR utilization).
We consider these KPIs both together and independently.

A surgical suite’s organization is robust if it can maintain the same level of

Robust performance, being able to resist variability of surgery durations or patient
arrivals, without a need to adapt.

A resilient surgical suite’s organization can maintain the same level of

Resilient performance despite strong disruptions such as emergency arrivals or
cancellations.

To reach our objective, we rely on the operating schedule, which is the ranked sequence of
surgical cases to be performed in each operating room of the surgical suite. For instance, Figure 2
represents a provisional operating schedule for a 4-OR surgical suite with 18 patients. We
represented in blue the time during which surgeries can be performed: all 4 rooms open at 8am,
OR#1 and OR#2 close at 6pm and OR#3 and OR#4 close at 4pm. In this specific case, ORs are
dedicated to a specialty and different surgeons can perform surgeries in the same room.

Figure 2 - Example of a provisional operating schedule for a 4-OR surgical suite.

3
Introduction

As we will see in Chapter II, the construction of the operating schedule is referred to in the
scientific literature as the OR planning and scheduling problem. Its activities comprise decisions
from four hierarchical levels: strategic, tactical, offline operational, distributing and online
operational (Figure 3):
– At the strategic level (long-term, up to 10 years), hospital management aligns the available
surgical suite’s resources with the forecasted patient demand. This level includes identifying
what are the required resources (capacity planning problem) and distribute them between
the different surgical specialties (capacity allocation problem).
– The tactical level (medium-term, between 6 and 12 months) considers the master surgical
scheduling problem (MSSP). It consists in building a cyclic schedule that describes the
different shifts available for surgeons to operate on patients. This schedule is usually
decided every 6 or 12 months and it spreads over 1 or 2 weeks.
– The offline operational level (short term, weekly) tackles the surgery scheduling problem
(SSP) and is divided into advanced scheduling and allocation scheduling. It consists in
assigning a date, a start time, and resources to each surgery of the following week. When
the SSP is resolved in a stochastic environment, it can also be called predictive disruption
management (Kamran, Karimi, and Dellaert 2020).
– At the online operational level (real-time), the OR manager coordinates the resources,
handles disruptions, and makes real-time decisions to smooth the OR schedule execution
(Hans and Vanberkel, 2012). These decisions include (1) scheduling new non-elective cases
in the current schedule, (2) deciding whether to keep or postpone a case that might go
overtime, and (3) re-organizing the schedule when patients don’t show or when surgeries
are canceled. The online operational level can be found in the literature under different
names such as “OR management operational decision-making on the day of surgery”
(Franklin Dexter et al. 2004) or “reactive management” (Kamran, Karimi, and Dellaert
2020).

4
 Introduction

Figure 3 – The 4 decision levels of planning and scheduling activity

The scientific community has shown an increasing interest in planning and scheduling in the
operating room. The literature is abundant on strategical, tactical, and offline operational planning
of ORs; it focuses mostly on offline decisions. However, in an environment as uncertain as the
surgical suite, multiple important gaps can appear between the provisional and the performed
schedule. To the best of our knowledge, there are few studies on disruption management at the
online operational level. Indeed, as mentioned in (Guerriero and Guido 2011), "Few papers
consider on-line scheduling, aimed at modifying an existing schedule since urgent and emergency
arrivals".
Several reasons could explain this fact. First, higher decision levels impact and condition the
lower ones. Thus, if the strategic, tactical, and offline operational levels are not mastered, the
benefits of improving the online level are greatly reduced. Second, working at the operational
level requires to make decisions quickly. Therefore, optimization methods with little or no
requirement in terms of reaction time (such as the ones used for higher decision levels) may not be
suitable. Third, making decisions at the operational level brings the difficult task of finding (or
developing) mathematical models that consider both the complexity and the stochasticity
of the real world. This is all the more true for online operational decisions. Fourth and finally, the
lack of research work on the operational level may be the consequence of the strong involvement
of human resources in the surgical suite organization. Indeed, since decisions are not all based
on explicit rules, teams are left to choose their own. The current modus operandi is to tackle
organizational issues on the spot during schedule execution: the staff finds empirical solutions to
each disruption before moving on to the next one. At the end of the day, except for serious
incidents that need the involvement of the hierarchy or outside services, the staff does not revisit
past problems.

5
Introduction

All these reasons leave us room to improve the disruption management strategies. This is why, in
this research work, we wish to build on the existing academic work and further improve the online
aspects of the operational level activity management. We make the three following hypothesis: (1)
the allocated resources are fixed, (2) the master surgical schedule is fixed, (3) the provisional
schedule is fixed.

Within this context, we propose the following general research question: “How can we
promote and maintain the performance of a surgical suite’s organization under
uncertainties?”

To answer this question, we propose: (1) A methodology for a prospective analysis of the
provisional schedule, (2) A methodology for a retrospective analysis of the performed
schedule, and (3)A methodology for real-time decision-making virtual training destined to
OR managers.

We illustrate the positioning of these 3 methodologies using Figure 4. Based on our literature review
and on-site observations, we have decided to propose an offline digital twin-based decision
support system (DT-DSS) to improve the decisions made at the online operational level.
This means that this DT-DSS directly supports the prospective analysis of the provisional schedule
and of the retrospective analysis of the performed schedule. This DT-DSS could be used by the
OR manager to indirectly prepares the schedule execution and provides feedback on it. We detail
and justify this choice in Chapter III.

Figure 4 – Research Positioning

We discuss the green boxes’ problematics of Figure 4 in Table 2, and we use a PDCA structure
(plan, do, check, act) (Sokovic, Pavletic, and Pipan 2010) to decompose our research question
into 5 sub-questions. We discuss this methodology in more details in Chapter III. Since we only
focus on offline decisions, we exclude step #2 of the research scope.

6
 Introduction

As a proof of concept, we propose a study case in which we apply these three

methodologies to a surgical suite inspired by the Hôpital Privé de La Baie (GIE Vivalto Santé,
France). To do so, (1) we develop a digital twin-based decision support system for the
prospective and retrospective simulation and analysis of the operating room schedule
execution, and (2) we infer from this prototype a standardized methodology to conceive
and build such a tool in any surgical suite.

Table 2 – Proposed methodology

# Step When? Objective Research Questions Proposition

Q1. How can we anticipate

disruption before schedule
Before execution?
Plan schedule Prospective
1 Plan schedule Q2. How can we assess
execution analysis
execution the quality of the
predictive disruption
management?

Execute the
During
provisional schedule
2 Do schedule / /
while managing
execution
uncertainties

Q3. How can we assess

the quality of the reactive
After Provide feedback on disruptions management?
Retrospective
3 Check schedule the schedule Q4. How can we identify analysis
execution execution whether performance lack
stems from offline or
online decisions?

After Take measures to Q5. How can we train OR

Virtual
4 Act schedule normalize or managers to disruption
Training
execution improve the process management?

7
Introduction

These led us to propose several scientific and technical contributions; we list them in Table 3.
Table 3 - Suggested contributions

# Domain Contribution Q1 Q2 Q3 Q4 Q5 Chapter(s)

Prospective methodology to Chapter III

assess the quality of the
1 Methodology X X + Chapter
predictive disruption
management. VIII

Retrospective methodology to Chapter III

2 Methodology assess the quality of the reactive X X + Chapter
management. VIII

Methodology to train the OR

manager on disruption
3 Methodology X Chapter III
management in a virtual
environment.

Methodology to conceive and

build a digital twin-based
4 Methodology X X X X X Chapter III
decision support system of a
surgical suite.

Proposition of process models

for provisional schedule
5 Model X X X X X Chapter IV
execution and performed
schedule execution.

Methodology to clean and

complete the patient pathway
6 Data X X X X X Chapter V
timestamps of a real-world
surgical suite database.

Prototype of a digital twin-

based decision support system Part II +
7 Result X X X X X
applied to a real-world case Part III
study.

8
 Introduction

After this introduction, the remainder of this manuscript is divided into 3 parts that are themselves
divided into 9 Chapters.
In Part 1, we present the context of our research and the problematics we focus on. First,
based on our on-site observations, staff interviews and database analysis, we present the general
functioning of a surgical suite (Chapter I). Second, we provide a background and related works on
the planning and scheduling problem, on the disruption management problem, and on digital twin
and simulation tools (Chapter II).
In Part 2, we discuss our proposed methodologies and our digital twin-based decision
support system (DT-DSS). In Chapter III, we present our three methodologies and justify why
our DT-DSS is fit to perform them (contributions #3, #4 and #5). The rest of Part 2 is dedicated
to conceiving and building a DT-DSS for our specific study case (contributions #6 and #7). To
begin with, we discuss how we model and simulate schedule execution in a determinist
environment (Chapter IV). Then, we explain how we model duration variability using historical
data (Chapter V). Finally, in Chapter VI, we discuss how we model and simulate non-elective cases
arrivals and scheduling during schedule execution (contribution #1).
In Part 3, we apply our methodology on a one-day operating schedule. In Chapter VII, we
describe the study case we use, which is an operating day inspired from the database of the Hôpital
Privé de La Baie (contribution #2). Using our previously developed DT-DSS, we describe each step
and analyze the obtained results for: the prospective analysis and the retrospective analysis (Chapter
VIII).
Finally, we conclude by summarizing our work, the contributions provided, and presenting
future research perspectives.

9
 Introduction

PART 1 - CONTEXT AND

PROBLEMATICS

11
 Chapter I. The context of the surgical suite / operating room

CHAPTER I. THE CONTEXT OF THE SURGICAL SUITE / OPERATING

ROOM

For this applied research project, we worked with three different French hospitals: Hôpital Privé de
La Baie (HPB, GIE Vivalto Santé, Avranches, France), Centre Hospitalier d’Albi (CHA, Albi, France)
and Centre Hospitalier Intercommunal de Créteil (CHIC, Créteil, France). HPB is a private hospital while
CHA and CHIC are public. In France, it is common to use the term “hospital” (hôpital) to refer to
both public hospitals and private non-profit hospital facilities. The term “clinic” (clinique) refers to
private for-profit hospital facilities. Since there is no need to systematically make a difference in
this research, we will use the term hospital indifferently for any of these structures.
At the beginning of this PhD project, an expert committee was created with directors, surgeons,
anesthesiologists, and engineers from these three hospitals. Together, we ensured the relevant of
our research question, and defined the study perimeter (KPI, uncertainty, operational decisions to
focus on). This helped us shape our research work so that it could be beneficial for the healthcare
community.
In this first Chapter, we structure and synthesize the knowledge we gathered on the general working
of the surgical suites of our three partners. This knowledge is based on (1) on-site observations, (2)
staff interviews, and (3) OR software database analysis. Expert readers of the surgical suite
organization can skip this Chapter and only read the synthesis section 5 at the end.
First, we propose an overview of the working of a surgical suite. Second, we describe the different
types of pathways for patients admitted to surgery. Third, we give an in-depth presentation of the
different resources required in a surgical suite. Fourth, we make a short review on performance
indicators in the OR, and we conclude this Chapter 1 with a synthesis of the problems we raise.

13
Chapter I. The context of the surgical suite / operating room

1. Overview of the working of a surgical suite

The surgical suite is a complex environment that is constantly interacting with services that are
internal or external to the hospital. In this section, we describe the inner workings of the surgical
suite (1.1), and we present their relationship with outside services (1.2).

1.1. The inner workings of a surgical suite

In this subsection, we present an overview of the internal functioning of a surgical suite, and we
propose a high-level process map in Figure 5. We designed it based on our on-site observations,
staff interviews and literature review; it is applicable in each of our partners’ suites. Although,
deliberately not exhaustive, this map allows for a clear overview of the service’s complex internal
affairs.
The proposed process map is based on the process-oriented approach defined by the ISO 9001
standard5. Thus, it displays three types of processes; from top to bottom: decision-making
processes, operational processes and supporting processes. Each blue box represents a macro-
process that could be divided into several steps. The processes have inputs (left column) and
produce outputs (right column).

Figure 5 - High level process mapping for the surgical suite

Decision-making processes (first row) ensure the smooth execution of the operational processes
by planning and managing the activity. In the surgical suite, these processes focus on aligning
patient demand for care with the surgical suite resources while respecting the constraints imposed
by external services. We represented decision-making processes leading to the building of the

5 [Link]

14
 Chapter I. The context of the surgical suite / operating room

operating schedule (ordered list of surgeries to be realized during a specific day) and the staff
schedule. We propose a more in-depth description of decision-making processes in Chapter II; our
focus is on disruption management that occurs in both surgery scheduling and schedule execution.
Operational processes (second row) include the processes through which the suite provides
added value. As the suite’s core activity is to provide care to the patient, they integrate the entire
patient pathway. They are divided into three phases: the pre-operative phase (before surgery), the
peri-operative phase (during surgery), and the post-operative phase (after surgery, or recovery). We
describe the different patient pathways taking place in the suite in section 2; our study perimeter
encompasses these three phases.
Supporting processes (third row) do not themselves contribute to an added value. However, even
if they do not provide care to the patients, they are required for the proper execution of the
operational processes. We represent processes related to intra-hospital non-clinical units (green
boxes), to intra-hospital clinical units (orange boxes), and to equipment management (yellow
outline). Processes in white boxes are operated by the suite’s staff. These supporting processes can
degrade the everyday schedule execution if they are not properly synchronized with the operating
processes. For instance, a patient brought too early in the surgical suite will have to wait before
being received for preoperative care (hence decreasing patient satisfaction), but a patient arriving
too late will make the surgical team wait for them and keep an operating room idle (hence risking
the decrease of OR utilization and increase of staff overtime). In this research, we consider the
operating room cleanup.

The surgical suite’s organization revolves around three types of processes: Decision-making
processes (plan and manage the activity), operational processes (provide care to patients),
and supporting processes (allow the proper execution of the patient pathway).

Our objective is to improve the performance of operational processes by working on the

disruption management occurring during the surgery scheduling and the schedule execution
processes. We consider the three phases of the patient pathway and include the supporting
process that is the operating room cleanup.

In this section, we presented an overview of the working of a surgical suite and highlighted the
processes on which our work is focused. In the following section (1.2), we focus on the
relationships of the surgical suite with its outside environment.

1.2. The relationship between the surgical suite and outside units
In Figure 6, we propose a representation of the relationship established between the surgical suite
and external services to provide care to patients. This includes the operational and supporting
processes described in Figure 5. Each box represents one or several services: intra-hospital non-
clinical units (green boxes), outside hospital units (grey boxes), intra-hospital clinical units (orange
boxes), patients ward (blue boxes), surgical suite (white boxes). Arrows represent flows of patient,
information, material resources or service.

15
Chapter I. The context of the surgical suite / operating room

Figure 6 – Interactions between the surgical suite and external services

The surgical suite is strongly tied to external services. First, the surgical suite receives patients from
upstream wards (the inpatient unit, the outpatient unit, and the emergency ward). It supplies
patients to the same service, plus the intensive care unit (ICU). Second, patients can have delayed
access to the surgical suite because of intra-hospital diagnosis units such as the radiology or the
laboratory. Third, the surgical suite relies on resources from outside the hospital to complete its
capacity (equipment provided by laboratories, temporary workers…). Fourth, the efficiency of the
surgical suite is directly affected by intra-hospital non-clinical units such as: IT department,
technical services (procurement), cleaning services, sterilization unit, pharmacy (drugs), and
stretcher bearers.

The surgical suite is strongly tied to external units. These units can be either inside or
outside the hospital, and either clinical or non-clinical. The performance of the suite’s
organization is dependent on how well the patients flow, information flow, and service flow
with these units are managed.

Now that we have a better understanding of the inner functioning of the surgical suite (1.1) and of
its relation with other units (1.2), we will provide a detailed description of the patient pathways
going through the surgical suite (this includes the operational processes described in Figure 5).

16
 Chapter I. The context of the surgical suite / operating room

2. Patient pathways involving the surgical suite

The surgical suite’s core activity is to provide care to patients. Doing so, the service welcomes a
variety of patients, and thus, a variety of patient pathways are possible.

2.1. Patient features that influence the patient pathway

To begin with, we present three distinctive patient features that impact the patient pathway: the
admission type, the emergency level and the anesthesia type. Then, we present the steps followed
by patients before entering the suite within the suite, and after exiting the suite 2.5, based on these
features. Finally, we will address the uncertainties linked to patients during their passage at the
hospital, and more specifically, in the surgical suite. Patient features that influence the patient
pathway.

2.1.1. Patient type with respect to admission type: outpatients and inpatients
Patients receiving surgery are either outpatients or inpatients. Outpatients are admitted, undergo
surgery, and leave the hospital on the same day, while inpatients stay at least one night in the
hospital (S. Zhu et al. 2019). In other words, outpatients correspond to ambulatory surgery, while
inpatients correspond to conventional surgery. We describe below the other main differences
between inpatients and outpatients: admission date and transportation. We display a synthesis in
Table 4.
Admission date. Inpatients are usually admitted one or more days before the day of the surgery,
whereas outpatients are admitted on the day of the surgery. Thus, inpatients can be classified as
stand-by while outpatients can cancel, be no-show or arrive late (Duma and Aringhieri 2015;
Guinet and Chaabane 2003). It is interesting to note that, in some hospitals, inpatients have the
option to arrive on the day of their surgery like outpatients (but are hospitalized afterwards).
Transportation. Outpatients usually receive light surgeries and can thus enter the suite walking,
on wheelchair or on a stretcher. Inpatients, with heavier surgeries, mostly come in lying down on
a stretcher. Thus, patient admission type impacts the patient transportation process. Indeed,
stretcher-bearers (the staff in charge of transporting patients), stretchers, and wheelchairs are
limited resources of which the unavailability can hinder the schedule execution.

17
Chapter I. The context of the surgical suite / operating room

Table 4 - Main differences between inpatients and outpatients

# Characteristics Inpatients Outpatients

1 Type of surgery Ambulatory Conventional

2 Admission date Before surgery date On surgery date

Number of hospitalized
3 At least one None
nights

Where do patients start and

4 Inpatient wards Home
end their day?

Walking, sitting, lying

5 Usual transportation mode Lying down
down

Outpatients are admitted, undergo surgery, and leave the hospital on the same day, while
inpatients stay at least one night in the hospital. Outpatients and inpatients differ in terms
of type of surgery, admission date, number of hospitalized nights and patient pathway.

In our study, we consider both inpatients and outpatients; we model them in the same way:
they follow the same patient pathways and receive surgery in the same operating rooms.

2.1.2. Patient type with respect to emergency level: elective and non-elective patients
Patients receiving surgery can either be elective or non-elective patients. Elective patients can
wait before their surgery. They are scheduled weeks or months ahead of time (S. Zhu et al. 2019).
Non-elective patients arrive unexpectedly and need to be treated right away.
That being said, non-elective patients are not all prioritized equally: a patient with a peritonitis6 (an
acute inflammation of the peritoneum) will have a priority over a patient with a closed ankle
fracture (i.e. bone does not break through the skin). However, as displayed in Table 5 (Van Riet
and Demeulemeester 2015), the criteria to categorize non-elective patients are not standardized in
the scientific literature. References can be found in the article. The Category column specifies the
name given to the non-elective type studied, and the Target column specifies the ideal maximum
delay duration between the hospital admission and the surgery times. For instance, depending on
the article, the term “emergent” may refer to patients needing to undergo surgery within 30
minutes, 1 hour, 2 hours, 6 hours, or 24 hours.

6 [Link]

18
 Chapter I. The context of the surgical suite / operating room

Table 5 – Examples of categorization of non-elective patients found in the scientific literature. Taken
from (Van Riet and Demeulemeester 2015))

Category Target

Trauma Now

Emergent <30 min, <1 h, <2 h, <6 h, <24 h

Urgent <4 h, < 24h

Semi-urgent <8 h, <1/2 w

Add-on <24 h

Add(-on) elective No target: fill up free capacity

Non-urgent <24 h

Work-in [24 h – 1 w]

P1–P3 (emergent): <1 h, <4 h, <12 h

P1–P3: <8 h, <8–24 h, <24–48 h
Priority levels (P)
P1–P3: <6 h, <24 h, <78 h
P1–P5 (emergent): <45 min, 2 h, 4 h, 8 h, 24 h

The non-elective category definition disparity makes it crucial to specify it for this study.
Consequently, we adapt Table 5 definitions and propose in Table 6 three categories of patients that
differ based on their target.
Table 6 - Proposition of categories based on target for our study.

# Category Target

Semi-Urgent Non-
1 Surgery is maximum three days after the first admission.
Elective (SUNE)

Urgent Non-Elective
2 Surgery is on the same day as the first admission.
(UNE)

19
Chapter I. The context of the surgical suite / operating room

Elective patients can wait before their surgery. They are scheduled weeks or months ahead
of time. Non-elective patients arrive unexpectedly and need to be treated right away.

In our study we consider both types of patients and model them in different ways. We
consider elective patients, semi-urgent patients and urgent patients.

To conclude this part dedicated to patient type, we add that:

• Inpatients, outpatients, elective patients, and non-elective patients follow the same pathway
and/or use the same facilities, or not. This depends on the surgical suite’s organization and
can change from one site to another. For instance, a hospital with two surgical suites can:
(1) have one suite dedicated to outpatients and one suite dedicated to inpatients, (2) have
ORs dedicated to outpatients and ORs dedicated to inpatient within the same surgical suite,
(3) mix outpatients and inpatients in the same OR.
• The admission type and the emergency level features are independent: elective patients and
non-elective patients can be either outpatients or inpatients (and inversely). However, most
outpatients are also elective patients. This can be explained by the fact that the non-elective
patients usually require heavier surgeries and thus need more after-surgery care.

2.1.3. The four main types of anesthesia

The type of anesthesia the patient receives during surgery can change the patient pathway within
the surgical suite. We describe here the four main types of anesthesia: general anesthesia, sedation
or monitored anesthesia care, local-regional anesthesia, and local anesthesia7-8.
General anesthesia (GA). For a GA, the anesthesiologist and/or the nurse anesthetist put the
patient in a sleeping state where they have no awareness, nor sensations. The anesthesia can be
provided in a dedicated anesthesia room, or in the operating room itself – although its mostly
performed in the OR.
Loco-regional anesthesia (LRA). During an LRA, the patient stays awake while the
anesthesiologist (most of the time assisted by a nurse anesthetist), uses numbing medication on a
specific area of the body in order to prevent the patient from feeling pain. Common types of LRA
include spinal anesthesia, epidural anesthesia, and nerve blocks. Spinal and epidural anesthesia can
be used for childbirth of for heavy orthopedic surgeries below the waist such as total knee and total
hip replacements. On the other hand, nerve blocks are used for smaller areas such as an arm or a
leg (e.g. femoral nerve block or brachial plexus block). GA and MAC (see below) can be coupled
with LRA in order to reduce patient pain during recovery.
Sedation or Monitored anesthesia care (MAC)9. Sedation is provided by the anesthesiologist
and/or the nurse anesthetist; it makes the patient feel drowsy and relaxed without necessarily losing
consciousness. Mild sedation is often used for eye surgery (e.g. cataract surgery) and allows the

7 [Link]
8 [Link]
9 [Link]

20
 Chapter I. The context of the surgical suite / operating room

patient to answer questions and follow instructions. Moderate sedation may let the patient doze
off, although they would be able to wake up easily. With deep sedation, patients fall asleep but are
still able to breath on their own – contrary to GA. It is notably used for endoscopy or colonoscopy.
Local anesthesia (LA). LA can be provided without the supervision of an anesthesiologist or a
nurse anesthetist. Staff may inject medication (such as lidocaine) or apply a numbing cream on a
specific small area. LA is usually used to relief pain during short procedures (e.g. sewing a deep cut)
and is often coupled with MAC during minor outpatient surgery.
Some anesthesia types are provided in the operating room, others are provided in the PACU. Some
require anesthesia team members, others do not. From now on, we will discuss anesthesia not
based on their type but based on where they are provided, and by whom. We provide a synthesis
of the anesthesia type in Table 7.
Table 7 - Brief description of the four main types of anesthesia provided in the surgical suite.

Anesthesia Induction
Anesthesia type Patient State
team? Location

General
The patient is deeply asleep. Yes OR
anesthesia (GA)

A specific area of the patient’s body is Spinal and

Local-regional completely numbed using injected epidural: OR
Yes
anesthesia (LRA) medication. The most common types of Block:
LRA are spinal, epidural and block. PACU

Sedation or Depending on the sedation intensity, the

monitored patient is either deeply asleep (deep),
Yes PACU
anesthesia care dozing off (moderate), or able to follow
(MAC) instructions and answer questions (mild).

The patient is injected medication or is

Local anesthesia
applied cream in order to relief them from No OR
(LA)
pain.

Now that we have defined these patient features, we present the different patient pathways that
pass by the surgical suite and explain how these features impact them. We describe the patient
pathway as a multi-step process.

2.2. From taking a surgeon appointment to being ready for hospital admission
We present the patient pathway as a multi-step process. In this section, we focus on the period
from when the patient requires a surgeon consultation to the moment where the patient is ready
for hospital admission. We describe this process for elective patients (i.e. who can wait before
undergoing surgery), whether they are outpatient or inpatient. Indeed, for these specific steps,

21
Chapter I. The context of the surgical suite / operating room

inpatient and outpatient pathways barely differ. Note that there can be up to several months
between the surgery consultation and the surgery day.
The process consists in 5 mains phases: (1) taking an appointment with the surgeon, (2) surgeon
consultation, (3) anesthesiologist consultation, (4) patient pre-admission, and for outpatient only: (5)
nurse calls before the surgery day. We visually display the pathway in Figure 7.
We describe slightly more each step:
• First, the patient takes an appointment for a surgery consultation at the surgery
secretariat. This can be made in person, over the phone, or on the internet. See Figure 7.1.
• Second, on the day of the surgery consultation, the patient meets with the surgeon. They
discuss whether there is a need for surgery or not, and if it’s the case, whether it will require
conventional or ambulatory surgery. The surgeon also explains whether anesthesia will be
needed, and if yes, what type. The surgeon can also order more exams before making any
decision. Then, the patient schedules the date of the surgery with the surgeon or with
its secretariat. They also schedule the anesthesia consultation (if needed) as well as a
preoperative consultation (or pre-admission) with a nurse. See Figure 7.2.
• Third, during the anesthesia consultation (2 to 3 weeks before the surgery), the
anesthesiologist decides whether the patient can be administered the required anesthesia.
If not, the surgery is postponed. See Figure 7.3.
• Fourth, on the day of the pre-admission, a nurse from the surgery service explains and
organizes: (1) the admission, (2) the hospital’s exit and (3) the post-operative follow-up
with the patient. Other members of the medical and/or paramedical team can be present
if needed. Note that pre-admission is usually on the same day as the anesthesiologist
consultation. See Figure 7.4.
• Fifth, the patient waits for their admission day. In case of an ambulatory surgery (outpatient
case), a member of the suite staff will call the patient the day before the admission to give
them the hours at which they will be expected at the ambulatory service. Apart from that,
the inpatient and outpatient pathways are very similar. See Figure 7.5.

22
 Chapter I. The context of the surgical suite / operating room

Need for surgeon Surgery Anesthesia

Preadmission Day
consultation consultation day consultation day

• Ambulatory
Take an Surgery • Surgeon Anesthesia • Anesthesiologist service
Preadmission
appointment for • Surgeon consultation • Patient consultation • Patient nurse
surgeon secretariat • Patient
consultation

Patient goes back

Need for gome
Can the
surgery ?
patient receive
Wait for surgery
anesthesia ?
consultation day

Yes

No
Schedule the • Surgeon their
date of the secretariat
No • Anesthesiologist Day before the
surgery • Patient
Postpone surgery or their surgery
date secretariat
• Patient

Yes

Need for • Surgical suit

If oupatient: call
anesthesia ? • Anesthesiologist nurse
Schedule another patient at D-1
or their • Patient
anesthesia
secretariat
consultation
• Patient
Yes

Schedule • Surgeon or Patient waits for

anesthesiology their surgery day
secretariat Patient goes back
consultation gome
• Patient

No

• Surgeon or
Schedule their
preadmission secretariat
• Patient

Patient goes back

home

Figure 7 – Patient pathway from surgeon consultation to being ready for the hospital admission for
elective inpatients and outpatients.

For elective patients (whether they are inpatient or outpatient), the steps from needing a
surgeon consultation to being ready for hospital admission are (1) taking an appointment
for a surgeon consultation, (2) surgeon consultation, (3) anesthesiologist consultation, (4)
preadmission, and (5) if outpatient: wait for OR nurse call on the day before surgery. In
our study, we focus on the surgery day; we consider these steps to have already been
performed.

2.3. From the hospital admission to entering the surgical suite.

The hospital admission steps change depending on the admission type and the emergency level of
the patient. We first present the steps for an inpatient elective case, then for an outpatient elective
case, and then finally for a non-elective case (Figure 8).
Inpatient elective case. On the admission day, inpatients present themselves to the hospital for
administrative admission. They are moved to their bedroom in their specific ward where they will

23
Chapter I. The context of the surgical suite / operating room

sleep during their stay. A nurse recovers their latest exams and reminds them how their stay is
going to unwind. The patient also meets the surgeon and the anesthesiologist for a last check-up
before the surgery. On the surgery day (usually the next day), a nurse informs the inpatient of their
departure for the surgical suite and prepares them for surgery. Once the patient is ready, one or
two stretcher-bearers bring them to the surgical suite.
Outpatient elective case. The steps are quite similar for outpatients except that: (1) the admission
and the surgery are the same day, (2) the patient does not spend the night at the hospital, (3) the
patient does not meet the surgeon and anesthesiologist before entering the suite, (4) the patient is
admitted to the ambulatory ward instead of the surgery service, (5) outpatients are more likely to
go to the suite sitting on a wheel chair or walking, although they will still be accompanied by
stretcher-bearers.
Non elective patients. The process for non-elective patients is quite different10,11. A non-
elective patient can be either outpatient or inpatient depending on the severity of the surgery and
on the waiting time between the date of the patient arrival at the hospital and the patient surgery
date. First, the patient can either enter the emergency department through the pedestrian entry (if
they came by their own mean of transportation) or lying down on a stretcher (if they were brought
in by an ambulance). In both cases, they undergo administrative reception and triage in the
emergency department. During the triage, a specialized nurse assesses the emergency level of the
patient. Mildly severe patients are moved to the waiting area while severe cases are immediately
moved to the treatment area. In the treatment area, patients are examined by doctors and nurses
and prescribed further exams if needed. Between exams, patients are moved back to the waiting
area. Once the examination is completed, the doctor decides where to send the patient next: (a)
their home, (b) a hospitalization bed, (c) the surgical suite, (d) another care facility.

10
[Link]
11
[Link]

24
 Chapter I. The context of the surgical suite / operating room

Non-elective
patient

Need for
complementary
exams ?

Administrative • Receptionist
Yes
Reception • Patient

Ask for
complementary • Doctor
exams
• Reception
Triage / Nurse
organization Nurse
Reception
• Patient
Move to waiting
• Patient
room
No
Where to move
patient ?
Doctor • Doctor
Examination #2 • Patient
Move to waiting Move to
• Patient • Patient
room treatment area

Send patient to
Doctor • Patient Send patient What to do with
another care
Examination #1 • Doctor home patient ?
facility

• Nurse Hospitalize the Send patient to

Nurse care patient surgical suit
• Patient

Figure 8 - Non-elective patient pathway in the emergency wards.

We have detailed the patient pathway’s steps: (1) for elective patients, from requiring a surgery to
being transported to the surgical suite, and (2) for non-elective patients, from being admitted to
the emergency wards to being sent to the surgical suite. We now describe the patient pathway
within the surgical suite.

2.4. From entering to exiting the surgical suite: the patient pathway in the surgical
suite
In the preceding subsections, we described the patient pathway before entering the surgical suite
(2.2 and 2.3). In this subsection we discuss the process as the core of our study. We present the 3
main types of patient pathways within the surgical suite that we observed on-site:
a) Surgery with an induction in the OR (GA, spinal anesthesia, epidural anesthesia)
b) Surgery with an induction in the PACU (block, sedation)
c) Surgery without the need of the anesthesia team (local anesthesia)
To begin with, we describe in depth the patient pathway requiring induction in the OR (a). Then
we explain what changes between this pathway and the two other cases (b and c).
a - Surgery with induction in the OR. To better understand the patient pathway in the surgical
suite, we illustrate it with an example inspired by a patient pathway in HPB’s surgical suite. The
layout shown in Figure 9 is extracted from the complete surgical suite layout of HPB that is later
shown in Figure 13. The arrows represent the path followed by the patient in the suite and the
associated number the sequence of their steps. The example we present here is the one of an
elective inpatient coming for a hip replacement under general anesthesia.

25
Chapter I. The context of the surgical suite / operating room

We highlighted:
• the transfer area (white): the area between the surgical suite and the rest of the hospital.
• the patient waiting area (blue): the area where the patient waits before the surgery.
• the operating rooms (green): the rooms in which surgeries take place.
• the post-anesthesia care unit or PACU (yellow): the area where patients are transferred
to for recovery after their surgery.

Figure 9 – Illustration of a patient pathway in the surgical suite.

Preoperative phase. The preoperative phase starts with a stretcher bearer transporting the patient
from their inpatient wards (outside the surgical suite) to the transfer area (1). The patient is
welcomed in the suite by a nursing assistant that records the patient suite entry time, checks their
identity, and installs them on an operating table. Since general anesthesia does not require passing
by the PACU, the nursing assistant brings the patient directly to the patient waiting area (2). If a
PACU nurse if available, they come to put the patient on a drip. If not, the patient waits until the
nurses from the surgical team (2 OR nurses and one anesthetist nurse) do the checklist and bring
them to the operating room (3). This is the end of the preoperative phase and the start of the
peri-operative phase.
Peri-operative phase. In the operating room, the patient undergoes the following steps: setup,
induction, procedure, and reversal. During the setup, the staff finishes preparing the material and
installs the patient for the induction and/or the procedure. During induction, the anesthesiologist
(and/or the anesthesia nurse) provide anesthesia to the patient and install them for the procedure.
During the procedure (from incision to the suture), the surgeon and the OR nurses operate on
the patients while the anesthesia team makes sure the patient stays asleep. During the reversal, the
nurses close the patient’s wound and prepare them to leave the OR and enter the PACU.

26
 Chapter I. The context of the surgical suite / operating room

The surgeon and the anesthesiologist can stay during the entire peri-operative phase. However,
usually the surgeon only stays for the procedure (incision to suture) and the anesthesiologist only
stays for the induction (induction start to induction end). The nurses stay with the patient the entire
time they are in the OR. Once the reversal is over, the nurses move the patient to an available
PACU bed. This is the end of the peri-operative phase and the start of the post-operative one.
Post-operative phase. One nurse stays in the OR for the clean-up. The other one transports the
patient to a PACU bed (4) and prepares the OR for the next case. The patient is then under the
responsibility of the PACU nurses. Once the patient’s vitals have been stabilized, the nurses call
the stretcher bearer to transport the patient to the outpatient ward or to the patient’s hospitalization
ward (5). This marks the end of the surgical suite patient pathway.
In Figure 10, we represented all the steps of this patient pathway as well as the human resources
required for each one.

Figure 10 – Surgical suite patient pathway with in-OR induction.

b - Surgery with induction in the PACU. The main difference between this pathway and the
previous one is that the induction is neither realized at the same location and nor at the same
moment. To illustrate this, we represented the preoperative phase options in Figure 11’s flowchart:
• Each box is a patient pathway step.
• The diamond marks an intersection based on the type of anesthesia required: patients go
straight to the OR after their checklist (option 1), or receive anesthesia preparation,
induction and monitoring for either block LRA or sedation.
• The required resources are listed on the right of the boxes.
• RN stands for “registered nurse”. Nurses, even if not specialized, can be trained to (1) help
the anesthesiologist for LRA preparation and induction, and (2) perform sedation
preparation, induction and monitoring alone.

27
Chapter I. The context of the surgical suite / operating room

Once in the OR, the patients who received their induction in the PACU follow the following
process: setup, procedure, and reversal (no induction).

Suite Entry

Move patient to • Stretcher-bearer

preoperative area

• Nurse Assistant
Patient Reception

Type of
preoperative 3 - Sedation
care ?
1 - General, Spinal,
Epidural or Local
2 - Block

• RN
Prepare patient for Prepare patient for
• RN
LRA sedation

• RN
• Anesthesiologist
LRA induction Sedation induction
• RN

Wait for LRA to Wait for sedation to

work work

Patient Check-List • Nurse team

OR Entry

Figure 11 – Preoperative care patient pathway options.

c - Surgery without the need of the anesthesia team. This pathway is similar to the one with
an anesthesia inside the OR (a). The two differences are: (1) there is no induction at all, (2) neither
the anesthesiologist nor the anesthesia nurse are present throughout the entire patient pathway.
In this subsection, we have described how the anesthesia type required for the surgery could
influence the patient pathway. As a side note:
• The emergency level can influence the patient pathway in the following way: in case of a
vital emergency, the patient can be directly moved to an OR without having the
preoperative phase.
• The admission type can influence the patient pathway in the following ways: (1) outpatients
need to pass by the outpatient wards before exiting the hospital, thus, they need to leave
the surgical suite soon enough so that the outpatient wards is still open for enough time to
do so, (2) outpatients are mostly operated under anesthesia that do not put them in a deep
sleep so that their recovery in the PACU is faster (typically: no general anesthesia).

28
 Chapter I. The context of the surgical suite / operating room

We will now move on to the description of the final leg of a patient pathway passing by the surgical
suite.

2.5. After the surgical suite: from exiting the surgical suite to leaving the hospital
Once the reversal of a surgery is over, the patient can either: (1) be moved to the ICU if they require
close monitoring, or (2) be transported to the PACU for recovery.
In the first case, they stay in the ICU until their state is stabilized. They are then moved again to an
inpatient service until they are free to return home.
In the second case (after passing by the PACU), stretcher-bearers bring patients to their bedroom;
inpatients return to their surgical service, and outpatients return to the ambulatory service. Non-
elective patients follow either the inpatient or the outpatient pathway.
Inpatients are welcomed by a nurse who checks their vitals and manages their pain. In the evening
the patient has a medical check-up with both the surgeon and the anesthesiologist (not necessarily
at the same time): they discuss the outcome of the surgery and inform the patient on the treatment
follow-up. The inpatient will then have daily meetings with the surgeon, the anesthesiologist, the
nurses, and other staff members until the end of their stay to (1) manage the pain, and (2) ensure a
safe and quick recovery. More exams (blood test, radio, etc.) can be ordered if needed. Upon the
agreement of the medical team, the inpatient will be authorized to leave the hospital.
On the other hand, outpatients are monitored until the RN judge they are stable enough to leave.
A nurse calls them back a few days after the surgery to check-up on them one last time.

29
Chapter I. The context of the surgical suite / operating room

3. Surgical suite resources

In the previous sections, we presented an overview of how the surgical suite works (1.1), how it
interacts with outside services (1.2), as well as detailed the patient pathways involving the surgical
suite (2). Throughout these descriptions, we regularly referred to material resources (e.g. operating
rooms or PACU beds) and human resources (e.g. surgeon or anesthetist nurse). We will now give
a more in-depth description of these resources and the issues linked to their management.

3.1. Material resources

Material resources can be bought or rented. They refer to transportation means (stretchers or
wheelchairs), personal protective equipment, surgical instruments (which are sterilized and
reused), surgical supplies (which are thrown after being used once), anesthesia equipment,
drugs, imaging and visualization equipment (e.g., X-ray machines, endoscopes or
laparoscopes, cameras), surgical support equipment (surgical lights, surgical tables and
positioning devices), information and communication systems (operating room information
system, phones, emails…) and infrastructure resources (premises and storages). In Figure 12, we
represented all these material resources, and colored in green the ones related to the medical
domain.

Transpor-
tation Personal
Infrastructure
Protective
Resources
Equipment

Information and
Surgical
Communication
Systems Instruments

Material
Resources
Surgical
Surgical
Support
Supplies
Equipment

Imaging and
Anesthesia
Visualization
Equipment
Equipment
Drugs

Figure 12 - The different types of material resources required in a surgical suite.

30
 Chapter I. The context of the surgical suite / operating room

To illustrate this, we present an example of a real-world surgical suite infrastructure using the suite
layout of our partner HPB (Figure 13). We use colors to highlight the different areas of the service.
Some areas are dedicated to patients while other are for staff usage only. The zones used by the
patients are:
• the transfer area through which patients enter and exit the surgical suite (white)
• the preoperative patient waiting area (blue)
• the operating rooms in which the patients undergo surgery (OR, green). Operating rooms
can be of different types as they are not necessarily suited for all types of surgery.
• the post-anesthesia-care unit (PACU, yellow). This specific PACU has beds dedicated
to preoperative care: beds for locoregional anesthesia (LRA, orange) and beds for
ophthalmic induction preparation (OIP, brown). Note that although it is common to
find the LRA beds in the PACU, the OIP beds are a specificity of HPB due to their
important ophthalmology activity.
There are also areas that are only authorized for the staff:
• staff office, toilets, and break room (dark blue),
• sterile arsenal storages (SAS, black)
• surgery preparation room (Prep, red)
• clean airlock area (purple)
• dirty airlock area (pink)
• waste area (yellow).

Figure 13 – Commented layout of a real-world surgical suite

31
Chapter I. The context of the surgical suite / operating room

Now that we have described the material resources required by surgical suites, we will discuss the
human resources they need.

3.2. Human resources

3.2.1. Introduction
In the surgical suite, the medical staff and the paramedical staff provide direct care to the patient,
while the management team ensures the smooth planning and execution of the operating schedule.
The technical staff, although mostly not a part of the suite staff, contributes to the running of the
service. The surgical suite is thus a multidisciplinary service where individuals work together
towards the same goal - providing qualitative and safe care to the patients - but with different work
organizations.
In Figure 14, we detail the four staff categories found in the surgical suite (medical, paramedical,
management and technical); we color in green the members of the surgical team, meaning the staff
allowed to enter the OR with the patient.

Figure 14 – Surgical suite staff categories and the different occupations they consist of.

In the remaining of this subsection, we describe the different roles and missions of the staff without
extensively listing their tasks – especially not the medical ones12-13-14-15. Indeed, we only aim at giving
the reader an overview of everyone’s role and an understanding of how coordinated human
resources must be for the schedule to be smoothly executed.

12
[Link]
13
[Link]
14
[Link]
15
[Link]

32
 Chapter I. The context of the surgical suite / operating room

As a side note:
• Human resources can be (1) external to the hospital, (2) internal to the surgical suite, or (3)
external to the surgical suite but internal to the hospital.
• Individuals working in private hospitals can be private practitioners and are not strictly
speaking “staff”. We will not make the difference between employees and liberals.16-17

3.2.2. Medical staff

The suite medical staff is composed of licensed doctors (surgeons, anesthesiologists) and of
doctors in training such as interns and clinic directors. We discuss their missions and display them
in Figure 15.
The surgeon first meets the patient during a consultation to establish whether there is a need for
surgery. During the day of the surgery, they perform the surgical procedure and check the patient’s
state after they exited the PACU. Surgeons can either stay during the entire intervention or from
the incision to the suture (that they both perform). The surgeon can be assisted by another surgeon,
a surgeon in training or an operating room nurse (see 3.2.3)
The anesthesiologist first meets the patient during a consultation to establish whether they meet
the requirements to receive anesthesia. During the day of the surgery, they usually administer the
anesthesia to the patient and monitor their vital signs during the surgical procedure. The
anesthesiologist can be assisted by an anesthesiologist in training or an anesthetist nurse (see 3.2.3).
The anesthesiologist is responsible for the proper anesthesia of the patient but is not required to
provide the anesthesia themselves: it is possible for an anesthetist nurse to provide the anesthesia,
even if the anesthesiologist is not in the room. Thus, the anesthesiologist can stay during the entire
time the patient is in the operating room, only during the induction or the setup phase, or not at
all.

Surgeon Anesthesiologist

•Hosts surgery consultation •Hosts anesthesiology consultation

•Realizes the surgical procedure (incision •Provides and maintains the anesthesia
to suture) (OR entry to OR exit)
•Does follow-up after surgery •May be assisted by: anesthesiologist in
•May be assisted by: another surgeon, a training, anesthetist nurse
surgeon in training, OR nurses

Figure 15 – Brief description of the missions of the suite medical staff.

3.2.3. Paramedical staff

The suite paramedical staff consists in non-medical staff that provide direct care to the patient,
such as registered nurses, nurse assistants, and radio operators. Registered nurses can pass
additional certifications to specialize themselves and become either operating room nurses or

16 [Link]
17 [Link]

33
Chapter I. The context of the surgical suite / operating room

anesthetist nurses. The PACU is managed by PACU registered nurse. This position does not
require an additional diploma. First, we discuss the types of nurses in the surgical suite (Figure 16),
then describe the different types of OR nurses (Figure 17), before finishing with the missions of
the nurse assistants.
OR nurses are responsible for the smooth execution of the surgical procedure, the respect of
hygiene and safety rules, as well as the traceability of the products, equipment and surgical acts
provided during the intervention. OR nurses can fulfill three different roles in the OR. They can
be a circulating nurse, a scrub nurse or an instrumentist nurse. Usually the scrub nurse and
the instrumentist nurse are the same person.
Anesthetist nurses assist the anesthesiologist. Since the anesthesiologist is only responsible for
the quality of care provided to the patient and is not required to be present in the OR, the
anesthetist nurse role can go from assisting the anesthesiologist to providing and maintaining the
anesthesia on their own.
PACU nurses ensure the safe care of patients in the PACU. During the preoperative step, they
prepare patients requiring an LRA and assist the anesthesiologist providing it. They also welcome
the patient once they exit the OR and ensure that they are safely recovering from the surgery,
before sending them to their bedroom.

OR Nurse Anesthetist Nurse PACU Nurse

•Ensures execution of •Assists the •Participates in patient

surgical procedure anesthesiologist preparation for LRA
•Ensures respect of •Provides anesthesia •Ensures the patient's
hygiene and safety rules (induction) stafety in the PACU
•Ensures traceability of •Maintains anesthesia
resources

Figure 16 - Brief description of the missions of the suite paramedical staff (1/2).

During a surgical procedure, each OR nurse adopts a specific role: circulating, scrub and/or
instrumentist. The circulating nurse is the OR conductor and ensures the communication
between the sterile surgical team and the rest of the suite. They ensure the timely preparation,
documentation, and delivery of the surgical supplies to the OR medical team, manage the
documentation related to the patient, and record the in-room timestamps in the information
system. OR nurses stay in-room during the entire intervention, except the circulating nurse that,
when needed, can go fetch additional supplies or ensure communication with the rest of the suite.
The scrub nurse and the instrumentist nurse have overlapping missions: (1) before the surgery,
they help prepare the OR, the surgical supplies, and the patient, and (2) after the surgery they suture
incisions if the surgeon did not do it and perform the reversal. However, their focus is different.
During the surgery, the scrub nurse directly assists the surgeon by holding open incisions, halting
bleeding, cutting wires, and ensuring the surgeon has a clear vision of the surgical site. On the other
hand, the instrumentist nurse hands the instruments to the surgeon and is responsible for surgical
instrumentation and management of the sterile field.

34
 Chapter I. The context of the surgical suite / operating room

OR Nurse OR Nurse OR Nurse

Circulating Scrub Instrumentist
•Ensures communication •Prepares OR, surgical •Prepares OR, surgical
between the OR team supplies and patient supplies and patient
and the rest of the suite • Sutures incisions after •Sutures incisions after
•Prepares surgical supplies surgery surgery
•Enters data in the OR •Directly assists the •Hands the instruments to
software surgeon with the surgery the surgeon

Figure 17 - Brief description of the missions of the suite paramedical staff (2/2).

Finally, the missions of a nurse assistant18 are related to patients (welcoming the patient,
performing the identity check, helping to move the patient within the surgical suite), surgery
(dressing the doctors, preparing the surgical supplies that need to be sterilized), equipment supply,
maintenance, and premises cleaning. These missions can change from one site to another: some
hospitals do not authorize nurse assistant to enter the room while the patient is inside, other only
employ janitors for the cleaning.

3.2.4. Management staff

The surgical suite is also the workplace of the management team, which is made of the OR
manager and the OR coordinator. Our research focuses on helping the OR manager.
An OR manager is an experienced nurse (preferably a former OR nurse). They are the
intermediary between the medical staff and the paramedical staff, as well as between the surgical
suite staff and the board of directors (or more globally, the administrative services). The OR
manager ensures that the staff respects the surgical suite’s charter and adheres to the service project.
The OR manager and the OR coordinator are two different positions with two different sets of
missions; the OR manager takes care of the administrative tasks while the OR coordinator ideally
only focuses on the day-to-day smooth execution of the operating schedule and the handling of
disruptions. However, in some surgical suites, the OR manager is also in charge of ensuring the
proper execution of the operating schedule. It is interesting to note that we noticed an increasing
number of OR coordinators during the last decade. We describe the missions of an OR
coordinator in Chapter II. where we also discuss the processes and decisions linked to the schedule
execution.

3.2.5. Technical staff

Some supporting staff that are neither part of the medical, the paramedical nor the management
team, work in the suite or in close contact with it. Namely: stretcher-bearers, and hospital service
agents. Stretcher-bearers’ organization change from one site to the other: some suites are assigned

18 [Link]
operatoires-et-ambulatoire--reference-2021-748246/

35
Chapter I. The context of the surgical suite / operating room

dedicated stretchers while others are not. Hospital service agents can replace nurse assistants for
the cleaning activities.

3.2.6. The surgical team

During a surgical case, there are several professionals in the OR with the patient: the surgeon, the
anesthesiologist, an anesthetist nurse, and one to three OR registered nurses. The OR surgical team
is divided into the sterile (surgeon, scrub nurse, instrumentist nurse) and unsterile team members
(anesthesiologist, anesthetist nurse, circulating nurse). Additional members can include: a second
surgeon, a medical student, or a resident (a physician undergoing a specialization), a medical device
company representative or a radio operator. Depending on the surgical suite, there can be an
important number of individuals present at the same time in the operating room: it is extremely
important that they all know each other’s work so that they can properly collaborate. An illustration
is provided in Figure 18.

Figure 18 – Representation of the surgical team members during a surgery. 19

19
[Link]

36
 Chapter I. The context of the surgical suite / operating room

4. Performance evaluation in the OR

In this section and the remaining of the manuscript, we talk about schedule performance, schedule
robustness and schedule resilience through misuse of language. For instance, by “schedule
performance”, we mean the performance of the organization (the surgical suite) which followed
this specific schedule.

4.1. Brief review on performance in the OR

(Cima et al. 2011) presents the use of the industrial engineering and manufacturing oriented Lean
and 6 sigma techniques within a medical journal to improve OR efficiency. The authors point out
that efficiency improvement projects often focus on a limited number of ORs or on cases.
However, there are rarely studies on the surgical suite. In this article, they present a Value Stream
Mapping (VSM) of the entire suite, from the decision to operate to the patient's discharge from the
hospital (patient pathway in the OR), carried out by a multidisciplinary team. Each activity is
analyzed according to 3 domains: personnel, information, and time, with several value
improvement objectives: reducing volume variability, streamlining the preoperative process,
reducing non-operative time, eliminating redundant information, and promoting staff engagement.
Results show significant improvements in adherence to start times, and a reduction in overtimes.
The article concludes that "Process mapping, leadership support, staff engagement, and sharing
performance metrics are keys to enhancing OR efficiency".
(Marjamaa, Vakkuri, and Kirvelä 2008) describes the operating room management in terms of who
runs the operating room operationally.
(Michael Samudra et al. 2016) presents a classification of publications on the "OR planning and
scheduling" problem according to patient type, performance measures used, decision support,
services upstream and downstream of the OR, uncertainty, research methodology, and the test
phase. It highlights 3 pitfalls that explain why these methods are not applied: (1) the lack of a clear
choice of authors on whether to target researchers (contributing advanced methods) or
practitioners (providing managerial insights), (2) the use of ill-fitted performance measures in
models, and (3) the failure to understandably report on the hospital setting and method-related
assumptions.
(Heydari and Soudi 2016) addresses the predictive and reactive disruption management problem.
It provides a definition to the terms "schedule robustness” and “schedule stability”: “A primary
schedule is called robust if in confronting disruption, its performance does not exacerbate much.
Similarly, it is stable if the schedule produced after the disruption does not deviate much from the
primary schedule".

4.2. KPI used to assess schedule performance, robustness and resilience

When it comes to assessing schedule performance, robustness, and resilience, it is important to
clearly define what a performant, robust and resilient schedule is. Performance is defined based
on three key performance indicators (KPIs): (1) overtime, where lower overtime instances would
reflect better working conditions for the staff, (2) patience waiting times, where lower waiting
times would indicate an increased patient satisfaction, and (3) OR utilization, where higher

37
Chapter I. The context of the surgical suite / operating room

utilization rates ensure a profitable organizational efficiency. OR overtime, OR idle time, OR

overutilization and OR underutilization are indicators that come naturally when it comes to
studying OR usage. Figure 19 illustrates the different usage states of the OR. The represented OR
shift starts at 8am and ends at 6pm; 4 cases are scheduled throughout the day. We use the following
color code: (1) green: OR is supposed to be available, (2) orange: the OR is supposed to be closed
but is opened for overtime, (3) yellow: the OR is open but empty, (4) light blue: a patient in
undergoing surgery in the OR, and (5) deep blue: the staff is cleaning the OR between surgeries.

Figure 19 – Description of our OR usage related KPIs.

Below we describe how we compute each of our KPIs. Note that these are the duration side of the
KPIs. Divided by the OR shift length, it becomes a rate.
𝑂𝑅𝑢𝑡𝑖𝑙𝑖𝑧𝑎𝑡𝑖𝑜𝑛 = (1 ∩ 4) + (1 ∩ 5)
𝑂𝑅𝑖𝑑𝑙𝑒 𝑡𝑖𝑚𝑒 = (1 ∩ 3)
𝑂𝑅𝑜𝑣𝑒𝑟𝑡𝑖𝑚𝑒 = (2 ∩ 4) + (2 ∩ 5)
𝑂𝑅𝑢𝑛𝑑𝑒𝑟𝑢𝑡𝑖𝑙𝑖𝑧𝑎𝑡𝑖𝑜𝑛 𝑀𝑎𝑥𝑖𝑚𝑢𝑚 (0, 𝑂𝑅𝑠ℎ𝑖𝑓𝑡 − (𝑂𝑅𝑢𝑡𝑖𝑙𝑖𝑧𝑎𝑡𝑖𝑜𝑛 + 𝑂𝑅𝑜𝑣𝑒𝑟𝑡𝑖𝑚𝑒 ))

𝑂𝑅𝑢𝑛𝑑𝑒𝑟𝑢𝑡𝑖𝑙𝑖𝑧𝑎𝑡𝑖𝑜𝑛 𝑀𝑎𝑥𝑖𝑚𝑢𝑚 (0, 𝑂𝑅𝑢𝑡𝑖𝑙𝑖𝑧𝑎𝑡𝑖𝑜𝑛 + 𝑂𝑅𝑜𝑣𝑒𝑟𝑡𝑖𝑚𝑒 − 𝑂𝑅𝑠ℎ𝑖𝑓𝑡 )

A surgical suite’s organization is said to be robust if it manages to maintain the same level of
performance despite system input variability such as stochastic surgery without the need to adapt.
A surgical suite’s organization is said to be resilient if it manages to maintain the same level of
performance despite strong disruptions such as emergency arrivals or cancellations. Robustness
and resilience are defined based on the schedule performance gap between the discrete and
stochastic environments. We use the ANAP targets for surgical suite KPIs as performance targets20.

20 [Link]

38
 Chapter I. The context of the surgical suite / operating room

5. Chapter synthesis: selected on-site problematics

Properly running the surgical suite requires to synchronize the resources inside and outside the suite
with the patient pathway execution. Hence, a performant surgical suite’s organization relies on
resources communication, and coordination within the surgical suite, as well as between the surgical
suite and the outside world. This is complex because both the interactions with external services
and the suite resources are numerous.
First, the surgical suite staff is composed of a variety of human resources (medical staff,
paramedical staff, technical staff, administrative staff) which all require timely access to adequate
material resources (supplies, premises, storages…). Any lateness, absence, or error can lead to
a disruption and hinder the operating schedule execution. Indeed, if the right resource is
not available at the right place and at the right time, the patient pathway will stop. This
goes as much for a surgeon as for a registered nurse.
Second, due to the required interactions between the suite and outside units, dysfunctions
appearing in the suite impact outside units, and vice-versa. This is especially an issue as the default
staff behavior - whether it is within the suite or not - is to focus on their own tasks without taking
into consideration the global working of the hospital, thus neglecting the impact their actions can
have on the overall process. This is all the more true when the surgical suite can be seen as an
isolated environment, due to its strong hygiene and safety rules; the suite staff has few and
constrained contacts with the outside world. One example of what should be clearly communicated
is the ongoing state of the operating schedule and its related patient pathways. We give examples
in Table 8.
Table 8 – Example of dysfunctions coming from a lack of communication.

Service Why should they know about the Example of dysfunction

operating schedule?

Stretchers They should know the patient The stretcher bearers do not know
service so that they can move back where the patient is.
and forth between the patient service The operating schedule falls behind.
and the surgical suite

Suite Staff They should know what the schedule The OR nurses do not know what
is so that they can prepare the equipment to prepare for the next
adequate material. The schedule surgery.
must be made knowing the material The operating schedule falls behind.
resources’ limits (sterilization time
for boxes or endoscopes in case of
necessity at the same time, limited
number of cameras, etc.).

Consequently, the staff must maintain a high level of communication on each patient pathway
advancement - overall the operating schedule execution - so that they can act as a synchronized

39
Chapter I. The context of the surgical suite / operating room

team as well as prevent and deal with disruptions. Consequently, it is necessary to establish strong
communication and collaboration, both within the surgical suite, and between the surgical suite
and outside services. This requires understanding the surgical suite organization in its globality. The
staff needs to pay attention to the processes they impact, the processes they depend on, and the
processes they are responsible for. Resource coordination must be ensured at both the offline and
the online operational steps. In this context, the staff needs to have great communication,
collaboration, coordination, and adherence to established processes.

40
 Chapter II. Background and related works.

CHAPTER II. BACKGROUND AND RELATED WORKS.

In Chapter I, we have highlighted the complexity of the surgical suite. The smooth running of
surgical operations requires perfect synchronization of the human and technical resources involved.
The human factor is omnipresent and constantly thwarts synchronization, making the operating
room a place of uncertainty. In such a context, the planning and scheduling of interventions in
each room is a major challenge.
In Chapter II, we present works related to the operating room planning and scheduling problem
from different levels and horizons. We progressively refocus our literature review on the
operational level, with the management of disturbances, both predictively (before program
execution) and reactively (during execution). Throughout our research, we have observed a cruel
lack of methods and tools for managing disruptions at the operational level. Today, disruptions are
managed empirically, based on the experience of teams, by making decisions without any guarantee
of their effects. However, the scientific literature presents several very interesting contributions on
mathematical approaches to solving the planning and scheduling of operating programs subject to
random disturbances. Even if they prove to be effective in theory, we can only observe that these
approaches are not deployed or integrated into the various OR management software packages we
have seen in practice. This was confirmed in an exchange with international OR expert F. Dexter
(MD, PhD, Professor, Department of Anesthesia, University of Iowa). It is for this reason that we
assume that a digital twin of the operating room, capable of offering end-users a realistic and
faithful visualization and simulation of the surgical program, taking disturbances into account,
could undoubtedly be better accepted. We therefore conclude this chapter with a state-of-the-art
review of digital doubles and simulations used in the field of hospital processes and patient
pathways.

41
Chapter II. Background and related works.

1. Planning and scheduling problem

In this section, we provide a state of the art on papers tackling the planning and scheduling problem
in the surgical suite. This subject has been extensively studied within the literature and
many review papers exist (Przasnyski 1986; Franklin Dexter et al. 2004; Cardoen, Demeulemeester,
and Beliën 2010; Guerriero and Guido 2011; May et al. 2011; M. Samudra, Demeulemeester, and
Cardoen 2013; Demeulemeester et al. 2013; Abdelrasol, Harraz, and Eltawil 2014; Michael Samudra
et al. 2016; S. Zhu et al. 2019; L. Wang et al. 2021; Zonderland and Boucherie 2021; Rahimi and
Gandomi 2021; Harris and Claudio 2022; Van Riet and Demeulemeester 2015).
(Zonderland and Boucherie 2021) provide “a classification of literature reviews on patient planning
and scheduling decisions in healthcare” based on three dimensions: the service providing care
(ambulatory, emergency, surgical, inpatient, home), the level at which the decision is made
(strategical, tactical, offline operational, online operational), and the planning complexity (single
activity planning, multidisciplinary planning, care pathway planning). (L. Wang et al. 2021) provide
“the first literature review on comparing outpatient surgery scheduling with inpatient surgery
scheduling”. Their analysis is based on three dimensions: “the uncertainty incorporation, the
research methodology, and a scheduling performance comparison between both settings”.
The planning and scheduling problem is often divided into four hierarchical decision levels: long-
term strategic, medium-term tactical, short-term offline operational and real-time online
operational. The scientific community has shown a strong interest in the strategic, tactical, and
offline operational levels, and thus focused less on the online operational level. We proceed to
describe the strategic level, the tactical level, the offline operational level, and the online operational
level.

1.1. Strategic level

The strategic level deals with a long planning horizon as well as aggregated and forecasted
information. At this level, decisions are made to optimize the distribution of resources among the
different surgical specialties of the suite. The focus is notably on the number and the specialty of
potential future surgeries, the number of resources required, etc. The strategic level problem is
usually divided into the capacity planning problem, the capacity allocation problem, and the case-
mix problem (CPM):
• The capacity planning problem consists in determining the resources required to meet
patient demand in a cost-effective manner (Roshanaei et al. 2017b; Fügener, Schiffels, and
Kolisch 2017; Koppka et al. 2018).
• The capacity allocation problem consists in assigning the previously identified OR
capacity to each surgical specialty. (Koppka et al. 2018; Roshanaei et al. 2017b;
Vancroonenburg, Smet, and Vanden Berghe 2015).
• The case-mix problem refers to forecasting the number and type of surgeries that will we
performed in the suite. (Koppka et al. 2018; Liang, Guo, and Fung 2015; Castro and
Marques 2015).

42
 Chapter II. Background and related works.

(Choi and Wilhelm 2014) propose a prototypical non-linear stochastic programming model to
allocate each surgical specialty to a certain number of OR days, with the objective of minimizing
total expected costs due to penalties related to not accommodated patients, OR undertime and OR
overtime.

1.2. Tactical level: master surgery scheduling problem

The master surgical schedule (MSS) is a weekly cyclical schedule which describes the OR shifts
for each day. The master surgical scheduling is a tactical process. Every 6 to 12 months, the board
of directors meet with the representatives of the surgical suite staff: the OR manager, the OR
coordinator, at least one surgeon per surgical specialty, and at least one anesthesiologist. Together,
they study the past period surgical suite activity, the next period patient demand forecast, the future
human and material resources evolutions, and the potential surgical activity development
opportunities. They update the current master surgical schedule by adding, deleting, and modifying
OR shifts from the schedule.
(Bovim et al. 2020) proposes an optimization-simulation method to build a MSS made of dedicated
time slots for elective patients and flexible slots for handling potential non-elective patients. They
assess the impact on uncertainties related to surgery duration and patient length of stay in the
hospital using DES. (Makboul et al. 2022) discusses a robust optimization approach for tackling
both MSSP and the scheduling case assignment problem (SCAP). They do it for elective outpatients
on a one-week horizon while considering OR restrictions, surgery priority, resource availability
(OR, surgeon, ICU and PACU beds, etc.), and uncertainties (surgery duration and ICU bed
availability). They assess their solutions using the number of scheduled surgeries and the suite
utilization rate. (Chaabane et al. 2006) and (Kharraja, Albert, and Chaabane 2006) compare two
OR block scheduling strategies for elective cases: the first one considers individual surgeons, and
the second consider groups of surgeons. More papers on MSS can be found in (Koppka et al. 2018;
Vancroonenburg, Smet, and Vanden Berghe 2015).
There are three main scheduling strategies:
• Block scheduling (Koppka et al. 2018; Guido and Conforti 2017).
• Open scheduling (Hashemi Doulabi, Rousseau, and Pesant 2016; Vancroonenburg,
Smet, and Vanden Berghe 2015; W. Xiang, Yin, and Lim 2015b).
• Modified block scheduling (Vancroonenburg, Smet, and Vanden Berghe 2015; Molina-
Pariente, Fernandez-Viagas, and Framinan 2015; Van Huele and Vanhoucke 2015).
The block scheduling strategy consist in pre-allocating OR capacity to different individual
surgeons, groups of surgeons or surgical specialty. This is the most common strategy followed in
real-world surgical suites. For instance, in a MSS build using the block scheduling strategy, every
Monday the OR#1 could be available for surgeon A from 8am to 1pm and to surgeon B from 1pm
to 6pm. The open scheduling strategy is a more flexible solution that does not consider pre-
allocating OR shifts to surgeons or surgical specialties. Consequently, in a surgery schedule
following an open scheduling strategy two cases from two different surgeons, or even from two
different specialties can be scheduled in the same OR. The modified block scheduling
strategy is a mix between the two first strategies. In a modified strategy, the MSS is initially built

43
Chapter II. Background and related works.

according to a block strategy. As the execution day approaches, the dedicated block can then be
opened to other specialties if underutilization is most likely to happen in the OR. For instance,
pediatrics has access to OR#2 every Monday from 8am to 6pm. If on Thursday OR#2 shift has a
utilization rate of 30%, the OR manager might take the decision to open the shift to other
specialties that are struggling to fit all their cases in their own shifts.
Commonly used non-elective scheduling strategies exist. We refer the reader to (Vancroonenburg,
Smet, and Vanden Berghe 2015) for more information on it.

1.3. Offline operational level: the surgery scheduling problem

The surgery scheduling (“programmation”) depends on offline operational decisions. It is divided
into three main phases:
• Advance scheduling: setting a date and a room for each case (Roshanaei et al. 2017b;
Turhan and Bilgen 2017).
• Allocation scheduling: sequencing cases within each ORs. (Kroer et al. 2018; Roshanaei
et al. 2017a),
• Integration of advance scheduling and allocation scheduling
In real life surgical suites, the offline operational level is divided into different steps. First, during
the surgery consultation, the surgeon sets the admission type (outpatient/inpatient) and the
provisional duration of the surgery. They also indicate if the surgery will require supplementary
equipment or staff. The surgeon chooses the surgery date based on their own schedule (shifts,
patient waiting list, already scheduled patients).
Second, during the anesthesia consultation, the anesthesiologist accepts, postpones, or rejects
the surgery. They choose the anesthesia type and specify whether the case needs additional supply.
Finally, during the weekly scheduling meeting (cellule de programmation), the attendees (1) check
that the required resources (staff, operating room, surgical supplies) will be available, (2) identify
and tackle the potential disruptions linked to the schedule execution, (3) fix the sequence and the
starting time of the provisional surgeries, (4) freeze the agreed-upon provisional schedule –
meaning that staff cannot add patients for the following week without directly passing by the OR
manager, (5) send the schedule to each surgical specialty staff, and (6) send the patients’ provisional
suite entry times to the hospitalization services.
The scheduling meeting attendees ideally include the OR manager, the OR coordinator as well as
a representative for each surgical specialty (a surgeon and/or a secretary), and for the
anesthesiologists. The meeting can also include a representative for the pharmacy, the sterilization
service, the ambulatory service, and the surgical suite nurses. At worst, only the OR coordinator
and a representative of the doctors (usually an anesthesiologist) meet. The scheduling meeting
usually takes place on Thursday morning. During the meeting, the attendees discuss the schedule
of the following week (sometimes of the two following weeks).
(Guinet and Chaabane 2003) propose to solve the surgery schedule problem over a one to two
weeks horizon in two phases: (1) assign an OR to each patient, and (2) assign a rank to each patient.

44
 Chapter II. Background and related works.

The article focuses on the first step and solve it using an “assignment model with resource capacity
and time-window additive constraints”.
(Perdomo, Augusto, and Xie 2006) propose a Lagrangian relaxation approach to solve the surgery
scheduling problem while considering both the operating rooms and the post-operative beds.
(Augusto, Xie, and Perdomo 2008) does the same but also considers stretchers. (Lamiri et al. 2008)
propose to use column generation as a decomposition approach to tackle the surgery scheduling
problem while considering the same three types of resources (stretchers, operating rooms and
PACU beds) and while minimizing a patient completion time-based criterion. (Augusto, Xie, and
Perdomo 2010) use a Lagrangian relaxation-based method to study the possibility of using
operating rooms for patient recovery after surgery when the PACU is full. They consider the same
resources and assess their results based on several patients’ completion times indicators.

1.4. Online operating level: the schedule execution problem

We have previously discussed the strategic, tactical, and offline operational decision levels of the
scheduling problem. These levels allow to build a provisional schedule that shall be executed on
surgery day. In this section, we discuss the online operational level which consist in managing the
schedule execution.
On the day of the surgery, the OR coordinator is responsible for the smooth execution of the
provisional schedule. Since it is known in advance that this initial schedule will be disrupted by
uncertainties, the goal of the OR coordinator is not to obtain a performed schedule identical to the
provisional schedule. Instead, they use the provisional schedule as a guide to take the real-time
(online) decisions that will allow them to manage the daily disruptions and reach the performance
targets set for the service.
The performed schedule describes how the schedule unraveled during a past day. It is often
different from the provisional schedule, as uncertainties create deviations between the two
schedules. It comprises elective cases, old non-elective, and new non-elective cases. Usually, the
ongoing schedule execution is represented as in Figure 9: one operating room per row.
The OR coordinator ensures the smooth execution of the schedule. They coordinate the staff and
the equipment so that the patient is provided adequate care, and they take real-time decisions to
deal with uncertainties inherent to the surgical suite functioning. These decisions are only
organizational – never medical. They use the provisional schedule but know in advance that they
are going to deviate from it as disruptions are an inherent part of the surgical suite activity. Below
is a list of decisions to be taken in real-time:
We can note several things about these real-time decisions. First, they can vary from one
coordinator to the other. Indeed, coordination strategy depends on the coordinator personality.
For instance, some are more risk-averse than others. Second, real-time decisions are taken on the
spot, usually based on both the medical and the organization related expertise of the coordinator
– who most of the time is a former and experienced operating room registered nurse. There is no
generally used tool for the coordinator, except maybe their phone and the timeline of the ongoing
schedule. When making a decision, the coordinator has no means to assess thoroughly its possible

45
Chapter II. Background and related works.

consequences. Three, except in the case of a grave incident, there is no feedback or retrospective
analysis on the decisions taken by the coordinator.

46
 Chapter II. Background and related works.

2. The predictive and reactive disruption management problem

2.1. Introduction
Disruptions – and thus schedule modifications - are inherent in the OR schedule execution
(Franklin Dexter et al. 2004). Indeed, uncertainties impact patients, human resources, and material
resources. Thus, planning and scheduling article reviews show an increasing interest in stochastic
approaches to scheduling. We first discuss the type of uncertainties that can be found in the surgical
suite. Then, we address the strategies developed in the literature review to provide a remedy to
these disruptions before they take place (predictive disruption management), and upon their
occurrence (reactive disruption management).

2.2. Uncertainties
In this section, we provide a non-exhaustive review of the different types of uncertainties in the
surgical suite: duration uncertainty, arrival uncertainty, resource uncertainty, and care requirement
uncertainty (S. Zhu et al. 2019).
Patient activity duration in the surgical suite (preoperative care, setup, procedure…) depends on
patient condition, surgeon skill, surgery type and several other factors (Molina-Pariente, Fernandez-
Viagas, and Framinan 2015; Koppka et al. 2018; Kroer et al. 2018; Ng et al. 2017). Activity duration
is thus highly stochastic, and its modeling can have a strong impact on the quality of both the
planning and scheduling (Guda et al. 2016), and the disruption management problems. The three
distributions usually used by the scientific community are the log-normal, gamma and normal ones
(S. Zhu et al. 2019). Other methods exist such as using Monte Carlo simulation.
Patient arrival uncertainty include unpredictable arrival times of outpatients in the hospital
(Kroer et al. 2018; Rachuba, Imhoff, and Werners 2022; Latorre-Núñez et al. 2016; Guda et al.
2016). Indeed, outpatients are admitted and discharged in the same day. Since they are not in the
hospital at the start of the schedule execution day, it is more complicated to control their arrival
times. Approaches to limit this uncertainty include adding some slack time to give more time to
the patients to arrive (Cardoen, Demeulemeester, and Beliën 2010).
Patient arrival uncertainty can also refer to the arrival of non-elective cases. These cases can
be semi-urgent (meaning they do not need to receive surgery on their admission day) or urgent
(meaning they must receive surgery on their admission day).
Resource uncertainty translates the fact that human and material resources might not always be
available at the right time and the right place for the patient (Hashemi Doulabi, Rousseau, and
Pesant 2016; Castro and Marques 2015; Vancroonenburg, Smet, and Vanden Berghe 2015). Since
the surgical suite processes heavily rely on resource synchronization, this uncertainty can lead to
patient waiting times or even to case postponement. (Erdem, Qu, and Shi 2012) proposes a reactive
surgery scheduling model that reschedules not only elective cases but also resources upon the
arrival of non-elective cases.
Care requirement uncertainty (refer to the fact that professionals cannot always know in
advance what care patients will need during their stay at the hospital. Thus, as the patient situation

47
Chapter II. Background and related works.

evolves throughout their stay at the hospital, it can lead to their surgery being canceled or
postponed. For instance, during our on-site observations, we witnessed a patient having their case
cancelled while in the OR; the staff had found a rash on their leg when they were about to get a
hip replacement.

In our study, we focus on uncertainties related to resource availability, activity durations

and non-elective case arrivals.

2.3. The disruption management problem at the operational level

In this section, we rely on two literature reviews. (Franklin Dexter et al. 2004) review reactive
disruption management methods at the online operational level, and (Kamran, Karimi, and Dellaert
2020) discuss both predictive and reactive strategies at the online and operational levels. We
distinguish the articles based on whether they only tackle offline disruption management, or both
offline and online.

2.3.1. Online Operational disruption management

At the online operational level, the objective is to take real-time decisions to ensure a smooth
schedule execution despite uncertainties inherent to medical practice and surgical suite
organization(Franklin Dexter et al. 2004) define OR inefficiency as “the sum of two products:
hours of underutilized OR time multiplied by the cost per hour of underutilized OR time plus
hours of overutilized OR time multiplied by the cost per hour of overutilized OR time” and
describe OR efficiency as “the value that is maximized when the inefficiency of use of OR time
has been minimized”. They propose to make online operational decisions based the following 4
ordered priorities: ensuring patient safety, not cancelling cases, maximizing OR efficiency, reducing
patient waiting time. We will refer to this framework as the 4-priority framework (4PF) in the future
as it comes out often in their article. They list the following online operational decisions (reactive
disruption management strategy):
• Scheduling add-on cases (Zhou and Dexter 1998; F. Dexter, Macario, and Traub 1999;
Franklin Dexter and Traub 2002; Franklin Dexter, Macario, and Traub 1999)
• Fill the schedule gaps (Zhou and Dexter 1998)
• Move already scheduled cases (Franklin Dexter et al. 2003) (F. Dexter 2000) (Franklin
Dexter et al. 2004)
• Assign staff (F. Dexter, Macario, and O’Neill 1999; Franklin Dexter et al. 2004)
• Prioritize limited resources and personnel (F. Dexter and Traub 2000a) (Lebowitz 2003)
• Prepare patients (F. Dexter and Traub 2000b)
• Sequence urgent cases (Zhou et Dexter 1998)
We discuss some of the articles referenced in (Dexter et al. 2004a) and in the literature below.
Scheduling add-on cases. (F. Dexter, Macario, and Traub 1999) use computer simulation to
assess how many hours of add-on elective cases can be scheduled into open OR time for 10
scheduling methods. The different algorithms are the following: Best Fit (with or without fuzzy
constraints), Worst Fit (with or without fuzzy constraints), Best Fit Descending (with or without

48
 Chapter II. Background and related works.

fuzzy constraints), Worst Fit Descending (with or without fuzzy constraints), Worst Fit Ascending,
and Hybrid. Elective cases are defined as cases which can wait at least 3 days for their surgery to
be performed (e.g. an elective patient admitted on Monday can be operated during the next
Thursday or after). Elective cases are considered as “Add-on” when they are scheduled after a
specified cut-off time (e.g. after the weekly scheduling meeting). Their results are likely to reflect
reality for surgical suites with a few add-on elective cases per day.
(Franklin Dexter and Traub 2002) assess two methods to schedule an elective case into an OR:
Earliest Start Time (i.e. the case is scheduled into the first available OR) and Latest Start Time (i.e.
the case is scheduled into the last available OR that allow them to finish the surgery without
overtime; otherwise, the case is scheduled in the first available OR). Their study perimeter is the
following: surgeons and patients choose the day of the surgery, cases cannot be cancelled, and
staffing to maximize the efficiency of OR utilization. First, they show that (1) Earliest Start Time
is rational economically and allows to maximize OR efficiency if the suite is already nearly full, (2)
Latest Start Time is best at balancing the OR utilization between the services, (3) the difference of
utilization between the two methods is only a few methods per OR. Second, they use computer
simulation to assess the impact of surgery duration uncertainty on the performance of these two
heuristics and show that it amounts to only a few minutes per OR. They conclude that there is no
need for strong restrictions on elective add-ons for facilities which aim at ensuring (in this order)
patient safety, patient and surgeon access to OR, and surgical suite efficiency.
Within the 4PF (4 priority framework), (Dexter et al. 2004a) study case sequencing decisions. They
take the following example: OR#1 and OR#2 have both 8.5 h of allocated time. Turnover times
are 0.5 h. A 2.5 h case is scheduled in OR#1 requiring a microscope by surgeon A and a 4 h case
is scheduled in OR#2 by surgeon B. They address the following question: should we schedule a 4
h case in OR#2 requiring the same microscope after the turnover, or should we schedule multiple
shorter add-ons? The first option implies a 100% utilization rate, meaning that any delay would
cause an overtime. To answer that question, they propose a 1.5% accurate statistical method to
compute the probability of one surgery lasting less time than another based on at least 2 historical
durations for each of them.
Scheduling add-on cases and Sequencing urgent cases. (Franklin Dexter, Macario, and Traub
1999) discuss how to optimally sequence non-elective cases. They suggest different methods such
as: minimizing the average patient and surgeon waiting times, applying FIFO scheduling, and
ranking cases based on medical priority.
Filling schedule gaps. (Zhou and Dexter 1998) assess whether an add-on case can be added to
the schedule without leading to overtime by predicting the upper bound of its duration. They
conclude that the prediction bounds were accurate if based on both the surgeon and the procedure
type, and if the case durations were assumed to follow a log-normal distribution.
Moving already scheduled cases. (Franklin Dexter et al. 2003) explain that although it can be
constraining, moving the last case of a day from one OR to another one can improve OR efficiency.
The survey they conducted with physicians show that the OR overtime should be reduced by at
least an hour for the OR change to be perceived as worth it.

49
Chapter II. Background and related works.

(F. Dexter 2000) discusses the information required to move the last case of the day in one OR to
another OR that is idle to decrease overtime labor costs. They compare the overtime per case if
the OR management has access to (1) the exact duration prediction of the case, and (2) historical
duration data for the cases. They show that knowing the exact case duration reduces overtime by
less than 5 minutes than using historical case durations. They conclude that “The use of other
information technologies to assist in the decision of whether to move a case, such as real-time
patient tracking information systems, closed-circuit cameras, or graphical airport-style displays, can,
on average, reduce overtime by no more than only 2 to 4 min per case that can be moved”.
Assigning staff. If there are still surgeries to be performed once the OR shift is over, the surgical
team can either continue in overtime or be replaced by another team. (F. Dexter, Macario, and
O’Neill 1999) discuss the information required to establish a relief strategy for anesthesiologists at
the end of the OR shift. They show that although knowing the exact duration remaining in cases
minimizes anesthetist staffing costs, using historical case durations performs almost as well. They
conclude that “Few additional staff hours would have been saved by supplementing our relief
strategy with other methods to monitor case durations (e.g., real-time patient tracking systems or
closed-circuit cameras in operating rooms)”.
Prioritizing limited resources and personnel. A patient pathway requires the synchronization
of various human and material resources. This complexity increases with the number of patients
as they rely on the same resources (although supposedly not at the same time). (F. Dexter and
Traub 2000a) discuss how to use statistical decision theory based on historical case duration to
decrease the impact of resources on OR scheduling, and thus increase equipment and OR
utilization. They study a specific situation where resource will first be used by the first case of an
OR#1, and then by the first case of another OR#2, knowing that the OR#2’s first case has a higher
probability of lasting longer than the OR#1’s first case.
(Lebowitz 2003) use a Monte Carlo Simulation of a surgical suite to show that scheduling short
procedures first can decrease staff overtime without reducing the number of surgeries performed.
Preparing patients. Real case durations are usually longer or shorter than their predicted value.
Consequently, surgeries may need to start earlier or later than predicted. Since patients need to be
prepared before going to the surgical suite, the time at which they are supposed to be ready needs
to be updated throughout the day. (F. Dexter and Traub 2000b) use simulation to determine at
what time a patient should be ready for surgery on the day of surgery. They assume that the
historical case durations follow a log-normal distribution and compute prediction bounds while
reducing to 5% the risk of OR staff having to wait for the patient.
(F. Dexter, Traub, and Lebowitz 2001) propose a method to compute the delay between two
surgeons working one after the other in the same OR on the same day by using both analytical
expression and Monte Carlo simulation.
Below are mother articles that discuss other decisions.
(Stuart et al. 2010, 20) propose a robust reactive surgery assignment model that minimizes
cancellations of already scheduled patients and maximizes the throughput of non-elective cases.
They focus on a single operating room suite. At the end of each surgery, they re-solve the surgery

50
 Chapter II. Background and related works.

schedule problem while considering the disruptions that occurred since the last reschedule. The
scheduling strategy can be inferred a type of block scheduling policy. They model uncertainties on
surgical durations (log-normally distributed), non-elective case arrivals (exponential distribution
with an average inter-arrival time of 225 minutes), and resource availability (human or material).
These disruptions can lead to either early or late start times for the scheduled patients.
(Stuart and Kozan 2012) suggest a reactive scheduling model with the goal of maximizing the
weighted number of expected in-time patients. The tool can delay, reschedule, or add additional
non-elective cases while respecting the constraint on a single OR suite capacity. They model
uncertainties on surgical durations by adding a slack time at the end of each case, non-elective case
arrivals, and resource availability (human or material).
(Erdem, Qu, and Shi 2012) study a rescheduling problem of elective patients in case of non-
elective patient arrivals. They consider the overtime cost of the suite and/or the PACU, the cost
of postponing or preponing elective cases, and the cost of turning down the non-elective cases in
their objective function. They include both the surgical suite and the PACU in their study.
(He and Xiang 2013) tackle the rescheduling problem when a lack of resources makes it
unfeasible. They consider uncertainty on resource availability, maintain the patient-surgeon
coupling after the schedule modification, and aim at minimizing the duration between initial case
start and modified case start.

2.3.2. Online and offline operational disruption management

(Dios et al. 2015) present a scheduling and rescheduling decision support system for elective
cases and suite resources (surgeons and nurses) with the objective of minimizing the number
of changes to already scheduled cases, the patient waiting time and the resource re-scheduling cost.
They develop their too so that they can generate the surgery schedule based on an elective patient
waiting list up to 6 months in advance, while allowing for manual last-minute changes in case of
disruptions such as non-elective arrivals. Their DSS is currently in use in one of the largest hospitals
in Spain.
(Bruni, Beraldi, and Conforti 2015) develop a reactive scheduling model for a multiple OR
suite. They include both non-elective arrivals and duration uncertainty. They model three reactive
disruption management strategies: scheduling cases in overtime, swapping cases, and rescheduling
all the remaining cases. They consider a block scheduling strategy.
(Addis et al. 2016) propose a rolling horizon approach for the offline scheduling and
rescheduling problem. They use an elective case waiting list that considers continuously joining
new patients and where each patient is assigned a maximum waiting time before being scheduled.
They assume a block scheduling strategy. Their study includes two sources of uncertainty: patient
arrival in the waiting list and surgery durations.
(Heydari and Soudi 2016) consider a predictive/reactive assignment and sequencing model for a
multiple identical OR suite. They consider both elective and non-elective cases, and they include
uncertainties on activity duration and non-elective arrivals. They build an initial schedule able to
absorb disruptions with minimum effect on already elective cases. Doing so, they define two sets

51
Chapter II. Background and related works.

of performance indicators comprising robustness (the ability to maintain performance despite

disruptions) and stability measures (the ability of a schedule to not deviate despite disruptions).
(Ceschia and Schaerf 2016) propose a dynamic patient admission scheduling in which they
integrate the constraints brought by the surgery scheduling. To do so, they include constraints on
the utilization of ORs for patients requiring surgery: each OR capacity must be respected, and each
OR is only assigned a set of surgical specialties. They consider elective and non-elective patients.
(Ballestín, Pérez, and Quintanilla 2019) build a scheduling and rescheduling model for elective
cases that minimizes the percentage of late patients. First, they build a tentative schedule two
weeks before the planning period with the objective of minimizing the number of tardy patients.
Second, a few days before schedule execution, they rebuild the schedule based on the changes that
happened since the first scheduling. They propose several strategies based on the specificities of
elective cases to tackle the changes from the first to the second schedule.
As a synthesis for this part, we propose Table 9 extracted from (Kamran, Karimi, and Dellaert
2020).

In our study we focus on the disruption management at the offline and at the online
operational level.

52
 Chapter II. Background and related works.

Table 9 - Review of Predictive and Reactive Disruption Management for the surgery scheduling and the disruption management problems. Taken from (Kamran,
Karimi, and Dellaert 2020). The abbreviations stand for planning (P), scheduling (S), replanning (RP), rescheduling (RS), patient booking strategy (PBP), DM
(disruption management), reactive disruption management (RDM), predictive and reactive disruption management (PRDM).

Patient Model/Problem
P and and/ RP and/
Reference DM
or S or RS PBP
approach Elec. Emerg. Deter. Stoch.
Stuart et al. (2010) x O1 RDM x x x
Ceschla and Schaerf (2014) x x B1 RDM x x x
Bruni et al. (2015) x x B RDM x x x
Dios et al. (2015) x x O RDM x x
Addis et al. (2016) x x B PRDM x x x
Ballestin et al. (2019) x x B RDM x x
Stuart and Kozan (2012) x O1 RDM x x x
He and Xiang (2013) x O RDM x x x
Zhang et al. (2014) x x O RDM x x
Nouaouri et al. (2011) x O RDM x x x
Erdem et al. (2012) x O1 RDM x x x
Shu and Subbaraj (2015) x B2 RDM x x
Heydari and Soudi (2016) x x O PRDM x x x x
Soudi et al. (2019) x O PRDM x x x
Akbarzade et al. (2019) x B2 RDM x x x
Kamran et al. (2019) x x MB PRDM x x x
Kamran et al. (2020) x x MB RDM x x x
O The way they approach to patient booking policy can be inferred a type of block scheduling policy as well.
1

B1 The way they approach to patient booking policy can be inferred a type of modified block scheduling policy as well.
B2 The way they approach to patient booking policy can be inferred a type of open scheduling policy as well

53
Chapter II. Background and related works.

3. A digital twin for the surgical suite based on discrete event simulation

3.1. Discrete-event simulation

Simulation is “the imitation of the operation of a real-world process or system over time” (Banks,
n.d.; Robinson 2004). Discrete-Event Simulation (DES) is a computer-modeling technique used to
model the behavior of a system as a sequence of events in time. The key components of a DES
mode include entities, events, a simulation clock, queues, resources, activities, and processes. We
give a brief definition of each of these components:
- An entity is an object that move through the modeled system (example: a patient admitted
for surgery).
- An event is an instantaneous occurrence that change the state of the system (example: a
patient arrives in the surgical suite; a patient exits the surgical suite).
- The simulation clock is a virtual clock that tracks the simulation time progression. It
determines the order of occurrence of each event.
- Queues allow to model waiting entities (example: a patient waiting room).
- Resources are the required componaents for the system to perform its action (example:
a surgeon, a nurse).
- An activity is considered as the smallest unit of work. It has a finite duration, and it is
assumed to provoke state changes at its start and at its end (example: a procedure, a
reversal).
- A process is a sequence of activities and events in a chronological order that allows to
model all or part of the behavior of the targeted system (example: the preoperative care
process).
Since the 1950s, the research community has shown great interest in the use of DES in fields such
as manufacturing, supply chain management, military operations, computer, and network design.
In a standard manufacturing DES model, raw materials are introduced into the system and are
directed to queues for processing. Then, they undergo transformation and/or consumption
through different processes, before leaving the system in the form of finished goods. A similar
modeling can be applied to the healthcare sector. First, a sick or injured patient enters a hospital
and is sent to a waiting room. Then, the patient is then treated (transformed) via various processes
(surgery, radiological exam, etc.), before exiting the hospital as a healthy patient (Forbus and
Berleant 2022).
DES has been used in the healthcare sector for more than 20 year (Evans, Unger, and Gor 1996)
and several reviews are available (Thorwarth and Arisha 2009; Günal and Pidd 2010; S. Liu et al.
2020; X. Zhang 2018; Vázquez-Serrano, Peimbert-García, and Cárdenas-Barrón 2021; Forbus and
Berleant 2022).
(Forbus and Berleant 2022) reviews five years of literature (2017-2021) and show that not only the
interest in DES in healthcare increases by the year, but it also expands to new areas of the healthcare
system. They identify four primary purposes of DES use in the healthcare area: disease progression
management (study the long-term economic impact of treatments), health screening protocols

54
 Chapter II. Background and related works.

(study of the effects of properly orienting patients towards specific care pathways), health behavior
modeling (study of diseases caused or exacerbated by personal lifestyle choice), and healthcare
system operations (the equivalent of traditional manufacturing operations management in the
healthcare sector). This later use of DES, which focuses especially on resource utilization,
scheduling and capacity planning is the one that interests us in this study. (Günal and Pidd 2010)
review the use of DES for performance modelling in the healthcare sector. They focus on care
provided by hospitals such as outpatient, inpatient, day-case and emergency care. They address the
attempts at building whole hospital simulations and the challenges it brings. They conclude that
most articles are unit specific, facility specific and can only be used at an operational level.
Some these reviews highlight the fact that DES is also used in the surgical suite. Indeed, many
articles can be found (Marcon and Dexter 2006; Gul et al. 2011; Saremi et al. 2013; Lehtonen et al.
2013; Niu et al. 2007; Ma and Demeulemeester 2013; Peng, Qu, and Shi 2014; J. Brown et al. 2014;
Baesler, Gatica Fuentes, and Correa 2015; van der Kooij, Mes, and Hans 2014; Saadouli et al. 2015;
W. Xiang, Yin, and Lim 2015a; Bam et al. 2017; Koppka et al. 2018). Subjects of interests include
the assessment of the performance of OR management strategies (Schoenfelder et al. 2021; Persson
et al. 2017; Allen, Taaffe, and Ritchie 2014; M’Hallah and Al-Roomi 2014) and OR scheduling
(Schultz and Claudio 2014; Ewen and Mönch 2014; S. Wang et al. 2016; Roshanaei et al. 2017a).
Below, we briefly discuss a few other articles where DES is applied to solve issues related to the
surgical suite.
(Bovim et al. 2020) proposes to solve the MSSP with a simulation-optimization. First, they use
two-stage stochastic optimization model is used to develop a MSS. Second, they use DES to test
the MSS in a stochastic environment with uncertainties related to the surgery duration and the
hospital length of stay, as well as to provide scenarios for the optimization model.
(Yahia et al. 2017) develop a Design and Engineering Methodology for Organization-based
simulation model to provide a more comprehensive view of the planning and scheduling problem
in the surgical suite. They their DES model with AnyLogic and they use it (1) to assess the
operational performance of the CMP and MMSP, (2) as a simple process and ontological
representation.
(Duma and Aringhieri 2015) propose to use simulation-based optimization to assesses whether a
case with at risk of going overtime should be cancelled or assigned overtime. They consider elective
and non-elective cases, the impact of the training level disparity among surgical teams, as well as
uncertainties on patient arrivals, patient length of stay, and surgery durations. Both patient-centered
and facility centered indices are used to assess the performance.
(Z. Zhang and Xie 2015) use a simulation-based optimization model to tackle the appointment
scheduling problem in a mutli-OR suite. They consider uncertainties on surgery durations and
assess performance through the costs generated by surgeon waiting time, OR idle time and OR
overtime.

55
Chapter II. Background and related works.

3.2. Digital twin

The notion of “digital twin” is not a recent development. Its origins track back to the 1960s
when NASA’s Apollo space program introduced the idea of “twin” space vehicles. NASA built
two identical space vehicles, deployed one into space and kept the other on Earth to replicate and
predict the behavior of its space twin (Boschert and Rosen 2016). In 2003, Michael Grieves
introduced the concept of digital twin as we know it today: “a digital equivalent to a physical
system”21. However, it is only in 2010, that NASA provided the first definition of a digital twin
within the context of their own field: “an integrated multi-physics, multi-scale, probabilistic
simulation of a vehicle or system that uses the best available physical models, sensor updates, fleet
history, etc., to mirror the life of its flying twin”22.
Today, the development of advanced information technologies such as IoT has opened doors for
a new generation of digital twins. Although the scientific literature abounds with literature
reviews on the digital twin, there is still not a single, widely accepted definition of what is
a digital twin (Jones et al. 2020; Semeraro et al. 2021; M. Liu et al. 2021; VanDerHorn and
Mahadevan 2021). For instance, (Grieves and Vickers 2017; Abramovici, Göbel, and Dang 2016;
Rosen et al. 2015) describe a DT as “a set of virtual information constructs that fully describes a
potential or actual physical manufactured product from the micro atomic level to the macro
geometrical level”, while (Schluse and Rossmann 2016; Schroeder et al. 2016) defines it as “an
integrated multiphysics, multiscale simulation of a vehicle or system that uses the best available
physical models, sensor updates, fleet history, etc., to mirror the life of its corresponding flying
twin”.
Some common characteristics of what a digital twin should still be inferred from this pool
of available definitions: a digital twin must be individualized, high-fidelity, real-time, and
controllable (M. Liu et al. 2021). In other words, a digital twin must (1) be the virtual replica of a
designated physical twin, (2) be able simulate the physical twin’s behavior in real-time or almost
real-time, and (3) allow a bidirectional communication with the physical twin. Furthermore, in
Novembre 2021, the International Organization for Standardization published the norm ISO
23247-1:202123 where they provide an overview and general principles of a digital twin framework
for manufacturing (Shao, Frechette, and Srinivasan 2023).
Note that an offline digital twin describes the case where real-time communication is not critical.
Rather, an offline digital twin would connect to the physical system periodically. This periodic
connection is an important distinction between an offline digital twin and a traditional simulation
model (Shao and Helu 2020).

21 [Link]
[Link]
22 [Link]
23 [Link]

56
 Chapter II. Background and related works.

Within this study, we define a digital twin as an individualized and high-fidelity virtual
replicate of a physical twin (a product, a process, or an organization). The digital and the
physical twins exchange data via a bidirectional communication canal. The physical twin
provides data so that the digital twin can replicate its behavior in real or near-real time; the
digital twin provides data to the user and the digital twin so that decisions can be made in
the real-world.

The industrial applications of digital twin span across the different lifecycles phases: design
(F. Xiang et al. 2019; Lutters 2018; Caputo et al. 2019), manufacturing (Z. Zhu, Liu, and Xu 2019;
Leng et al. 2019; Knapp et al. 2017), service (Aivaliotis et al. 2019; Xie et al. 2019) and retire (X. V.
Wang and Wang 2019).
Although less developed than in the industry, digital twins are also used in healthcare.
(Erol, Mendi, and Doğan 2020) define a digital twin as a “digital replica that allows modeling the
state of a physical asset or system”. They propose the digital twin of a patient which shows the
same physical characteristics and changes that the real patient. Their tool can be used for diagnosis
and treatment process monitoring. (Y. Liu et al. 2019) propose a framework of the cloud healthcare
system based on digital twin (CloudDTH) in order to monitor, diagnose, and predict aspects of
individual health. Their tool specifically targets the elderly population. (Elayan, Aloqaily, and
Guizani 2021) define a digital twin as “a virtual replica of a physical asset that reflects the current
status through real-time transformed data”. Their article proposes and implement an intelligent
context-aware healthcare system using the DT framework.
The above articles describe patient focused digital twin for monitoring individual and diagnosing
health problems. However, it is possible to apply this diagnosis and monitoring concept to the
healthcare organization itself. (Abdallah Karakra et al. 2018; 2019; A. Karakra et al. 2020; Abdallah
Karakra 2021; Abdallah Karakra et al. 2022) build a discrete event simulation based digital twin for
real-time monitoring and near-future prediction of patient pathways in the hospital. (Obinna C.
Madubuike and Anumba 2022; Song and Li 2022; Obinna C. Madubuike and Anumba 2023;
Obinna Chimezie Madubuike, Anumba, and Agapaki 2023) tackles the discuss the application of
digital twin in healthcare facility management.

In our case, we propose to build the digital twin of a multi-OR surgical suite. Our digital
twin is thus the virtual replica of a group of intertwined processes that aim at answering
patient demand (surgical interventions) while using specific resources (the operating rooms
and the staff). Note that the study of material resources other than the operating rooms are
out of scope. The digital twin is used as a decision-support system on a daily basis and thus,
the daily input of data is equivalent to having real-time data input for an online digital twin.
We make the hypothesis that we can build the digital twin of a surgical suite using Discrete
Event Simulation.

57
Chapter II. Background and related works.

4. Chapter synthesis
In this three-section Chapter, we have discussed the state of the art surrounding our problematic
and defined our study perimeter. The goal of this chapter was to bridge between the on-site
observed problematics and the current advancements and solutions provided by the scientific
community to define our study perimeter.
To begin with, we have described the planning and scheduling problem linked to the working of
surgical suite. Then, we discussed the reactive and the predictive levels of the disruption
management problem. After that, we addressed the concept of digital twins, how they can be used
in both the healthcare and industrial sectors. Finaly, we presented a brief review on the use of
discrete event simulation in the healthcare environment, and more precisely, in the surgical suite.
Consequently, we have set our study aim as follows: providing a decision-support system for the
operational decision level to deal with both the reactive and predictive disruption management
problem. To do so, we propose to build a discrete event system digital-twin-based decisions
support system (DT-DSS). For this purpose, we use Flexsim Healthcare®: a 3D simulation and
data analysis software.
Our choice of discrete event simulation is justified by the fact that it:
• Is adapted to complex system modeling: we can model a multiplicity of resources and
processes (patient pathways) in parallel.
• Inherently respects resource constraints. For instance, if a patient acquired a surgeon, the
surgeon will not be able to intervene on another patient at the same time.
• Is compatible with the computation of KPIs and the construction of dashboards, and thus
with performance analysis.
• Allows to model and simulate operating schedule execution.
• Is compatible with the implementation of an experimenter. This allows to easily create
what-if scenarios. An experimenter allows to easily configurate scenarios for multiple
replications – meaning to integrate stochasticity – and to compare the global KPI for all
the replications, or the KPI for a specific replication.
• Is compatible with the implementation of an optimizer.
• Allows to perform risk assessment, and sensitivity analysis.
• Allows flow visualization and provides a dynamic visual interface (we can see the system
evolution across time).
• Can provide a pedagogical environment. It can enhance non-expert user trust in the tool
with the visualization aspect. This is especially important as our application field is the
healthcare area. 3D figures are easier to understand than mathematical equations.
• Allows uncertainties modeling such as duration variability and non-elective arrivals.
• Provides a risk-free environment and a training environment.

58
 Chapter II. Background and related works.

PART 2 - PROPOSED
METHODOLOGY AND TOOL

59
Chapter III. Solution proposal

CHAPTER III. SOLUTION PROPOSAL

In Chapter II, we have described the four different planning and scheduling decision levels of a
surgical suite. First, the long-term strategic level deals with the capacity planning, the capacity
allocation, and the case-mix problems. Second, the medium-term tactical level tackles the master
surgery schedule problem (MSSP). Third, the short-term offline operational level related to the
surgery scheduling problem (which can be divided into the advanced and the allocation scheduling
problems). Fourth and finally, the real-time online operational level consists in the schedule
execution problem. We have then discussed the disruption management problem that takes place
at the operational levels. It is divided into two main issues: (1) in predictive disruption management
(PDM), disruptions are anticipated and dealt with before they occur; (2) in reactive disruption
management (RDM), the disruptions are only tackled after they have been already realized.
Upon studying the on-site and scientific problematics linked to the surgical suite, we have decided
to focus on the disruption management at the operational level and have thus made the following
hypotheses: we consider the allocated resources (strategic level), the master surgical schedule
(tactical level) and the provisional schedule (offline operational level) to be fixed and staff-validated
in advance. In other words, we do not wish to build nor to improve the provisional schedule. This
has led us to address the following general research question: “How can we promote and maintain
the performance of a surgical suite’s organization under uncertainties?” which we have divided into
5 more specific research questions:
Q1. How can we anticipate disruption before schedule execution?
Q2. How can we assess the quality of the predictive disruption management?
Q3. How can we assess the quality of the reactive disruption management?
Q4. How can we identify whether performance lack stems from offline or online decisions?
Q5. How can we train OR managers to disruption management?
This has required us to specify the study perimeter. Thus, we have chosen to:
• Assess the performance using 3 KPI’s: (1) the maximum patient waiting time in the surgical
suite, (2) the average utilization of the surgical suite’s ORs and (3) the average staff
overtime in the surgical suite.
• Focus on three types of uncertainties: non-elective patient arrivals, activity duration
variability and resource availability.
• Study one type of online operational decision: the scheduling of non-elective patient
arrivals.
• Use Discrete Event Simulation and Digital Twins to tackle our research questions.

60
 Chapter III. Solution proposal

To provide decision support on online operational decisions, we can either:

(1) Before schedule execution (offline tool): anticipate what would be the best online
operational decision strategy to follow based on the specific provisional schedule.
(2) During schedule execution (online tool): provide insights into what online
operational decision to take in real-time.
(3) After schedule execution (offline tool): provide feedback on the online operational
decisions that were taken during schedule execution.
Either way, the principal function of our tool will be: “Maintain and improve the performance of
a surgical suite’s organization under uncertainties”. Note that the term “maintain” is
particularly important. We make the hypothesis that the provisional schedule is staff-validated
and fixed. Thus, we cannot change the provisional schedule - even if we show before schedule
execution that the performance targets may not be reached. This means that, with certain
provisional schedules, it might be impossible to obtain a performant performed schedule. In other
words: the fact that the performed schedule is not performant does not always mean that the online
operational management was not performant; it could also indicate that the provisional schedule
was suboptimal.
Example: if the provisional schedule utilization is at 95% and several non-elective cases are added
to the schedule, the overtime resulting from it might not be avoidable.

Based on our literature review and on-site observations, we have decided to propose a
digital twin-based offline decision support system to improve the decisions taken at the
online operational level. This means that we propose a decision support system that directly
supports the analysis of the provisional and the performed schedules, and that thus
indirectly prepares the schedule execution and provides feedback on it. Indeed, to the best
of our knowledge, the offline decision support system we propose does not exist today.
Indeed, in most surgical suites, the staff has access to an OR software to build the
provisional schedule and to extract data to retrospectively compute KPIs. However, there
has not been any tool developed to anticipate and assess different decision strategies to
follow during the real-time execution of the schedule, or to provide feedback on how the
schedule execution was carried out.

In the remainder of this Chapter we describe a methodology for a prospective analysis of the
provisional schedule (contribution #1), a methodology for a retrospective analysis of the
performed schedule (contribution #2), and a methodology for real-time decision-making virtual
training destined to OR managers (contribution #3).
In Chapter IV, V and VI, we propose as a proof of concepta study case in which we apply the three
first methodologies to a surgical suite inspired by the Hôpital Privé de La Baie. To do so, (1) we
develop a digital twin-based decision support system for the prospective and retrospective
simulation and analysis of the operating room schedule execution, and (2) we infer from this
prototype a standardized methodology to conceive and build such a tool in any surgical suite
(contribution #4). We describe this contribution #4 in the last section of the Chapter.

61
Chapter III. Solution proposal

1. Prospective analysis of the provisional schedule

In this section, we propose a methodology for a prospective analysis of the provisional schedule.
The objectives are to anticipate disruptions before schedule execution (Q1), and to assess the
quality of the predictive disruption management (Q2). This methodology corresponds to our
contribution #1.
We propose a five-step analysis of the provisional schedule. To begin with, we simulate the
provisional schedule execution to assess the resource synchronization (step #1) and compute the
schedule performance (step #2). Then we consider the impact of stochastic durations on the
performance to assess the schedule robustness (step #3). After that, we include additional non-
elective arrivals, and compute the schedule performance for different scheduling strategies (step
#4). Finally, we simulate the provisional schedule execution in a stochastic environment that has
both stochastic durations and stochastic non-elective arrivals to assess both the robustness and
resilience of the provisional schedule at the same time (step #5).
We synthesize these four steps in Table 10. Each row corresponds to one analysis step, and each
column corresponds to a model parameter.
Table 10 – Prospective analysis steps

Additional Non-
# Objective Activity durations
elective arrivals

1 Assess feasibility (resource synchronization) Deterministic No

2 Compute Performance Deterministic No

3 Assess robustness Stochastic No

4 Assess resilience Deterministic Yes

Assess the best non-elective case scheduling

5 Stochastic Yes
strategy based on the provisional schedule

A detailed study case will be presented in Chapter VIII. We briefly describe each of these steps
below:
• Step 1 - Resource synchronization. The material and human resources are synchronized
if it made such that the patient does not have to wait for them. For example, there will be
waiting time if: (a) Two patients are scheduled for surgery in the same OR at the same time
(therefore material resource constraints are not respected), or (b) Two patients are
scheduled for surgery with the same surgeon at the same time (therefore human resource

62
 Chapter III. Solution proposal

constraints are not respected). We discuss the modeling and simulation of resources in a
surgical suite digital twin in Chapter IV.
• Step 2 – Performance. We run the provisional schedule in a deterministic environment
and compute the KPIs. We compare each of them to the ANAP targets. The KPIs’
description is available in Chapter I.
• Step 3 – Robustness. For the robustness (definition in Chapter 1 we run the provisional
schedule with stochastic durations and compute the gap between the new KPI values and
the ones of step 2, and to the ANAP targets.
• Step 4 – Resilience. For the resilience (definition in Chapter 1), we run the provisional
schedule using deterministic durations, and we test different combinations of non-elective
arrival scenarios with non-elective scheduling scenarios. The resulting KPIs are compared
with the ones of step 2, and to the ANAP targets.
• Step 5 – Simulating the provisional schedule execution. We run the provisional
schedule in a stochastic environment where the durations are variable and where there are
non-elective arrivals, and we test the non-elective scheduling strategies. We compare the
KPIs with the ones of step 2, and to the ANAP targets.
As a side note, the results of the provisional analysis can be used as feedback for the offline
operational decisions (surgery scheduling problem), although this was not our primary goal.

63
Chapter III. Solution proposal

2. Retrospective analysis of the performed schedule

In this section, we propose a methodology for a retrospective analysis of the performed schedule.
The objectives are to assess the quality of the reactive disruptions management (Q3), and to identify
whether performance lack stems from offline or online decisions (Q4); we aim at providing
feedback on the schedule execution that led to obtain the performed schedule. This methodology
corresponds to our contribution #2.
In this section, we propose a three-step methodology for a retrospective analysis of the performed
schedule: (1) replay the performed schedule as-is (i.e. in a deterministic environment) to assess
whether the resource constraints are respected in the performed schedule, (2) assess the performed
schedule performance, and (3) identify the performance gap between the performed and the
provisional schedule and assess whether they are a consequence of offline or online operational
decisions. We synthesize these three steps in Table 11. Each row corresponds to one or several
analysis steps, and each column corresponds to a model parameter.
Table 11 - Retrospective analysis steps

Activity Additional Non-

# Objective
durations elective arrivals

1 Compute Performed Schedule Performance Deterministic No

Assess the impact of implementing other

2 Deterministic No
decisions to tackle the disruptions.

3 Identify performance gap root causes Deterministic No

A detailed study case will be presented in Chapters VII and VIII. We briefly describe each of these
steps below:
• Step 1 – Performed schedule performance. We run the performed schedule in a
deterministic environment and compute the KPIs. The KPIs’ description is available in
Chapter I.
• Step 2 – Test other decisions. We simulate the performed schedule in a deterministic
environment to assess the impact of different solutions to tackle the disruptions. For
instance, we test other scheduling decisions for non-elective arrivals. We compute the KPIs
and compare them with the ones of step 1.
• Step 3 – Performance gap root causes. The performed schedule performance is strongly
dependent on the provisional schedule performance. For instance, if there is scheduled
overtime in the provisional, there is a high probability that there will be overtime in the
performed schedule too. Consequently, we identify the performance gap and study whether
they result from improvable surgery scheduling (offline level), or from the real-time OR
management (online level).

64
 Chapter III. Solution proposal

3. A training environment for the OR manager

In this section, we propose a methodology for training an OR manager on disruption management.
This answers our research question: how can we train OR managers on disruption management?
(Q5). This methodology corresponds to our contribution #3.
We have described a decision support system able to model the execution of an operating suite
schedule in a stochastic environment (duration variability and additional non-elective arrivals) while
following a specific OODS. But what if we allowed the user to test their own personal non-elective
schedule strategies themselves? We propose to use this virtual environment to train the OR
coordinator on schedule execution.
We represent this in Figure 20.

Figure 20 - Complete framework

65
Chapter III. Solution proposal

4. Proposition of a standardized method to build a surgical suite digital twin

We propose a 5-step standardized method to build the digital twin of a surgical suite using discrete
event simulation. We illustrate it in Figure 21, and describe its steps below.

Figure 21 - Illustration of the steps of a standardized method to build to build a surgical suite digital twin.

1 – Gather data - Qualitative. The first step consists in gathering data related to the targeted
surgical suite. To do so, we perform interviews with the surgical suite staff and perform on-site
observations. The objective is to have a deep understanding of how the specific suite works.
Indeed, although all suites share some common processes, they also all display some local solutions
that can be interesting to model. For instance, the process to ask the stretcher-bearers to bring the
next patient to the surgical suite usually changes from one suite to the other. First, we recommend
speaking with at least one representative of each profession in the suite, namely: a surgeon, an
anesthesiologist, an OR nurse, an anesthesiologist nurse, a nurse assistant, and a PACU nurse. It

66
 Chapter III. Solution proposal

can be interesting to understand how services in contact with the surgical suite function as well (i.e.
the inpatient and outpatient wards, the sterilization, the stretcher-bearer services, etc.). Second, we
also suggest starting by mapping the three operational suite processes (the preoperative,
perioperative and post-operative phases), then to move on the supporting and decision-making
processes related to them. Finally, it is interesting to assist in the surgery scheduling meetings and
to exchange with the OR manager or the staff responsible for the reactive management disruption.
1 – Gather data - Quantitative. We extract quantitative data from the OR software database. This
step can be delicate as patient data are sensitive and must be anonymous. At the end of this first
step, we propose a tentative study perimeter in terms of the process, resources, uncertainties,
disruptions, decisions, and KPIs are all considered.
2 – Treat and analyze data. This step focuses on the data extracted from the OR software
database. They usually include timestamps which are a must when modeling the surgical suite
activity. These timestamps can however be incorrect and must be corrected (Chapter V). It might
be possible that the data available in the database does not allow to study the perimeter proposed
in the first step so some adjustments can be made.
3 – Build a determinist model. Once we have all the data, we build the deterministic model of
the studied surgical suite. First, we map the observed process. Second, we add the fixed resources.
Third and finally, we build the KPI dashboard.
4 – Implement uncertainties. Depending on the study perimeter, we implement uncertainties in
the surgical suite processes. This include thinking of the number of replications required for each
stochastic scenario to be representative of reality.
5 – Develop the decision support system. As for step 4, we implement the decisions based on
the perimeter that was previously identified. We suggest proposing two decision-making modes to
the user: a manual mode in which the user can configure the model reaction to any disruption in a
flexible way, and an automatic mode in which the DT-DSS tries many already modeled disruption
management strategies.
This tool (A) models schedule execution of either a provisional or a performed schedule, (B) in
either a deterministic or a stochastic environment, while (C) respecting the resources and the
patient constraints, (D) applying online operational scheduling strategies, and (E) computing KPIs.
In the following Chapters, we describe how we develop the DT-DSS for our specific study case.
In Chapter IV, we describe how we model the schedule execution based on the operational
processes (patient pathway), and the supporting processes (sub-functionalities A & C). In Chapter
V, we detail how we model stochastic decisions (sub-functionality B). In Chapter VI, we explain
how we model stochastic additional non-elective arrivals and the scheduling strategies to add them
to the ongoing schedule (sub-functionalities B & D).

67
Chapter III. Solution proposal

5. Chapter synthesis
In this Chapter, we have presented our solution proposal: a methodology for prospective analysis,
a methodology for a retrospective analysis, a training methodology and a standardized method to
build a surgical suite digital twin. In the following chapters of this part II, we will describe how we
build a prototype to apply these first three methodologies.

68
 Chapter IV. Modeling and simulation of the schedule execution

CHAPTER IV. MODELING AND SIMULATION OF THE SCHEDULE

EXECUTION

In the previous Chapter, we have proposed a framework for the continuous improvement of online
operational decisions in a surgical suite, and we have justified that a digital-twin based decision
support is a satisfactory tool to perform all the functionalities required by this framework. We
developed a digital twin-based decision support system for the prospective and retrospective
simulation and analysis of the operating room schedule execution, and we inferred from this
prototype a standardized methodology to conceive and build such a tool in any surgical suite.
In this new Chapter, we describe how we model and simulate the schedule execution for
either a provisional or a performed schedule. We use the available information in the OR
database to model the surgical suite’s processes, while respecting the constraints imposed by
patients and the availability of resources. Note that lack of data in the OR database is a constraint
that one might face in most of the existing surgical suites.
At this stage of the study, we consider a deterministic environment and we do not model
uncertainties. Consequently: (1) the durations are deterministic, (2) there are no additional non-
elective arrivals and (3) we only consider operational and supporting processes. Indeed, since we
are modeling the execution of schedules in a deterministic environment, all the events are known
in advance and there is no need to make decisions throughout the execution.
Our objective is to model the execution of either a provisional or a performed schedule. Ideally,
we would like a model that considers the totality of the operational processes, the supporting
processes, the human resources, and the material resources. However, as we show in Table 12, we
only have access to 7 patient pathway timestamps as well as the surgeon, the anesthesiologist, and
the operating room IDs. We therefore limit our modeling process to what can be derived from the
available data.
In this Chapter, we first propose a solution to model schedule execution despite a lack of data. We
use and justify different models for provisional schedule execution (detailed process with strict
constraints on resources) and performed schedule execution (aggregated process with flexible
constraints on resources). Second, we illustrate this method with our study case. We provide a
synthetic conclusion at the end of the Chapter.

69
Chapter IV. Modeling and simulation of the schedule execution

Table 12 – Difference between the required and the available information in our study case

Type of information Required Information Information available in the database

Start time and end time for all the activities of the patient pathway; the
activities being:
A - For surgery with induction in the OR: Patient Reception, Patient
Checklist, Setup, Induction, Procedure, Reversal, Move to PACU, PACU
Patient pathway Monitoring For all patients: Suite Entry, OR Entry,
timestamps Incision, Suture, OR Exit, PACU Entry,
B - Surgery with induction in the PACU: Patient Reception, prepare
patient for Induction, Induction, Wait for anesthesia to work, Patient PACU Exit
Checklist, Setup, Procedure, Reversal, Move to PACU, PACU Monitoring
C - Surgery without the need of the anesthesia team: Patient Reception,
Patient Checklist, Setup, Procedure, Reversal, Move to PACU, PACU
Monitoring

Supporting process
OR cleanup start, OR cleanup end, etc. /
timestamps

Surgeon ID, Anesthesiologist ID, All nurses ID (OR, anesthesia, PACU),

Human Resources Surgeon ID, Anesthesiologist ID
Nurse Assistant(s) ID, Stretcher-bearers ID

OR ID, Preoperative bed ID, PACU bed ID, Other material resource ID:
Material Resources transportation means surgical instruments, surgical supplies, imaging and OR ID, Preoperative bed type
visualization equipment, surgical support equipment.

70
 Chapter IV. Modeling and simulation of the schedule execution

1. Proposal of different modelling approaches depending to available data

1.1. Surgical suite process

The surgical suite’s process (SSP) is complex and require the intervention of various professions
and equipment at different times and locations. Since it is common to not have the data required
to model them available in hand, let “aggregated SSP” be a model based on the available
timestamped data in the OR software, and let “detailed SSP” be a model based on the
available timestamped data in the OR software and our on-site observations. These SSP
models differ in terms of the exhaustivity of the process mapping and in terms of timestamps’
value plausibility: the aggregated SSP includes less steps, but has more plausible timestamps value
than the detailed SSP.
We illustrate the difference between an aggregated process and a detailed process for the
preoperative phase of a surgery requiring an LRA in Figure 22. The aggregated process (blue)
consists in one activity, while the detailed process consists in 4 activities (green). The aggregated
process relies on timestamps recorded in the database (blue), while the detailed process relies on
timestamps recorded in the database (blue) and on timestamps that are not recorded in the database
(green). Therefore, to simulate a detailed process, we need to estimate the value of the missing
timestamps (green) based on on-site observations.

Figure 22 – Illustration of the difference between an aggregated process (top) and a detailed process
(bottom) for the preoperative care of a surgery requiring an LRA.

We propose different approaches for the provisional and the performed schedules because their
simulation have different goals.

71
Chapter IV. Modeling and simulation of the schedule execution

The objective when simulating the performed schedule execution is to acquire a simulated
execution which is as close as possible to the performed schedule execution. (i.e. the performed
and the real patient room entry times are identical), then to exhaustively model the activity (i.e.
model the different steps for the whole preoperative patient pathway).
On the other hand, the provisional schedule is naturally based on estimations and hypotheses. For
instance, timestamps and durations are estimated because they are not known in advance and the
resource synchronization is assessed in an empirical manner. On the other hand, the simulating the
surgical suite schedule exhaustive is a priority as we are trying to where and when the schedule
could be disrupted (gives a better idea).
Consequently, we model the provisional schedule execution using the detailed SSP, and
we model the performed schedule execution using the aggregated SSP.

1.2. Resource constraints

To exhaustively model the surgical suite’s processes, we need information on the resources used:
what resource is needed? When? Where? For how long? However, as for the timestamps, the
information related to resources is not always available in the OR database. Furthermore, it is
common for the staff to not always follow the best practices (e.g. joint patient reception by the OR
nurses and the anesthesia nurse) or to adapt their process to the ongoing state of the surgical suite
(e.g. during a slow day, the anesthesiologist can perform all their patients’ induction while during a
busy day they might let the anesthesia nurse induce patient).
We illustrate this in Figure 23 for the perioperative phase of a patient requiring induction in the
operating room: in the best-case scenario (left), the surgeon, the anesthesiologist and the nurses
are all present from the “room entry” to the “room exit”. In reality (right), it is possible that the
surgeon is only present for the procedure and that the anesthesiologist completely delegates the
induction and the patient monitoring to the nurse anesthetist. Thus, applying resource constraints
to the SSP is a complex task.
Let “strict resource constraints” be the (possibly incomplete) constraints described in the OR
database and let “flexible resource constraints” be the constraints on the operating room only.
During the schedule execution of the performed schedule, we consider the recorded timestamps
to be representative of the reality of what actually took place and must therefore be respected
during the simulation. As we mentioned, applying resource constraints is tricky. Since the resource
that is less likely to be inaccurate is the OR ID, we use flexible resource constraints during the
schedule execution of the performed schedule. When we prospectively simulate the provisional
schedule execution, we make the hypothesis that the processes are perfectly performed, thus we
use strict resource constraints for the provisional schedule execution.

72
 Chapter IV. Modeling and simulation of the schedule execution

Figure 23 – Illustration of the difference between the theoretical best situation (left) and an example of
real situation (right) of the human resource presence during the perioperative phase of a patient requiring
induction in the operating room.

73
Chapter IV. Modeling and simulation of the schedule execution

2. Application to our study case

2.1. Introduction
In the previous section, we proposed a method to model the surgical suite organization despite the
missing data in the OR software database. We have suggested to use a detailed process with strict
constraints on resources for the provisional schedule execution, and to use an aggregated process
with flexible constraints for the performed schedule execution.
In this section, we first describe how these two types of resource constraints can be translated to
our study case. Second, we present the aggregated and the detailed processes we use. Third and
finally, we add technical notes on specific hypotheses we had to make because of our simulation
tool and study case.

2.2. Flexible and strict resource constraints

As we mentioned in Table 12, for each patient in the schedule, the OR database provides us with
information related to human resources (surgeon ID and anesthesiologist ID), and material
resources (OR ID and preoperative care bed type). Depending on the resource constraints we apply
to the surgical suite’s process, we can decide whether or not resources are: considered in the study,
nominative or non-nominative, and limited or unlimited.
To model resources in our model, we create entities that we gather in a “resource group”. Thus an
instance of the resource groups is any entity part of that group. Examples of resource groups are
surgeons, anesthesiologists, PACU beds…

2.2.1. Resources to be included

In agreement with our expert committee (P1C3), we decided to simplify the study and not consider
all the material resources:
H1. The study only considers the OR, the preoperative care location beds, and the PACU beds as material
resources. Therefore, we do not consider the following material resources: transportation means, surgical
instruments, surgical supplies, imaging and visualization equipment, and surgical support equipment.
Based on our on-site observations, we know that the anesthesia process in the OR can involve (1)
the anesthesiologist alone, (2) the anesthesia nurse alone, (3) both at the same time. Since the OR
database does not specify which option is followed, we make the following hypothesis:
H2. The study considers all surgical suite human resources except the anesthesia nurse: the surgeons, the
anesthesiologists, the OR nurses, the PACU nurses, the OIP nurse, the nurse assistants, and the stretcher-
bearers.

74
 Chapter IV. Modeling and simulation of the schedule execution

Based on our on-site observations, we know that the OR cleanup is performed by either (1) nurses
that were in the OR, (2) assistant-nurses, (3) both at the same time. Since the OR database does
not specify neither the cleanup start and cleanup end timestamps nor which option is followed, we
make the following hypothesis:
H3. There is one assistant-nurse per OR that systematically performs the OR cleanup once the patient exits the
OR.
Based on our on-site observations, we know that the OR nurses are systematically assigned an OR,
and thus the patients that are scheduled within. They are also assigned a role (circulating, scrub, or
instrumentist). However, since the OR database does not mention this information, we make the
following hypothesis:
H4. Each OR is assigned one OR nurse. This nurse is assigned to all patients assigned to that specific OR.
H5. The nurse schedule is identical to the OR schedule.

2.2.2. Resources to be modeled as nominative

We make the difference between nominative and non-nominative resources. Resources are
nominative when they are linked to a unique ID. When a resource is nominative, the patient can
be only cared for by the assigned resource in the schedule. When a resource is not nominative, the
care is provided to the patient by any resource of the same type available at that point.
Using nominative resources in a model is more constraining than using non-nominative ones: in
the first case you can only pick one specific instance of the resource group, while in the second case
you can pick any instance of the resource group. Using nominative resources when possible in the
model also provides a simulation closer to reality: the specific resource that was used in real-life is
used in the model.
In real life, the surgeon, the anesthesiologist, the anesthesia nurse, and the OR nurse know to which
OR they are assigned and thus which patient they are assigned to as well: they are all nominative
resources. However, in the OR database, only the surgeon and the anesthesiologist IDs are
mentioned, thus:
H6. Surgeons and anesthesiologists can be modeled as nominative resources.
H7. Anesthesia nurses and OR nurses cannot be modeled as nominative resources.
Human resources such as OIP nurses, PACU nurses and stretcher-bearers participate regularly in
the patient pathway. The rule of the thumb is usually the following: idle staff takes care of patients
in demand of care. Their IDs are not included in the database; thus we make the following
hypothesis:
H8. PACU nurses, OIP nurses, and the stretcher-bearers are modeled as non-nominative resources.
The same can be said for material resources such as PACU beds and the preoperative care bed: the
ones idle are used when needed; they are not assigned to a specific patient. The OR is the only
material resource that is assigned to the patient.
H9. Preoperative beds and PACU beds are modeled as non-nominative resources.
H10. ORs can be modeled as nominative resources.

75
Chapter IV. Modeling and simulation of the schedule execution

2.2.3. Limited or Unlimited Resources

In this section and the remaining of the manuscript, we talk about limited or unlimited resources
through misuse of language. By limited/unlimited resources, we mean limited/unlimited entities in
a resource group. If resources are limited during schedule execution, a patient can try to acquire an
already busy resource and must wait until the resource is idle again. If resources are unlimited, this
waiting state never occurs.
We illustrate the impact of limited resources in Figure 24 with the example of a single
anesthesiologist working in two ORs at the same time – note that this is common practice in
surgical suites. We consider two operating room shifts (grey) that each have one scheduled patient
(dark blue and light blue) which both require an anesthesiologist for their induction. In the first
case, the anesthesiologist resource is unlimited, and the two induction steps are performed
simultaneously. In the second case, the anesthesiologist resource is limited: the anesthesiologist
starts by the Patient #1 induction, then moves to the Patient #2 induction (green). Patient #2 must
wait for the anesthesiologist to perform their induction because they are already occupied with
performing the induction of Patient #1 (orange).

Figure 24 - Example of the Impact of the Anesthesiologist Being a Limited Resource

Now we briefly address the difference between unlimited resources and not considered resources.
If a resource is modeled as unlimited, it will never provoke waiting times. If the resource is not
even considered (i.e. not present in the model), the model will not even require the resource. In
other words, in either case, the steps’ timestamps and durations do not change. The difference
resides in the fact that with unlimited resources, we can still compute KPIs for the resource (such
as the utilization rate, the idle rate, etc.) and visualize the resource interactions with the rest in the
model. Modeling unlimited resources can be an intermediate step between not modeling resources
and modeling limited resources.

76
 Chapter IV. Modeling and simulation of the schedule execution

Concerning the human resources in our model: (1) we do not have access to the number of OIP
nurses and the number of PACU nurses, (2) each OR is assigned an OR nurse, (3) we have the list
of surgeon and anesthesiologist IDs. Since we can decide to hypothesize the number of OIP nurses
and PACU nurses, all human resources can be considered either limited or unlimited based on
what our goal is.
Concerning the material resources: we do not have access to the number of beds dedicated to the
preoperative care, nor the number of PACU beds. However, we can make a hypothesis on this
number. Besides, we know exactly how many ORs exist.
Note that nominative resources are always limited.
We summarize the way we model resources in Figure 25. Blue resources are always modeled as
non-nominative resources. Green resources can be modeled as nominative resources. Resources
with an infinite sign are always unlimited if they are considered in the model. Resources without an
infinite sign can be modeled as limited resources.

Figure 25 - Which resources can be modeled as infinite resources?

2.2.4. Resource Constraints for Our Specific Study Case

We apply the solution proposed in 1.2 to model our resource constraints based on the available
data and mix it with the notion of nominative/non-nominative resources, and limited/unlimited
resources. Strict resource constraints are the (possibly incomplete) list of constraints described
in the OR database and flexible resource constraints are the ones imposed on the operating
room only.

77
Chapter IV. Modeling and simulation of the schedule execution

We remind the reader that in our study case:

• Each patient is linked to a nominative surgeon, a nominative anesthesiologist (if needed),
a nominative OR, and a preoperative care location type (the bed used however is non-
nominative).
• Since each OR is assigned an OR nurse, and each patient is assigned an OR: each patient
is consequently assigned an OR nurse.
• Patients can require a PACU nurse or an OIP nurse depending on their patient pathway.
Table 13 and

Table 14 describe the flexible and the strict constraints on resources for our study case.
For the performed schedule execution, we use flexible constraints to respect the database
timestamps: human resources are not considered, and material resources are non-nominative and
unlimited, except for the OR that are nominative and limited. We decided to keep the OR
nominative and limited as this is the most trustable resource-related information and that it is at
the core of the schedule: it would not make sense to attempt to replay a performed schedule by
using different ORs than the ones that were used. Not that we could have modeled human
resources as non-nominative and unlimited but that we decided to not consider them for
simplification. This could be a work perspective.
For the provisional schedule execution, we use strict constraints as we try to envision the schedule
execution in its globality and in the perfect case. Human resources are nominative, when possible
(surgeon, anesthesiologist and OR nurse), and non-nominative + unlimited in the other case.
Regarding material resources: we kept nominative ORs and made an estimation of how many beds
are available in the preoperative area in the PACU; beds are thus limited resources.
Table 13 – Description of the flexible and strict constraints on material resources

Resource Flexible Constraint Strict Constraint

Operating Rooms Nominative + Limited Nominative + Limited
Preoperative beds Non-nominative + Unlimited Non-nominative + Limited
PACU beds Non-nominative + Unlimited Non-nominative + Limited

Table 14 – Description of the flexible and strict constraints on human resources

Resource Flexible Constraint Strict Constraint

Surgeons / Nominative + Limited
Anesthesiologists / Nominative + Limited
OR nurse / Nominative + Limited
PACU nurses / Non-nominative + Unlimited
OIP nurse / Non-nominative + Unlimited

78
 Chapter IV. Modeling and simulation of the schedule execution

In the next section, we will describe the surgical suite’s process and show for which steps resources
are necessary.

2.3. Aggregated surgical suite process with flexible constraints on resources

To model the performed schedule execution, we use an aggregated surgical suite process
with flexible resource constraints. The process we describe in this section is the one we modeled
in our simulation tool.
Based on the proposition in 1.1, we use the timestamps available in the database (Suite entry, OR
entry, incision, suture, OR exit, PACU entry, PACU exit). Since they are all part of the patient
pathway, the surgical suite processes are reduced to the patient pathway.
We add a step “Move to Suite” at the start of the process, and a step “Exit Suite” at the end of the
pathway. These steps are used to model the patient being transported in and out of the surgical
suite. “Move to Suite” provides time for the patient to be created at the suite entry, and to walk
towards the preoperative area. “Exit suite” provides time for the patient to move from the PACU
to the suite exit and disappear. This has two advantages: (1) the patient does not appear at the
preoperative location, but at the suite entry, (2) the patient does not disappear at the PACU but at
the suite exit. This requires us to make the following hypotheses:
H11. The timestamp labeled “Suite entry” in the database marks the start of the preoperative care. We
rename it “Preoperative Care Start”.
Consequently, the aggregated pathway follows the 7 initial timestamps available in the database
plus the suite entry and the suite exit. Is it made of 10 steps: (1) move to suite, (2) preoperative
care, (3) setup or setup + induction, (4) procedure, (5) reversal, (6) move to PACU, (7) PACU
monitoring, and (8) exit suite. Steps 1-2 are preoperative, steps 3-5 are perioperative, and steps 6-
8 are post-operative. Steps 1 and 10 are added to facilitate the modeling (see technical notes below).
The aggregated patient pathway and its related resources is presented in Figure 26, and each step
is described in Table 15. In Figure 26, the timestamps with an asterisk are modified/added
timestamps, the steps with two asterisks are added steps.

79
Chapter IV. Modeling and simulation of the schedule execution

Figure 26 - Description of the aggregated surgical suite process with flexible resource constraints:
timestamps, steps and required resources of the patient pathway.

Note that in Figure 26 we represented all the different possible patient pathway types. First, during
the preoperative care, the patient can either (1) receive LRA induction in a dedicated LRA bed, (2)
receive sedation induction in a dedicated sedation bed, (3) not receive induction and wait in a
dedicated waiting area bed. Second, in the operating room the patient can either (1) receive an
induction, or (2) not receive an induction. Since all combinations are possible, this amounts to 6
different patient pathways. During the preoperative care, the patient can either lie down on a LRA
dedicated bed, sedation dedicated bed or waiting area bed. On a side note, the PACU bed is
required during the “Move to PACU” step because the patient cannot leave the OR before being
sure there is a PACU bed available for them.
In Table 15, we describe the different steps of the aggregated patient pathway: the first column
numbers the steps, the second names the steps, the third describes what happens in the model
during the step (and thus what we can see in the 3D view), and the fourth details the resource
acquisition and release mechanics. In the step description we make a difference between acquiring
the “first available” (non-nominative) resource, or the “assigned” (nominative) resource.
Patients trying to acquire already used resources must wait. For example, a patient who has finished
the reversal step will stay in the OR if they did not acquire a PACU bed. To navigate the process,
each patient is assigned an OR ID, an anesthesia type, a duration for each step (see Chapter V),
and an arrival time.
Now that we have described how we the model performed schedule execution using an aggregated
surgical suite process with flexible constraints on resources, we discuss how we model the
provisional schedule execution using a detailed surgical suite process with strict constraints on
resources.

80
 Chapter IV. Modeling and simulation of the schedule execution

Table 15 – Description of the Steps of the Aggregated patient pathway

# Step name Step Description Resources

The patient is created at the suite entry.

To acquire: preoperative
1 Move to Suite The patient walks to the first available bed
preoperative bed and lays on it.

Preoperative The patient stays in their preoperative Already acquired:

2
care bed. preoperative bed

Setup or setup The patient walks to their assigned OR and To acquire: OR

3
+ induction lays on it. To release: preoperative bed

4 Procedure The patient stays in the OR. Already acquired: OR

5 Reversal The patient stays in the OR. Already acquired: OR

The patient walks from the OR to the first To acquire: PACU bed
6 Move to PACU
available PACU bed and lays on it. To release: OR

PACU Already acquired: PACU

7 The patient stays in their PACU bed.
monitoring bed

The patient walks to the suite exit and

8 Exit Suite To release: PACU bed
disappears.

2.4. Detailed surgical suite process with strict constraints on resources

We have explained how we modeled an aggregated SSP by using only the timestamps available in
the OR software. Now, we explain how we map the SPP based on the OR software data, on-site
observations, and interviews with the staff.
To obtain the aggregated pathway, we (1) started with the aggregated process, (2) detailed the
preoperative care phase, (3) implemented the strict resource constraints, and (4) added the OR
cleanup phase. We obtain 6 different patient pathways: the preoperative phase change depending
on the patient anesthesia type, and the perioperative care and the post-operative care phases only
differ in terms of the resource constraints applied.
First, we describe the perioperative and post-operative phases that are common to all patient
pathways (Figure 27), then we represent the preoperative patient pathway in case of no induction
before entering the OR (Figure 28), LRA induction (Figure 29), and ophthalmologic sedation
preparation (Figure 30).

81
Chapter IV. Modeling and simulation of the schedule execution

In the perioperative and post-operative phase of the detailed SSP with strict resource constraints,
the patient requires both materiel (OR and PACU bed) and human (surgeon, anesthesiologist and
OR nurse) resources. During the perioperative phase, the surgical team can either perform a setup
(case a) or perform a setup and an induction (case b). The anesthesiologist is required for case b
only. To model this patient pathway, we made the hypothesis that human resources are needed for
the shortest time possible, meaning that the surgeon must only be here for the surgical procedure
and the anesthesiologist for the induction:
H12. Human resources are required in the model when their absence would stop the process in real-life.
In the preoperative phase without induction, the checklist is long because this is the first time that
the patient meets with a surgical suite staff member:
H13. Modeling: The checklist performed by the OR nurse is “long” if the OR nurse is the first staff
member that the patient meets; otherwise it is considered “short”.
H14. In the preoperative phase for LRA induction and the preoperative phase for OIP, the checklist is
short because the patient has met the suite staff before.
Concerning the supporting process that are not included in the patient pathway, we add the
simulation of OR cleanup during provisional schedule execution – and not in the performed
schedule execution.

Figure 27 - Description of the detailed surgical suite process with strict resource constraints: timestamps,
steps and required resources of the patient pathway (perioperative and post-operative phases)

82
 Chapter IV. Modeling and simulation of the schedule execution

Figure 28 - Description of the detailed surgical suite process with strict resource constraints: timestamps,
steps and required resources of the patient pathway (preoperative phase without induction).

Figure 29 - Description of the detailed surgical suite process with strict resource constraints: timestamps,
steps and required resources of the patient pathway (preoperative phase for LRA induction).

83
Chapter IV. Modeling and simulation of the schedule execution

Figure 30 - Description of the detailed surgical suite process with strict resource constraints: timestamps,
steps and required resources of the patient pathway (preoperative phase with ophthalmology sedation
induction)

84
 Chapter IV. Modeling and simulation of the schedule execution

3. Additional modeling hypotheses applied to our simulation tool

In this section we discuss modeling hypotheses that are not directly linked to the process mapping
of the schedule execution.

3.1. Anesthesia type label and preoperative care option

During the provisional schedule execution, we propose (1) three preoperative care options: no
induction, LRA, and OIP, and (2) two perioperative care options during setup: with induction, and
without induction. These options are based on the anesthesia type that the patient is receiving. We
briefly explain how we treated the patient anesthesia type to define which preoperative and
perioperative care pathway they would be assigned to.
We find 24 different anesthesia labels in the database. We note that:
• Patient’s anesthesia type label can either be “none” or one of 23 anesthesia types.
• Some labels include notes such as “wished by the patient”, “to be decided on the day before
the surgery”, “to be decided with the patient on the day of the surgery”.
• Either one or two anesthesia types can be recorded per labels (i.e. “General anesthesia +
Truncular block under ultrasound”).
Based on our on-site observations, patients with different anesthesia types can follow the same
patient pathway. We make the new hypothesis related to database treatment:
H15. Data treatment: When the label is an anesthesia type plus a note similar to “wished by the
patient” or “to be decided”, we use the anesthesia type.
H16. Data treatment: We use general anesthesia when: (1) there is only the note “to be decided”,
(2) there is a hesitation between general anesthesia and another type of anesthesia.
H17. Data treatment: We group the anesthesia under the following global anesthesia types: GA,
LRA, OPH, SED and AL (see Table 16)
In Table 17, we display how the anesthesia type conditions the patient pathway. For instance, GA,
SED and SA imply that the patient will receive preoperative care in a waiting area bed, and that an
anesthesiologist will be required during the setup step in the OR in order to perform an induction.

85
Chapter IV. Modeling and simulation of the schedule execution

Table 16 – Anesthesia Type Grouping

Anesthesia type Abbreviation Labels included

General anesthesia GA General Anesthesia

Local-regional LRA Axillary Block, Popliteal Sciatic Block, Truncular Block,

anesthesia Nerve Block Under Ultrasound, Loco-Regional
Anesthesia, Peribulbar Anesthesia, Sciatic Block, Crural
Block, Catheter

Ophthalmologic OPH Sub-Tenon Anesthesia24

induction

Sedation SED Sedation

Local anesthesia LA Local Anesthesia, WALANT25

Spinal anesthesia SA Spinal Anesthesia, Short Spinal Anesthesia, Epidural

Table 17 – Patient Pathway Options based on Anesthesia Type.

Preoperative Induction in room

Anesthesia type
bed type required?

General anesthesia, Sedation, Spinal

Waiting Area bed Yes
anesthesia.

Local-regional anesthesia LRA bed No

Ophthalmologic induction OIP bed No

Local anesthesia Waiting area bed No

24 [Link]
25 [Link]

86
 Chapter IV. Modeling and simulation of the schedule execution

3.2. Patient movements modeling

In this subsection, we explain how we model patient movements in the surgical suite
during schedule execution. Patients must change location throughout the pathway:
1. From the Suite entry to the preoperative location
2. From the preoperative location to the OR
3. From the OR to the PACU
4. From the PACU to the Suite exit.
Concerning movements #1 and #4, we know that:
• The preoperative step duration is from the Suite Entry to the OR entry timestamps and is
realized in a preoperative bed.
• The PACU monitoring duration is from the PACU Entry to the PACU Exit timestamps
and is realized in a PACU bed.
• We want the patient to start the preoperative step and finish the PACU monitoring at the
expected times in the expected place.
• In our model, patients spawn at the Suite entry location and disappear at the Suite exit
location.
Thus, we add a one-minute step called “Suite Arrival” step after patient creation and another one-
minute step “Suite Exit” after PACU monitoring (so before patient disappearance). During these
steps, the patient walks from one location to another. This allows to respect both the time and the
location of Preoperative Care and PACU Monitoring. In other words, we rename Suite Entry in
Preoperative Care Start and add a new Suite Entry one minute before, and we add Suite Exit one
minute after PACU Exit.
Secondly, since movement #2 is short and has not related timestamps in the OR software, we
include it in the step that follows it. Moving the Patient to the OR is included in the Setup (or Setup
and Induction).
Finally, movement #3 has its own timestamps (OR exit and PACU entry) so we modeled it as an
independent step of the pathway.

3.3. Modeling OR schedules

Note that in our model, the operating rooms are the only resources on a schedule. All other
resources (staff and beds) are available at any time if they are not already acquired by a patient. We
model ORs schedules such that:
• Simulated and real operating room schedules (opening time, closing time and shift
durations) are identical.
• All patients scheduled for the day will either undergo surgery on that specific date or be
canceled/postponed. Since patients can exit the model by either finishing their pathway or
by being canceled (at the entrance of the Suite or before the OR entry), this means that
even if the OR shift is supposed to be over, scheduled patients will still receive surgery.
This is referred to as “overtime” and can also take place in real-life.

87
Chapter IV. Modeling and simulation of the schedule execution

For each performed case, we have access to the operating room ID, the shift’s start time and the
shift’s end time. We obtained a total of 14925 shifts. However, upon further analysis, we note
occurrences of shifts either overlapping each other or succeeding each other without a break. The
cases are the following:
(1) Shifts A and B are identical.
(2) Shift B starts before the end of shift A (or shift A ends after the start of shift B).
(3) Shift A and shift B start at the same time.
(4) Shift A and shift B end at the same time.
Therefore, we delete shifts duplicates and merge overlapping and successive shifts. This allows us
to have a clean database for the OR shifts.

88
 Chapter IV. Modeling and simulation of the schedule execution

4. Chapter synthesis
In this Chapter, we have described how we model the schedule execution for either a provisional
or a performed schedule.
To model the performed schedule execution, we use an aggregated process that solely uses
the information available in the database after schedule execution. Preoperative and post-operative
processes are the same for each patient. The perioperative process varies depending on whether
the patients need an induction in the OR, or not. The resource constraints are flexible: we do
not consider human resources, and we consider a limited number of nominative operating rooms;
there are no other material resources. Consequently, patient waiting times can only happen if two
patients require the same OR at the same time.
To model the provisional schedule execution, we use a detailed process that relies on database
information, staff interviews and on-site observations. Patients can have different preoperative care
(no induction, LRA, OIP), and different perioperative care (induction, no induction). Post
operative care is identical for each patient. We apply strict constraints on human and material
resources’ usage. Patient waiting time can occur for two reasons: (1) two patients require access to
the same OR, the same surgeon, the same anesthesiologist, or the same OR nurse, or (2) if there
are no more preoperative/post-operative beds available.
We synthesize this in Table 18 and Figure 31. We use the DT-DSS to simulate schedule execution,
whether it is provisional or performed. First, we added surgical suite processes, human resources,
and material resources in the virtual environment (simulation tool). Second, we implemented
parameters (schedule type, process type, constraint type) to be able to change the environment
configuration. Third and finally, we inputted the schedule description in the model. All this allows
us to extract the description of the schedule execution as the output.
Table 18 – Description of process type and resources for the performed and the provisional
schedule execution.

Schedule
# Process Type Human resources Material Resources
Type

1 Aggregated: relies on Limited and

Performed None
database information Nominative OR

2 - Limited and - Limited and

Detailed: relies on Nominative Surgeon, Nominative OR
database information, Anesthesiologist and
Provisional - Limited and not-
staff interview and on- OR Nurse nominative
site observations - Unlimited PACU preoperative and post-
nurse and OIP nurse operative beds

89
Chapter IV. Modeling and simulation of the schedule execution

Figure 31 – Representation of our DT-DSS so far.

90
 Chapter V. Computing durations

CHAPTER V. COMPUTING DURATIONS

In Chapter III, we have identified which functionalities our DT-DSS must perform to support our
analysis framework, and we have shown that our study case environmental constraints require to
adapt these functionalities. In Chapter IV we have described how we model the schedule execution
of either provisional or performed schedules. In this Chapter, we describe how we compute
the deterministic and stochastic durations of each activity of the surgical suite processes.
Activity durations can be either deterministic or stochastic. A deterministic duration is a discrete
value known in advance. A stochastic duration is a value gathered from a statistical or empirical
law. Using the pseudo-random number generator algorithm, we are able to generate a sequence of
values to be assigned to the stochastic duration variable. This can be done after providing a certain
“seed” number to the algorithm. We are able to re-run the exact same scenario with the exact same
sequence of pseudo-random duration values upon providing the same seed value. This serves for
the purpose of replicability of experiments. Upon providing different seed numbers however, we
are able to run multiple replications of a scenario, each of which possess a different sequence of
duration values. Our discrete-event simulation model allows us to run scenarios in multiple ways:
deterministic scenarios in which durations remain the same, or stochastic scenarios in which
durations change from one replication to another by automatically varying the previously
mentioned seed number. The performed schedule can only have deterministic durations as we are
only replicating something that already happened. However, the provisional schedule can be
modeled either in a deterministic or stochastic manner.
In Chapter IV, we discuss how we model schedule execution and list the timestamps and the
activities that we model in our DT-DSS. Some of these timestamps are recorded in the database,
others are not.
In Figure 32, we represent the detailed patient pathway timeline (Chapter II). The color code is the
following: green for timestamps and durations provided by the database, blue for timestamps and
durations not provided by the database, and grey for timestamps and durations added for modeling
purposes (section 3-Chapter IV). To compute the deterministic and stochastic durations of
each activity, we need to ensure the coherence of the green data and to estimate the blue
data. This will be later detailed and explained in the reminder of this Chapter.
A general correction and computation process is illustrated in Figure 33. Each step corresponds to
a section of this Chapter. Indeed, since the staff’s priority is providing care to the patient,
timestamps can be omitted or recorded earlier or later than when they occurred. Consequently,
we need to ensure the coherence of timestamps recorded in the database and compute the
missing ones before being able to use them in the DT-DSS.

91
Chapter V. Computing durations

Table 19 – Who records the surgical suite processes’ timestamps?

# Timestamps Recorded in database by…

Nurse assistant(s) responsible for

1 Suite Entry
welcoming the patient in the suite

2 PACU Entry, PACU Exit PACU nurses

Suite Entry, Room Entry, Incision, Suture,

3 OR nurses
Room Exit

LRA induction Start, LRA induction end,

4 Not recorded
Meet with nurses, Suite Exit

Figure 32 - Detailed Patient Pathway Timeline

Compute Compute
Delete Incorrect Compute
Missing Provisional
Performed Performed
Performed Timestamps and
Timestamps (1) Durations (2)
Timestamps (3) Durations (4)

Figure 33 – Steps followed to correct and compute timestamps and durations

92
 Chapter V. Computing durations

1. Correcting the initial timestamps

1.1. Introduction
As mentioned in the introduction, the staff is responsible for manually recording the timestamps
in the database. Since this task is not their priority, it can sometimes be neglected: timestamps can
be recorded at the activity start time, before it, after it, or not at all. This can lead to computing
erroneous durations and thus to not being able to simulate schedule execution in a reliable manner.
It should be noted that since we are studying the performed schedule, we use the aggregated patient
pathway (Figure 34) with flexible resource constraints. The aggregated patient pathway in the
surgical suite is divided into three main phases: preoperative care (before surgery), perioperative
care (during surgery) and post-operative care (after surgery, recovery). These processes can be
divided into steps (preoperative step, setup, procedure, reversal, moving to PACU, PACU
monitoring), that are delimited by timestamps (suite entry, room entry, incision, suture room exit,
PACU entry and PACU exit).

Figure 34 – Timestamps and Steps of an Aggregated Patient Pathway

We cannot determine whether an isolated timestamp is wrongly or rightly recorded

because we have no comparison point. We can observe the surgical suite processes from several
points of views (POV): the patient, the human resources (surgeon, anesthesiologist, OR nurse,
PACU nurse, OIP nurse), or the material resources (preoperative bed, operating room, PACU
bed). Ensuring timestamp coherence would require checking whether the patient pathway
timestamps are in the right order, and whether each resource is used only by one patient at a time.
However, as we mentioned before, we study the aggregated patient pathway with flexible resource
constraints, which is a model that only considers the patient and the ORs (Chapter IV).
Consequently, we ensure the coherence of the timestamps from the patient POV and the
OR POV. We describe both POV in Figure 35. The first describes the patient POV (patient
pathway), and the second timeline describes the OR POV. It is a succession of “operating times”
(patient in room) and “turnovers” (waiting for patient to enter the room).

93
Chapter V. Computing durations

Figure 35 – Illustration of the surgical suite processes from the patient POV (top) and the operating room
POV (bottom)

We illustrate timestamp incoherence using patient POV in Table 20. On the first row, the
timestamps were recorded one after the other: they are all coherent. On the second row, the
incision, suture, and room exit timestamps were also recorded one after the other and are thus
coherent with each other. However, we can see that room entry was allegedly performed after the
incision, which is impossible. Thus the room entry and the incision timestamps are not coherent
with each other. A question remains, which one should we keep?
Table 20 - Example of coherent and incoherent timestamps.

# Room Entry Incision Suture Room Exit

1 9:00 9:15 9:45 10:00

2 9:20 9:15 9:45 10:00

Based on our on-site observations and staff interviews, we noted that the staff may not have time
to record the timestamp of a specific step (n) at the right time. When recording the timestamp of
the next step (n+1), they may realize their omission and record a late timestamp for step n.
Consequently, we make the following hypothesis:
H18. Data treatment: For two supposedly successive timestamps, if the first one is recorded later than
the second one, then we delete the first timestamp.
In our Table 20 example, this means that we consider the incoherent timestamp to be the room
entry from row#2. Thus we delete it and keep the incision timestamp.
In this subsection, we consider timestamps from the patient POV. We correct them by ensuring
that, for each patient pathway, for two consecutive timestamps, the successor should always come
after the predecessor one (e.g., the patient leaves the room after the reversal, not the contrary).
When it is not the case, we delete the supposedly preceding timestamp and keep the successor.

94
 Chapter V. Computing durations

In this section, we study timestamp incoherence within each patient pathway (are all the timestamps
in the right order?) and for each OR (is there always only one patient in the OR?).

1.2. Finding incoherent timestamps using the patient point of view

Let us consider the examples displayed in Table 21. Room entry and incision are recorded for two
patients. In the first case, the OR nurse recorded the values of the timestamps successively (9:15am
is after 9am). In the second case, the OR nurse either forgot to record room entry or had another
more urgent task to do. When the surgeon incised the patient, the OR nurse recorded the incision
and then the room entry. Thus the room entry is recorded after incision. This can occur for any
consecutive timestamps.
Table 21 – Example of coherent and incoherent values two consecutive timestamps in the same patient
pathway

# Room Entry Incision Reversal Observation

Time Time Duration

1 9:00 9:15 15minutes Coherent: keep both

timestamps.

2 9:20 9:15 -5minutes Incoherent: delete

t_room_entry.

1.3. Finding incoherent timestamps using the operating room point of view
For two patients A and B using the same OR successively, we delete Room Exit (A) if Room Entry
(B) is recorded before Room Entry (A). In Figure 36, we represent the timelines of two patients
using the same OR. The records show that patient B enters the OR before patient A leaves it,
which is impossible. Thus, we keep Room Entry (B) and delete Room Exit (A).

Figure 36 – Illustration of incoherence timestamps in the light of operating room usage (OR POV).

95
Chapter V. Computing durations

1.4. Correction
In Figure 37, we show the number of timestamps recorded in the database initially (blue), after the
correction using patient POV (orange), and after the correction using OR POV (grey). Table 22
displays the number of timestamps at each step, and the gap between the initial number and the
number after correction #2. Table 23 does the same but under the form of percentages. We focus
on the results displayed in the column “Percentage after correction #2 (OR POV)” of Table 23:
all timestamps are recorded more than 78% of the time, except suite arrival which is recorded at
17%. This is because suite arrival was only recorded one year compared to 4 years for the other
timestamps. These 17% still amount to 12,238 timestamps, which is sufficiently large for what we
want to do.

80000

70000
Recorded in the Database
Number of Timestamps

60000

50000

40000

30000

20000

10000

0
Suite Arrival Room Entry Incision Suture Room Exit PACU Entry PACU Exit
Timestamp

Before correction Correction #1 (Patient POV) Correction #2 (OR POV)

Figure 37 - Comparison of the Number of Timestamps after Correction from the Patient POV
(orange) and the OR POV (grey).

96
 Chapter V. Computing durations

Table 22 – Number of timestamps before correction, after correction using patient POV, and
after correction using OR POV.

Gap between
Number after Number after
initial and
Timestamps Initial Number correction #1 correction #2
correction #2
(Patient POV) (OR POV)
number
Suite Arrival 12,278 12,238 12,238 40
Room Entry 68,091 67,864 67,864 227
Incision 56,255 56,154 56,154 101
Suture 66,829 65,176 65,176 1,653
Room Exit 67,947 66,138 63,143 4,804
PACU Entry 67,842 63,776 63,776 4,066
PACU Exit 67,723 67,723 67,723 0

Table 23 - Percentage of timestamps before correction, after correction using patient POV, and
after correction using OR POV.

Percentage Gap between

Percentage
Initial after correction initial and
Timestamps after correction
Percentage #1 (Patient correction #2
#2 (OR POV)
POV) Percentage
Suite Arrival 17.16 17.10 17.10 0.06
Room Entry 95.15 94.83 94.83 0.32
Incision 78.61 78.47 78.47 0.14
Suture 93.38 91.07 91.07 2.31
Room Exit 94.94 92.42 88.23 6.71
PACU Entry 94.80 89.12 89.12 5.68
PACU Exit 94.63 94.63 94.63 0

97
Chapter V. Computing durations

2. Computing durations

2.1. Introduction
Now that we have deleted the incoherent timestamps from our database, we can compute the
performed discrete durations between each remaining couple of timestamps. These durations will
allow to: (1) simulate the performed schedule execution in our DT-DSS, (2) compute discrete
provisional durations, and (3) compute stochastic provisional durations.

2.2. Computing initially available durations

For each two consecutive timestamps, we compute the duration between them. Table 24 displays
the number and the percentage of durations we can compute based on the current available
timestamps. As we can see, some durations are missing. For instance, we only have access to 17%
of the preoperative phase duration, and the highest percentage of durations recorded is 88%
(PACU monitoring). In the remaining of this section, we explain how we computed the missing
durations.
Table 24 – Number and percentage of computed durations per duration type.

Number of duration Percentage of durations

# Duration Type
computed in the database computed in the database
1 Preoperative care 12,195 17.04
2 Setup (+ induction) 55,827 78.01
3 Procedure 53,943 75.38
4 Reversal 60,469 84.50
5 Moving to PACU 59,279 82.83
6 PACU Monitoring 63,657 88.95

2.3. Impactful criterion of the operating duration

Based on our staff interviews, we know that the surgeon, the anesthesiologist, the anesthesia type,
and the surgery type can have a strong influence on the patient pathway durations. We cannot base
our study on the surgery type because it is not standardized in the database we are using. Thus, we
use the CCAM codes instead - the CCAM is the Common Classification of Medical Procedures
(Classification Commune des Actes Médicaux). It is used by the French social security system (Sécurité
Sociale) to pay doctors and reimburse insured people26. Consequently, we use the surgeon ID,
the anesthesiologist ID, the CCAM codes and the anesthesia type to group patients.
Upon trying to create patient groups based on all four criteria, we notice that very few cases have
all four-information recorded. Thus, we group patients based on what data is available and end up
with6 possible combinations. Table 25. describes the grouping criteria we use. The last column

26 [Link]

98
 Chapter V. Computing durations

specifies in which patient group is attached a cataract surgery described by the CCAM code
“BFPP001”, performed by surgeon A and anesthesiologist B, and receiving and OIP.
Table 25 – Proposition of grouping criteria.

Surgeon Anesthesiologist CCAM Anesthesia Example of grouping

#
ID ID codes type criterion

1 No No No No /

2 Yes Yes No Yes [A, B, OIP]

3 Yes No No Yes [A, OIP]

4 Yes No No No [A]

5 No No Yes No [BFPP001, OIP]

6 Yes Yes Yes Yes [A, B, BFPP001, OIP]

2.4. Method for each correction

For each grouping criterion, for each duration type, we follow these steps:
1. We extract the cases where (i) the grouping criterion is respected, and (ii) the duration is
not null.
2. We randomly split the obtained sample in two subsets (30% - 70%).
3. We compute the median and the mean of the durations in the 70% subset.
4. We compare the true value of the duration in the 30% subset to the estimated value (median
and mean) in the 70% subset.
5. Let k be a gathering criterion, j an activity, i a patient respecting the criterion and having
the activity duration computed, and n the number of patient i. We compute the Weighted
Absolute Percentage Error WAPE(j,k) according to the formula:
∑𝑛
𝑡=1 | 𝐴𝑡 −𝐹𝑡 |
𝑊𝐴𝑃𝐸𝑑𝑢𝑟𝑎𝑡𝑖𝑜𝑛 𝑡𝑦𝑝𝑒,𝑐𝑟𝑖𝑡𝑒𝑟𝑖𝑜𝑛 = ∑𝑛
,
𝑡=1 | 𝐴𝑡 |

where 𝐴𝑡 is the actual (true) value of the median or mean durations in the 30% subset
and 𝐹𝑡 is the forecasted median or mean duration computed from the 70% subset. The
reasoning behind our choice of error metric is to provide ourselves with an easily
understandable percentage value without having to worry about the effect of short near-
zero values on the overall percentage metric.
We display our results in Table 26. The durations are abbreviated as D1 (preoperative care), D2
(setup + induction), D3 (procedure), D4 (reversal), D5 (move to PACU), D6 (PACU monitoring).
The grouping criterion is abbreviated as S (surgeon ID), A (anesthesiologist ID) and T (anesthesia
type) and C (CCAM codes). The methods are abbreviated as MED (median) and AVG (average or
mean).

99
Chapter V. Computing durations

Table 26 – Values of the WAPE for each duration type (column) and each estimation method
(row). The worse the WAPE is the more the colors tend to be red; the better the WAPE is the
more the colors tend to be green.

Grouping
# D1 D2 D3 D4 D5 D6 Method
Criterion
0 3.43 5.42 12.58 17.16 8.84 7.04 S+A+T MED
1 5.78 5.78 14.98 16.85 8.42 6.41 S+T MED
2 10.79 8.18 21.12 21.2 11.16 9.68 S MED
3 5.78 4.95 5.64 13.09 6.49 5.64 C MED
C+S+A
4 3.13 2.85 3.7 11.08 6.89 4.33 MED
+T
5 0.73 0.23 0.2 0.64 2.47 0.2 S+A+T AVG
6 0.34 0.4 1.48 1.62 0.3 0.15 S+T AVG
7 0.66 0.25 0.81 0.8 1.33 0.08 S AVG
8 0.3 0.39 0.44 0.47 0.13 0.71 C AVG
C+S+A
9 0.68 1.06 0.52 0.1 1.07 0.62 AVG
+T
1
0.24 0.31 2.63 0.83 0 0.47 / AVG
0
1
13.16 16.28 35.01 37.87 10.56 21.16 / MED
1

2.5. Estimate missing durations

For each duration type, we rank the correction methods according to the WAPE (the lower being
the better). For each case, for each duration, we associate the best correction type available. If
correction is based on more than 2 values (3 or more), then we apply the correction. If not, we
move on the next best correction type. We can compute all the durations. However, we are not
sure yet that they are coherent with the available timestamps.

100
 Chapter V. Computing durations

3. Computing missing timestamps

3.1. Introduction
In the previous sections we have cleaned the database of incoherent timestamps, computed
durations between each remaining couple of successive timestamps, and used these newly
computed durations to estimate the value of the remaining missing.
We now have access to a discrete duration for each activity of each patient. However, we do not
know whether these durations guarantee the coherence of timestamps that we had established in
section 1. Consequently, in this section, we compute each missing timestamp based on the
durations and assess whether it keeps the processes coherent.
We remind our reader that:
• When dealing with two supposedly successive timestamps, we consider the successor to be
more reliable than the predecessor. Therefore, if these two timestamps are incoherent with
each other, we delete the earlier timestamp and keep the latter one (H18).
• The notion of coherence is based on the respect of the surgical process from the patient
POV and from the OR POV.
To compute our missing timestamps, we implement the following steps:
1. Correct pathway extremities: suite entry and PACU exit (section 3.2).
2. Correct each patient pathway independently: incision, suture and PACU entry (section 3.3).
3. Correct patient pathways by considering them dependent on each other: room exit and
room entry (section 3.4).
The steps are all repeated until the number of timestamps corrected stops increasing. Indeed, as
we will see, the timestamps are dependent on each other, so correcting one can help correct another
one. We detail each of the steps below.
In this section, we will constantly use the same color code in each figure: we want to compute the
missing timestamps (orange), based on the available previous/following timestamps (green) and
the available durations between them (green).

101
Chapter V. Computing durations

3.2. Correct pathway extremities: suite entry and PACU exit

For each patient, when both the 𝑡 𝑟𝑜𝑜𝑚 𝑒𝑛𝑡𝑟𝑦 and the 𝑑 𝑝𝑟𝑒𝑜𝑝𝑒𝑟𝑎𝑡𝑖𝑣𝑒 𝑐𝑎𝑟𝑒 are available, we compute
𝑡 𝑠𝑢𝑖𝑡𝑒 𝑒𝑛𝑡𝑟𝑦 = 𝑡 𝑟𝑜𝑜𝑚 𝑒𝑛𝑡𝑟𝑦 − 𝑑 𝑝𝑟𝑒𝑜𝑝𝑒𝑟𝑎𝑡𝑖𝑣𝑒 𝑐𝑎𝑟𝑒 . We illustrate this in Figure 38.

Figure 38 – We compute “Suite Entry Time” based on “Room Entry Time” and “Preoperative Care
Duration”.

For each patient, when both the 𝑡 𝑃𝐴𝐶𝑈 𝑒𝑛𝑡𝑟𝑦 and the 𝑑 𝑃𝐴𝐶𝑈 𝑚𝑜𝑛𝑖𝑡𝑜𝑟𝑖𝑛𝑔 are available, we compute
𝑡 𝑃𝐴𝐶𝑈 𝑒𝑥𝑖𝑡 = 𝑡 𝑃𝐴𝐶𝑈 𝑒𝑛𝑡𝑟𝑦 + 𝑑 𝑃𝐴𝐶𝑈 𝑚𝑜𝑛𝑖𝑡𝑜𝑟𝑖𝑛𝑔 . We illustrate this in Figure 39.

Figure 39 - We compute “PACU Exit Time” based on “PACU Entry Time” and “PACU Monitoring
Duration”.

3.3. Correct each patient pathway independently: incision, suture, PACU entry
For each missing timestamp that is not at the patient pathway extremity (𝑡 𝑠𝑢𝑖𝑡𝑒 𝑒𝑛𝑡𝑟𝑦 or 𝑡 𝑃𝐴𝐶𝑈 𝑒𝑥𝑖𝑡 ),
we identify whether we have the required elements to correct it. To correct 𝑡𝑖 , we need to respect
the following conditions (illustrated in Figure 40 and Figure 41):
𝑡𝑖−1 ≠ 0
𝑑(𝑡𝑖 , 𝑡𝑖+1 ) ≠ 0
𝑑(𝑡𝑖−1 , 𝑡𝑖 ) ≠ 0
𝑡𝑖+1 – 𝑑(𝑡𝑖 , 𝑡𝑖+1 ) > 𝑡𝑖−1 (𝑐𝑜𝑛𝑑𝑖𝑡𝑖𝑜𝑛 1)
𝑡𝑖−1 + 𝑑(𝑡𝑖−1 , 𝑡𝑖 ) < 𝑡𝑖+1 (𝑐𝑜𝑛𝑑𝑖𝑡𝑖𝑜𝑛 2)

102
 Chapter V. Computing durations

If all conditions are respected, we compute 𝑡𝑖 = 𝑡𝑖+1 − 𝑑(𝑡𝑖 , 𝑡𝑖+1 ). Note that we correct the
timestamps based on the latest timestamp. This is coherent with the fact that for two consecutive
timestamps, we consider the second one to be more reliable.

Figure 40 – Illustration of cases where 𝑡𝑖+1 – 𝑑(𝑡𝑖 , 𝑡𝑖+1 ) > 𝑡𝑖−1 is respected (top timeline) or not respected
(bottom timeline).

Figure 41 – Illustration of cases where 𝑡𝑖−1 + 𝑑(𝑡𝑖−1 , 𝑡𝑖 ) < 𝑡𝑖+1 is respected (top timeline) or not
respected (bottom timeline).

103
Chapter V. Computing durations

In the following figures, we illustrate how to respectively compute:

• 𝑡 𝑖𝑛𝑐𝑖𝑠𝑖𝑜𝑛 based on 𝑡 𝑠𝑢𝑡𝑢𝑟𝑒 and 𝑑 𝑝𝑟𝑜𝑐𝑒𝑑𝑢𝑟𝑒 (Figure 42).

• 𝑡 𝑠𝑢𝑡𝑢𝑟𝑒 based on 𝑡 𝑟𝑜𝑜𝑚 𝑒𝑥𝑖𝑡 and 𝑑 𝑠𝑢𝑡𝑢𝑟𝑒 (Figure 43).

• 𝑡 𝑃𝐴𝐶𝑈 𝑒𝑛𝑡𝑟𝑦 based on 𝑡 𝑃𝐴𝐶𝑈 𝑒𝑥𝑖𝑡 and 𝑑 𝑃𝐴𝐶𝑈 𝑚𝑜𝑛𝑖𝑡𝑜𝑟𝑖𝑛𝑔 (Figure 44).

Figure 42 – We compute “Incision Time” based on “Suture Time” and “Procedure Duration”.

Figure 43 - We compute “Suture Time” based on “Room Exit Time” and “Reversal Duration”.

Figure 44 - We compute “PACU Entry Time” based on “PACU Exit” and “PACU Monitoring
Duration”.

104
 Chapter V. Computing durations

3.4. Correct patient pathway by considering them dependent on each other: room
entry, room exit

3.4.1. Correction of room exit time

We try to correct 𝑡𝑖𝑟𝑜𝑜𝑚 𝑒𝑥𝑖𝑡 using the following methods:

(1) Relying on its successor (Figure 45).
(2) Relying on its predecessor (Figure 46).
(3) Making the hypothesis that moving to the PACU takes 30 seconds (Figure 45) based on
on-site observations.
Each of the methods requires to respect a set of constraints. When one set of constraints is not
respected, we move on to the next method. If no set is respected, we cannot correct the timestamp.
Method 1. If the following constraints are respected, we compute 𝑡 𝑟𝑜𝑜𝑚 𝑒𝑥𝑖𝑡 = 𝑡 𝑃𝐴𝐶𝑈 𝑒𝑛𝑡𝑟𝑦 −
𝑑 𝑚𝑜𝑣𝑒 𝑡𝑜 𝑃𝐴𝐶𝑈 :
𝑟𝑜𝑜𝑚 𝑒𝑛𝑡𝑟𝑦
𝑡𝑖+1 ≠ 0 | 𝑐𝑎𝑠𝑒𝑖 𝑖𝑠 ℎ𝑒 𝑙𝑎𝑠𝑡 𝑐𝑎𝑠𝑒

𝑡𝑖𝑠𝑢𝑡𝑢𝑟𝑒 ≠ 0
𝑃𝐴𝐶𝑈 𝑒𝑛𝑡𝑟𝑦
𝑡𝑖 ≠0

𝑑𝑖𝑚𝑜𝑣𝑒 𝑡𝑜 𝑃𝐴𝐶𝑈 ≠ 0
𝑃𝐴𝐶𝑈 𝑒𝑛𝑡𝑟𝑦
𝑡𝑖 − 𝑑𝑖𝑚𝑜𝑣𝑒 𝑡𝑜 𝑃𝐴𝐶𝑈 < 𝑡 𝑠𝑢𝑡𝑢𝑟𝑒
𝑃𝐴𝐶𝑈 𝑒𝑛𝑡𝑟𝑦 𝑟𝑜𝑜𝑚 𝑒𝑛𝑡𝑟𝑦
𝑡𝑖 − 𝑑𝑖𝑚𝑜𝑣𝑒 𝑡𝑜 𝑃𝐴𝐶𝑈 ≤ 𝑡𝑖+1 | 𝑐𝑎𝑠𝑒𝑖 𝑖𝑠 𝑡ℎ𝑒 𝑙𝑎𝑠𝑡 𝑐𝑎𝑠

Figure 45 – We compute “Room Exit Time” based on “Suture Time”, “PACU Entry Time” and “Move
to PACU Duration” of the same patient, as well as “Room Entry Time” of the next patient.

105
Chapter V. Computing durations

Method 2. If the following constraints are respected, we compute 𝑡 𝑟𝑜𝑜𝑚 𝑒𝑥𝑖𝑡 = 𝑡 𝑠𝑢𝑡𝑢𝑟𝑒 +
𝑑 𝑟𝑒𝑣𝑒𝑟𝑠𝑎𝑙 :
𝑟𝑜𝑜𝑚 𝑒𝑛𝑡𝑟𝑦
𝑡𝑖+1 ≠ 0 | 𝑐𝑎𝑠𝑒𝑖 𝑖𝑠 𝑡ℎ𝑒 𝑙𝑎𝑠𝑡 𝑐𝑎𝑠𝑒

𝑡𝑖𝑠𝑢𝑡𝑢𝑟𝑒 ≠ 0
𝑃𝐴𝐶𝑈 𝑒𝑛𝑡𝑟𝑦
𝑡𝑖 ≠0

𝑑𝑖𝑟𝑒𝑣𝑒𝑟𝑠𝑎𝑙 ≠ 0
𝑟𝑜𝑜𝑚 𝑒𝑛𝑡𝑟𝑦
𝑡𝑖𝑠𝑢𝑡𝑢𝑟𝑒 + 𝑑𝑖𝑟𝑒𝑣𝑒𝑟𝑠𝑎𝑙 ≤ 𝑡𝑖+1 | 𝑐𝑎𝑠𝑒𝑖 𝑖𝑠 𝑡ℎ𝑒 𝑙𝑎𝑠𝑡 𝑐𝑎𝑠𝑒

𝑃𝐴𝐶𝑈 𝑒𝑛𝑡𝑟𝑦
𝑡𝑖𝑠𝑢𝑡𝑢𝑟𝑒 + 𝑑𝑖𝑟𝑒𝑣𝑒𝑟𝑠𝑎𝑙 ≤ 𝑡𝑖

Figure 46 - We compute “Room Exit Time” based on “Suture Time”, “PACU Entry Time” and “Reversal
Duration” of the same patient, as well as “Room Entry Time” of the next patient.

Method 3. We make the hypothesis that 𝑡𝑖𝑚𝑜𝑣𝑒 𝑡𝑜 𝑃𝐴𝐶𝑈 = 30 seconds. If the following constraints
𝑃𝐴𝐶𝑈 𝑒𝑛𝑡𝑦
are respected, we compute 𝑡𝑖𝑟𝑜𝑜𝑚 𝑒𝑥𝑖𝑡 = 𝑡𝑖 − 𝑑𝑖𝑚𝑜𝑣𝑒 𝑡𝑜 𝑃𝐴𝐶𝑈 :
𝑟𝑜𝑜𝑚 𝑒𝑛𝑡𝑟𝑦
𝑡𝑖+1 ≠ 0 | 𝑐𝑎𝑠𝑒𝑖 𝑖𝑠 𝑡ℎ𝑒 𝑙𝑎𝑠𝑡 𝑐𝑎𝑠𝑒

𝑡𝑖𝑠𝑢𝑡𝑢𝑟𝑒 ≠ 0
𝑃𝐴𝐶𝑈 𝑒𝑛𝑡𝑟𝑦
𝑡𝑖 ≠0

𝑡 𝑃𝐴𝐶𝑈 𝑒𝑛𝑡𝑟𝑦 − 𝑑 𝑚𝑜𝑣𝑒 𝑡𝑜 𝑃𝐴𝐶𝑈 > 𝑡 𝑠𝑢𝑡𝑢𝑟𝑒

𝑟𝑜𝑜𝑚 𝑒𝑛𝑡𝑟𝑦
𝑡 𝑃𝐴𝐶𝑈 𝑒𝑛𝑡𝑟𝑦 − 𝑑 𝑚𝑜𝑣𝑒 𝑡𝑜 𝑃𝐴𝐶𝑈 ≤ 𝑡𝑖+1 | 𝑐𝑎𝑠𝑒𝑖 𝑖𝑠 𝑡ℎ𝑒 𝑙𝑎𝑠𝑡 𝑐𝑎𝑠𝑒

106
 Chapter V. Computing durations

3.4.2. Correction of room entry time.

𝑟𝑜𝑜𝑚 𝑒𝑛𝑡𝑟𝑦
As for 𝑡𝑖𝑟𝑜𝑜𝑚 𝑒𝑥𝑖𝑡 , We try to correct 𝑡𝑖 using the several methods:
(1) Relying on its successor (Figure 47).
(2) Relying on its predecessor (Figure 48).
Each of the methods requires to respect a set of constraints. When one set of constraints is not
respected, we move on to the next method. If no set is respected, we cannot correct the timestamp.
𝑟𝑜𝑜𝑚 𝑒𝑛𝑡𝑟𝑦
Method 1. If the following constraints are respected, we compute 𝑡𝑖 = 𝑡𝑖𝑖𝑛𝑐𝑖𝑠𝑖𝑜𝑛 −
𝑑𝑖𝑠𝑒𝑡𝑢𝑝 :
𝑠𝑢𝑖𝑡𝑒 𝑒𝑛𝑡𝑟𝑦
𝑡𝑖 ≠0

𝑡𝑖𝑖𝑛𝑐𝑖𝑠𝑖𝑜𝑛 ≠ 0

𝑑𝑖𝑠𝑒𝑡𝑢𝑝 ≠ 0
𝑟𝑜𝑜𝑚 𝑒𝑥𝑖𝑡
𝑡𝑖−1 ≠ 0 | 𝑐𝑎𝑠𝑒𝑖 𝑖𝑠 𝑡ℎ𝑒 𝑓𝑖𝑟𝑠𝑡 𝑐𝑎𝑠𝑒
𝑠𝑢𝑖𝑡𝑒_𝑒𝑛𝑡𝑟𝑦
𝑡 𝑖𝑛𝑐𝑖𝑠𝑖𝑜𝑛 − 𝑑𝑖𝑠𝑒𝑡𝑢𝑝 > 𝑡𝑖

𝑡𝑖𝑖𝑛𝑐𝑖𝑠𝑖𝑜𝑛 − 𝑑𝑖𝑠𝑒𝑡𝑢𝑝 ≥ 𝑡𝑖−1

𝑟𝑜𝑜𝑚 𝑒𝑥𝑖𝑡
| 𝑐𝑎𝑠𝑒𝑖 𝑖𝑠 𝑡ℎ𝑒 𝑓𝑖𝑟𝑠𝑡 𝑐𝑎𝑠𝑒

Figure 47 - We compute “Room Entry Time” based on “Suite Entry Time”, “Incision Time” and “Setup
Duration” of the same patient, as well as “Room Entry Time” of the previous patient.

107
Chapter V. Computing durations

𝑟𝑜𝑜𝑚 𝑒𝑛𝑡𝑟𝑦 𝑠𝑢𝑖𝑡𝑒 𝑒𝑛𝑡𝑟𝑦

Method 2. If the following constraints are respected, we compute 𝑡𝑖 = 𝑡𝑖 +
𝑑𝑖𝑝𝑟𝑒𝑜𝑝𝑒𝑟𝑎𝑡𝑖𝑣𝑒 𝑐𝑎𝑟𝑒 :
𝑠𝑢𝑖𝑡𝑒 𝑒𝑛𝑡𝑟𝑦
𝑡𝑖 ≠0

𝑡𝑖𝑖𝑛𝑐𝑖𝑠𝑖𝑜𝑛 ≠ 0

𝑑𝑖𝑝𝑟𝑒𝑜𝑝𝑒𝑟𝑎𝑡𝑖𝑣𝑒 𝑐𝑎𝑟𝑒 ≠ 0
𝑟𝑜𝑜𝑚 𝑒𝑛𝑡𝑟𝑦
𝑡𝑖−1 ≠ 0 | 𝑐𝑎𝑠𝑒𝑖 𝑖𝑠 𝑡ℎ𝑒 𝑓𝑖𝑟𝑠𝑡 𝑐𝑎𝑠𝑒
𝑠𝑢𝑖𝑡𝑒 𝑒𝑛𝑡𝑟𝑦
𝑡𝑖 + 𝑑𝑖𝑝𝑟𝑒𝑜𝑝𝑒𝑟𝑎𝑡𝑖𝑣𝑒 𝑐𝑎𝑟𝑒 ≤ 𝑡𝑖−1
𝑟𝑜𝑜𝑚 𝑒𝑥𝑖𝑡

𝑠𝑢𝑖𝑡𝑒 𝑒𝑛𝑡𝑟𝑦
𝑡𝑖 + 𝑑𝑖𝑝𝑟𝑒𝑜𝑝𝑒𝑟𝑎𝑡𝑖𝑣𝑒 𝑐𝑎𝑟𝑒 < 𝑡𝑖𝑖𝑛𝑐𝑖𝑠𝑖𝑜𝑛

Figure 48 - We compute “Room Entry Time” based on “Suite Entry Time”, “Incision Time” and
“Preoperative care Duration” of the same patient, as well as “Room Entry Time” of the previous patient.

3.5. Results after trying to compute the missing timestamps

We display the number and percentage of total timestamps obtained after trying to compute
missing timestamps in Figure 49, Table 27 and Table 28. We are not able to reach 100% because
the correction requires to have access to already recorded durations and other timestamps which
is not always the case. However, we still increase the incision percentage record by 15% and the
suite arrival one by 78%. This is non-negligible as incomplete patient pathways cannot be simulated
in our surgical suite digital twin.

108
 Chapter V. Computing durations

80000

Recorded in the Database 70000

Number of Timestamp

60000

50000

40000

30000

20000

10000

0
Suite Arrival Room Entry Incision Suture Room Exit PACU Entry PACU Exit
Timestamp

Before correction Correction #1 (Patient POV)

Correction #2 (OR POV) Correction #3 (Missing Timestamps)

Figure 49 – Number of timestamps recorded in the database: before correction (orange), after the
correction from the patient POV (yellow), after correction from the OR POV (grey), and after
computation of the missing performed timestamps using performed durations (blue).

Table 27 - Number of timestamps recorded in the database.

Gap between
Correction Correction Correction
Before initial
Timestamps #1 (Patient #2 (OR #3 (Missing
correction schedule and
POV) POV) Timestamps)
correction #3

Suite Arrival 12,278 12,238 12,238 67,907 55,629

Room Entry 68,091 67,864 67,864 67,864 -227
Incision 56,255 56,154 56,154 66,686 10,431
Suture 66,829 65,176 65,176 66,973 144
Room Exit 67,947 66,138 63,143 64,832 -3,115
PACU Entry 67,842 63,776 63,776 64,473 -3,369
PACU Exit 67,723 67,723 67,723 67,842 119

109
Chapter V. Computing durations

Table 28 - Percentage of timestamps recorded in the database.

Gap between
Correction Correction Correction
Before initial
Timestamps #1 (Patient #2 (OR #3 (Missing
correction schedule and
POV) POV) Timestamps)
correction #3
Suite Arrival 17.16 17.10 17.10 94.89 77.73
Room Entry 95.15 94.83 94.83 94.83 -0.32
Incision 78.61 78.47 78.47 93.18 14.57
Suture 93.38 91.07 91.07 93.58 0.20
Room Exit 94.94 92.42 88.23 90.59 -4.35
PACU Entry 94.80 89.12 89.12 90.09 -4.71
PACU Exit 94.63 94.63 94.63 94.8 0.17

110
 Chapter V. Computing durations

4. Compute timestamps and durations for the provisional schedule

4.1. Introduction
In the previous section, we corrected the performed timestamps, computed the performed
durations of the aggregated surgical suite processes, and used them to compute missing timestamps.
In the database, we have access to the provisional ranks, provisional OR and provisional suite
arrival times. However, the timestamps and the durations of the provisional schedule are not
recorded. In this section, we describe the computations carried out for the provisional schedule:
the stochastic timestamps and durations (to simulate its execution in a stochastic environment),
and the discrete timestamps and durations (to simulate its execution in a determinist environment).
The provisional schedule durations can be estimated in several ways: surgeon estimate, empirical
distribution, statical distribution-fitting, etc. Each of these methods require to group the cases by
their surgery label – and even better, by their surgery label and their surgeon.
However, in the database provided by our collaboration partner, the surgery labels are not
standardized: the name of each procedure is manually written by the staff in the OR software. The
same surgery type can be referred to by several labels. For instance, a cataract can be referenced as:
“cataract”, “Cataract”, “cataract with sedation”, “cataract at 8am”, “cataract that had previously
been canceled”, etc. This non-standardization of the surgery label makes it complicated for
us to group the totality of the database cases by their procedure.
To build the provisional schedule of our study case despite this lack of data, we apply steps
illustrated in Figure 50.

1 - Create a 3 - Determine which

2 - Simplify the
standardized surgery computation method
original surgery label
label to use

4 - Apply the selected 5 - Apply the selected

method to compute method to compute 6 - Compute the
the discrete the stochastic timestamps
provisional durations provisional durations

Figure 50 - Compute Timestamps and Durations for the Provisional schedule

111
Chapter V. Computing durations

4.2. Step description

1 – Create a standardized surgery label. First, we create a standardized name for each surgery
type of the study case. The standardized name is a one-word label in lowercase that describes as
accurately as possible the real surgery label. For simplification reasons, if there are several
procedures, we only focus on the first one. We give examples in Table 29.
Table 29 – Example of Proposed Standardized Label for surgical procedures

# Original Database Label Proposed Standardized Label

1 Cure d’hydrocèle droite selon lord hydrocele, hydrocèle

2 Urétrotomie interne endoscopique uretrotomie, urétrotomie

3 Rtuv + biopsies de vessie rtuv

4 Rtuv + jj droite rtuv

2 – Simplify the original surgery label. Second, we simplify the original surgery label by replacing
all capitals by lowercase, and by deleting any space before or after the text.
3 – Determine which computation method to use. The methods are explained in the next step.
We illustrate this using Figure 51. Our objective is to compute all the determinist and stochastic
durations required to simulate the provisional schedule execution. Thus, for each process step (Ex:
preoperative duration), for a specific surgery type (Ex: cataract) performed by a specific surgeon
(Ex: surgeon A), we count the number of times the step duration is recorded in the database (How
many cataracts performed by surgeon A have a preoperative duration?). If there are more than
100 datapoints, we use method 1, else, we count the number of times the value is recorded for the
specific surgery type only (How many cataracts have a preoperative duration?). If there are more
than 100 datapoints, we use method 2, else we use method 3. The idea is that Method 1 allows us
to compute a better estimation than Method 2, that itself allows us to obtain a better duration than
Method 3. We have chosen the threshold of a 100 datapoints with the expectation of having a
representative sample of cases.

112
 Chapter V. Computing durations

Figure 51 – How do we chose the computation method for the durations of the provisional schedule?

4 – Apply the selected method to compute the discrete provisional durations. Fourth and
finally, we compute the provisional durations by applying the previously selected method. The
formulas are all detailed in

113
 Chapter V. Computing durations

Table 31 (deterministic durations) and Table 32 (stochastic durations). In each of these methods,
we identify groups of durations:
(1) Suite arrival and suite exit: these durations are never recorded in the OR software. They
are only used in the model to ensure that the patient starts and ends their pathway at the
right time and place. The duration of these steps is always equal to 1 minute.
(2) Preoperative care, patient in room, moving patient to PACU, and PACU
monitoring: the durations of all these steps is recorded in the OR software. They are not
strongly correlated.
(3) Setup (+ induction), procedure, reversal: these steps are all recorded in the OR
software. They are strongly correlated (usually a long procedure requires a long setup and
a long reversal; the contrary is true too).
(4) Putting a drip, long reception, short reception, LRA preparation, LRA, LRA
waiting, OIP, OR cleanup: these durations are never recorded in the or software. Thus,
based on our observations and interviews, we estimated their average durations: putting a
drip (5 minutes), long reception (15 minutes), short reception (5 minutes), LRA preparation
(15 minutes), LRA (30 minutes), LRA waiting (30 minutes), OIP (15 minutes), suite arrival
(1 minute), suite exit (1 minute).

115
Chapter V. Computing durations

As a side note:
• The OR cleanup duration depends on the perioperative duration (
• Table 30).
𝑂𝑅 𝑒𝑛𝑡𝑟𝑦 𝑂𝑅 𝑒𝑛𝑡𝑟𝑦
• If 𝑡𝑖𝑐𝑙𝑒𝑎𝑛𝑢𝑝 > 𝑡𝑖+1 then 𝑡𝑖𝑐𝑙𝑒𝑎𝑛𝑢𝑝 = 𝑡𝑖+1
Table 30 – Computing cleanup duration.
# Condition on perioperative phase duration D Cleanup Durations

1 𝐷 < 30 𝑚𝑖𝑛𝑢𝑡𝑒𝑠 5 minutes

2 30 ≤ 𝐷 ≤ 120 𝑚𝑖𝑛𝑢𝑡𝑒𝑠 15 minutes

3 𝐷 > 120 𝑚𝑖𝑛𝑢𝑡𝑒𝑠 30 minutes

Group#2 durations are computed as follows:

- Method#1. Each step duration is computed as the average step duration of cases with the
same surgery and the same surgeon.
- Method#2. Each step duration is computed as the average step duration of cases with the
same surgery only.
- Method#3. Each step duration is computed as a random value extracted once from the
uniform distribution dall surgeons ∗ 𝑈(𝑚𝑖𝑛, 𝑚𝑎𝑥), when min and max are model
parameters and dall surgeons is the average step duration of cases with the same surgery
only.
Group#3 durations are computed following the same formulas except that they are also multiplied
𝑑𝑠𝑢𝑟𝑔𝑒𝑜𝑛
by %𝑑𝑠𝑢𝑟𝑔𝑒𝑜𝑛 = 𝐴𝑉𝐺 ( patient in room ) for method#1, and %𝑑𝑎𝑙𝑙 𝑠𝑢𝑟𝑔𝑒𝑜𝑛 =
dsurgeon

𝑑𝑎𝑙𝑙 𝑠𝑢𝑟𝑔𝑒𝑜𝑛
𝐴𝑉𝐺 ( patient in room ) for method#2 and method#3.
d all surgeon

Group#4 durations are always equal to their constant value.

116
 Chapter V. Computing durations

5 – Apply the selected method to compute the stochastic provisional durations. We keep
the same step grouping.
Group#2 durations are computed as follows:
- Method#1. Each step duration is a random value extracted from the histogram made of all
the steps’ durations of cases with the same surgery and the same surgeon.
- Method#2: Each step duration is a random value extracted from the histogram made of all
the steps’ durations of cases with the same surgery.
- Method#3. Same as for the discrete duration.
Group#3 durations are computed following the same formulas except that they are also multiplied
𝑑𝑠𝑢𝑟𝑔𝑒𝑜𝑛
by %𝑑𝑠𝑢𝑟𝑔𝑒𝑜𝑛 = 𝐴𝑉𝐺 ( patient in room ) for method#1, and %𝑑𝑎𝑙𝑙 𝑠𝑢𝑟𝑔𝑒𝑜𝑛 =
dsurgeon

𝑑𝑎𝑙𝑙 𝑠𝑢𝑟𝑔𝑒𝑜𝑛
𝐴𝑉𝐺 ( patient in room ) for method#2 and method#3.
d all surgeon

Each step duration from group#4 durations is a random value extracted from d𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡 ∗
𝑈(𝑚𝑖𝑛, 𝑚𝑎𝑥), where min and max are model parameters, and d𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡 is the discrete duration.
6 – Compute the timestamps. The database provides us with the provisional room entry time.
Using this and the provisional durations, we can compute the other provisional timestamps if
necessary.

117
Chapter V. Computing durations

Table 31 – Synthesis of the methods used to compute the deterministic durations for the provisional schedule.

Duration
Method 1 Method 2 Method 3
Group

1 1 𝑚𝑖𝑛𝑢𝑡𝑒

A random value extracted once from

The average step duration of cases with the
The average step duration of cases with
2 same duration and the same surgeon: dall surgeons ∗ 𝑈(𝑚𝑖𝑛, 𝑚𝑎𝑥)
the same duration: 𝐴𝑉𝐺(dall surgeons )
𝐴𝑉𝐺(dsurgeon )
where min and max are model parameters.

Random value extracted once from

patient in room
%𝑑𝑎𝑙𝑙 𝑠𝑢𝑟𝑔𝑒𝑜𝑛 ∗ 𝐴𝑉𝐺 (dall surgeon )
patient in room
%𝑑𝑠𝑢𝑟𝑔𝑒𝑜𝑛 ∗ 𝐴𝑉𝐺( dsurgeon ) %𝑑𝑎𝑙𝑙 𝑠𝑢𝑟𝑔𝑒𝑜𝑛 ∗ 𝑑𝑝𝑎𝑡𝑖𝑒𝑛𝑡 𝑖𝑛 𝑟𝑜𝑜𝑚
∗ 𝑈(𝑚𝑖𝑛, 𝑚𝑎𝑥)
3 %𝑑𝑎𝑙𝑙 𝑠𝑢𝑟𝑔𝑒𝑜𝑛
𝑑𝑠𝑢𝑟𝑔𝑒𝑜𝑛
%𝑑𝑠𝑢𝑟𝑔𝑒𝑜𝑛 = 𝐴𝑉𝐺 ( patient in room
) 𝑑𝑎𝑙𝑙 𝑠𝑢𝑟𝑔𝑒𝑜𝑛 Where min and max are model parameters and
dsurgeon = 𝐴𝑉𝐺 ( patient in room
)
d all surgeon 𝑑𝑎𝑙𝑙 𝑠𝑢𝑟𝑔𝑒𝑜𝑛
%𝑑𝑎𝑙𝑙 𝑠𝑢𝑟𝑔𝑒𝑜𝑛 = 𝐴𝑉𝐺 ( patient in room )
d all surgeon

4 d𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡

118
 Chapter V. Computing durations

Table 32 - Synthesis of the methods used to compute the stochastic durations for the provisional schedule.

Duration
Method 1 Method 2 Method 3
Group

1 1 𝑚𝑖𝑛𝑢𝑡𝑒

Random value extracted from

Random value extracted from the Random value extracted from the histogram dall surgeons ∗ 𝑈(𝑚𝑖𝑛, 𝑚𝑎𝑥)
2
histogram made of dsurgeon . made of dall surgeon .
where min and max are model parameters.

Random value extracted from the Random value extracted from

histogram made of Random value extracted from the histogram %𝑑𝑎𝑙𝑙 𝑠𝑢𝑟𝑔𝑒𝑜𝑛 ∗ 𝑑𝑝𝑎𝑡𝑖𝑒𝑛𝑡 𝑖𝑛 𝑟𝑜𝑜𝑚 ∗
patient in room
%𝑑𝑠𝑢𝑟𝑔𝑒𝑜𝑛 ∗ dsurgeon
patient in room
made of %𝑑𝑎𝑙𝑙 𝑠𝑢𝑟𝑔𝑒𝑜𝑛 ∗ dall surgeon 𝑈(𝑚𝑖𝑛, 𝑚𝑎𝑥)

3
where min and max are model parameters
𝑑𝑠𝑢𝑟𝑔𝑒𝑜𝑛 𝑑𝑎𝑙𝑙 𝑠𝑢𝑟𝑔𝑒𝑜𝑛
%𝑑𝑠𝑢𝑟𝑔𝑒𝑜𝑛 = 𝐴𝑉𝐺 ( )) %𝑑𝑎𝑙𝑙 𝑠𝑢𝑟𝑔𝑒𝑜𝑛 = 𝐴𝑉𝐺 ( ) and %𝑑𝑎𝑙𝑙 𝑠𝑢𝑟𝑔𝑒𝑜𝑛 =
patient in room patient in room
dsurgeon d all surgeon
𝑑𝑎𝑙𝑙 𝑠𝑢𝑟𝑔𝑒𝑜𝑛
𝐴𝑉𝐺 ( patient in room )
d all surgeon

Random value extracted from d𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡 ∗ 𝑈(𝑚𝑖𝑛, 𝑚𝑎𝑥)

4
where min and max are model parameters.

119
Chapter V. Computing durations

5. Chapter synthesis
In this section, we proposed and illustrated a methodology to correct and complete the performed
and provisional schedule timestamps and duration values extracted from the real-life database to
feed the digital twin. Our contribution #6 is therefore two-fold. We begin by preprocessing our
available database values and then proceed to use these values with the surgical suite processes
described in Chapter IV to simulate both the performed and the provisional schedule with discrete
durations, as well as the provisional schedule with stochastic durations.

120
 Chapter VI. Modeling and scheduling non-elective cases

CHAPTER VI. MODELING AND SCHEDULING NON-ELECTIVE CASES

In Chapter IV, we have discussed how we model schedule execution in our DT-DSS using Flexsim
Healthcare®. This amounts to modeling operational and supporting processes. In Chapter V, we
have explained how we modeled the stochastic duration for each activity of the surgical suites’
processes. In this new Chapter (Chapter VI), we describe how we model and simulate non-elective
cases.
In this research, we have defined two types of non-elective cases: semi-urgent cases which must be
performed maximum three days after their admission, and urgent cases which must be performed
the same day of their admission. We do not consider elective add-on (Chapter I). In Figure 52, we
illustrate how the schedule evolves from the weekly staff meeting (D-1 week) to the end of the
execution day (D+1). The color code is the following: elective cases (green), semi-urgent cases
(yellow), urgent cases (orange). We represent:
1. The staff-validated provisional schedule at D-1 week: it only consists in elective cases.
2. The provisional schedule between the weekly staff meeting and the day before
schedule execution: it consists of the same elective cases plus semi-urgent non-electives
cases that were added throughout the week.
3. The provisional schedule at the start of the execution day: idem.
4. The performed schedule at the end of the execution day: on top of the previous
elective and semi-urgent cases, urgent case(s) have been scheduled.
In Table 33, we describe when and why we must perform non-elective case scheduling during the
application of our methodologies. We represent these steps on a timeline in Figure 53 to make it
more visual. It should be noted that the schedule at the start of the execution day, and the schedule
before the execution day are usually the same – they only differ if a case was added during the
night. The Training can take place at any time. Finally, in Table 34, we remind the reader of the
DT-DSS configuration for each of these instances.
In this Chapter, we describe our solution proposal, discuss how we modeled non-elective case
arrival, and non-elective case scheduling.

121
Chapter VI. Modeling and scheduling non-elective cases

Figure 52 – Illustration of how the schedule evolves from the weekly staff meeting to the end of
the execution day.

Table 33 - When and why do we perform non-elective case scheduling?

# Methodology Step Objective

1 Prospective Analysis 4 Assess the provisional schedule resilience.

Training on a Practice scheduling urgent cases during scheduling

2 /
Provisional Schedule execution based on a specific provisional schedule.

Test other scheduling solutions for the non-elective cases

3 Retrospective Analysis 2 that arrived during the schedule execution day (i.e. urgent
cases).

Practice scheduling the real urgent cases that arrived during

Training on a
4 / scheduling execution based on a specific performed
Performed Schedule
schedule.

Practice scheduling urgent cases based on a specific

5 Regular Training /
schedule.

122
 Chapter VI. Modeling and scheduling non-elective cases

Figure 53 - When and why do we perform non-elective case scheduling?

Table 34 – Description of the DT-DSS configuration for each time we must perform non-
elective case scheduling.

Schedule Constraint Pathway Duration

# Methodology Step
Type Type Type Type

Prospective 4
1 Provisional Strict Detailed Determinist
Analysis

2 Training / Provisional Strict Detailed Stochastic

Retrospective 2
3 Performed Flexible Aggregated Determinist
Analysis

4 Training / Performed Flexible Aggregated Determinist

5 Training / Any Any Any Any

123
Chapter VI. Modeling and scheduling non-elective cases

1. Solution proposal
First, for the prospective analysis and for the training on a provisional schedule, we want to
model non-elective arrivals at two moments: (1) after the weekly scheduling strategy, and (2) on
the day before schedule execution. In the first case, the provisional schedule consists in elective
cases. Consequently, we model the arrival of both semi-urgent and urgent non-elective cases. In
the second case, the provisional schedule consists in elective cases and semi-urgent cases that were
added throughout the week. Thus we model the arrival of only urgent non-elective cases. We call
these non-elective cases that were not in the initial schedule additional non-elective (ANE)
cases. An ANE case can either be a semi-urgent case or an urgent case.
Second, for the retrospective analysis and for the training on a performed schedule, we want
to model the same non-elective arrivals that were in the initial schedule to find other scheduling
solutions. Consequently, we do not use ANE cases for the retrospective analysis.
Finally, for regular training, the user is free to practice on a provisional or a performed schedule,
and to use either ANE or non-elective cases from the initial schedule depending on what they aim
to achieve.
We propose two types of non-elective scheduling in our DT-DSS: (1) an automatic mode where
the tool will schedule the non-elective case according to a specific strategy specified by the user
(first fit, best fit, worst fit), or (2) a manual mode where for each non-elective case to schedule
the tool will run the different possible scheduling scenarios, provide the results to the user and let
them choose which one they want to implement. The automatic mode is used for analysis and
the manual mode for training. We synthesize this in Table 35.
.

124
 Chapter VI. Modeling and scheduling non-elective cases

Table 35 - Description of non-elective modeling and scheduling parameters for the DT-DSS
configuration for each time we must perform non-elective case scheduling.

NE
Schedule Initial
# Methodology Time NE Source Scheduling
Type Schedule
Strategy

After the ANE (semi

Prospective
1 weekly Provisional Elective urgent + Automatic
Analysis urgent)
meeting

Training on a After the ANE (semi

2 Provisional weekly Provisional Elective urgent + Manual
Schedule meeting urgent)

Day before Initial schedule

Prospective Elective + (semi-urgent)
1 schedule Provisional Automatic
Analysis Semi-urgent + ANE
execution (urgent)

Training on a Day before Initial schedule

Elective + (semi-urgent)
2 Provisional schedule Provisional Manual
Semi-urgent + ANE
Schedule execution (urgent)

After Elective + Initial

Retrospective Schedule (semi
3 schedule Performed Semi-urgent Automatic
Analysis urgent +
Execution + Urgent urgent)

Training on a After Elective + Initial

Schedule (semi
4 Performed schedule Semi-urgent Manual
Performed urgent +
Schedule Execution + Urgent urgent)

1.1. Modeling additional non-elective (ANE) arrivals scenarios

Let “initial schedule” be either the provisional or the performed schedule that would have been
simulated in a deterministic environment. An ANE case is a non-elective case that is not in the
initial schedule. Our goal is to model scenarios of ANE arrivals in the surgical suite.
Upon entering the surgical suite, each ANE case is defined by (1) a suite entry time, (2) the list of
surgeons able to perform the surgery, (3) the list of ORs in which the case can be scheduled, (4) an
anesthesia type, (5) the case urgency level, and (6) the activities durations. Thus, before being
scheduled an ANE do not have an assigned surgeon, an assigned anesthesiologist, nor an assigned
OR. The anesthesia type defines which patient pathway the ANE case is going to follow.

125
Chapter VI. Modeling and scheduling non-elective cases

Since (a) the operating rooms of the surgical suite we consider are different and cannot receive all
surgical specialties, and (b) in some very rare instances a case can be scheduled in a room that is
not suitable for it, we make the following hypotheses:
H19. A case of specialty ‘a’ can be placed in room ‘b’ only if at least 10 cases of the same specialty were
scheduled in that room in the database.
In the database, since some specialties are similar to each other (orthopedics, orthopedics - upper
limbs, orthopedics – lower limbs, etc.) we do not associate each surgeon to one specialty. Instead,
we make the following hypothesis:
H20. A surgeon ‘a’ can perform a case from specialty ‘b’ only if they do a case from that specialty at
least once in the database.
H21. A surgeon can only perform a case of their specialty/specialties.
As a side note:
H22. There are very few vital emergencies at HPB, so we focus on non-vital ANE cases.
H23. Our case study occurs during a weekday. Thus, we only include weekday ANE cases. We do not
consider weekend ANE cases.
H24. We maintain our focus on the surgical suite. Consequently, we model the ANE cases from suite
entry to suite exit – meaning we do not model their pathway in the emergency department.

1.2. Creating ANE arrival scenarios

Each ANE scenario is defined by (1) the urgency level of the cases, (2) the number of incoming
cases, and (3) the cases’ arrival times window in the suite.
Urgency level and number of incoming cases. We consider the non-elective cases recorded in
the database to have realistic scenarios. First, we create two sets of cases: (a) days with at least one
semi-urgent case, and (b) days with at least one semi-urgent case or urgent case. Second, we
compute the number of cases per weekday for each separated set. Third, we divide these number
of days into 4 quartiles:
• The first quartile (Q1) consists in the days with the least number of arrivals (note that they
still have at least one arrival).
• The fourth quartile (Q4) consists in the days with the highest number of arrivals.
• The second and third quartile (Q2 and Q3) consists of the rest of the days.
The repartition into quartiles allows us to test different scenarios of NE arrivals. A day in Q1 can
be considered as a good day, Q2 and Q3 as a regular day, and Q4 as a bad day. Indeed the more
arrivals we have (and this disruptions), the worse the day non-elective cases are not trivial to
schedule. Note that we do not consider the weekdays as we have stated in a previous section.
Cases’ arrival time window. We propose 5 options for the arrival time window of ANE cases:
the arrivals can be randomly distributed between: (1) midnight and midnight, (2) 7am and 7pm, (3)
7am and 12am, (4) 12am and 2pm, (5) 2pm and 7pm. 7am is an hour before the earliest shift start.
7pm is one hour after the latest shift ends. The staff takes the lunch break between 12am and 2pm.
These scenarios correspond to ANE cases arriving during the entire day, during the OR shifts,

126
 Chapter VI. Modeling and scheduling non-elective cases

during the morning, during the lunch break, during the afternoon. This allows us to model up to:
(4 × 2 × 5) + 1 = 41 scenarios of additional non-elective arrivals.

1.3. Synthesis
Table 36 synthesizes the different parameters available to model ANE arrival scenarios.
Table 36 – Description of the parameters options to create ANE arrivals scenarios.

Time window during which

Urgency level of the ANE Number of incoming ANE
these ANE enter the suite

Semi-urgent + urgent None Midnight/midnight

Urgent Based on all days, 7am-7pm
Based on Q1 days 7am-12am
Based on Q2Q3 days 12am-2pm
Based on Q4 days 2pm-7pm

127
Chapter VI. Modeling and scheduling non-elective cases

2. Modeling non-elective scheduling scenarios

In the preceding section, we have described how we model ANE arrival scenarios. We now
describe how we schedule non-elective cases. We propose 6 scheduling strategies: (1) maintain the
initial scheduling (if possible), (2) first fit, (3) best fit, (4) worst fit, (5) manual scheduling, and (6)
push the case to a waiting line. Note that it is possible to maintain the initial schedule if we are
simulating a performed schedule execution; indeed in that case we already know where, and when
the non-elective has been scheduled in real life.

2.1. Automatic scheduling mode

The available automatic strategies in our DT-DSS are the following:
• First Fit (FF): schedule the case in the first available allowed OR. This means that the
non-elective case will be introduced between already scheduled cases.
• Best fit (BF): schedule the case after already scheduled cases in the allowed OR that will
have the least shift duration left at the end of the schedule execution.
• Worst fit (WF): schedule the case after already scheduled cases in the allowed OR that will
have the most shift duration left at the end of the schedule execution.
Note that in case the strategy is set to best fit or worst fit and that there is no OR with a shift
duration long enough to fit the non-elective case that needs to be scheduled, then the strategy
changes to minimizing the overtime. Consequently, it is possible for best fit and worst fit to
produce the same scheduling strategy. We illustrate these three strategies in Figure 54.

Figure 54 - Illustration of The Three Non-Elective Scheduling Strategies Allowed in our DT-DSS.

128
 Chapter VI. Modeling and scheduling non-elective cases

In Figure 54, the initial schedule consists in three ORs with a total of 12 patients. The simulation
runs until a non-elective case arrives (non-elective arrival time, red). At that time, OR#1 is
performing turnover #2, OR#2 is performing turnover #6 and OR#3 is performing surgery #11.
Depending on the simulation parameters the model will schedule the non-elective case in an OR.
We make the hypothesis that in this specific example all ORs are allowed for the case, but this
might not always be the case.
Case #13 can fit in any of the three ORs displayed in Figure 54 with the Best Fit strategy, case #13
is scheduled in OR#1 because it has the shortest shift duration left (first row). With the Worst Fit
strategy, case #13 is scheduled in OR#3 because it has the longest shift duration left (last row).
Finally, with the First Fit Strategy, case #13 is scheduled in OR#2 because it is the first room to
finish its ongoing case.
Once the non-elective case has been assigned to an OR#m at the nth rank, we assign it a surgeon
and an anesthesiologist. We follow these rules:
• For the surgeon:
1. If the surgeon performing the surgery (𝑛 − 1) of OR#m can perform surgery n,
then it is assigned to the case.
2. If not, if there is a surgeon that is present at the suite that day that can perform
surgery n, then it is assigned to the case n.
3. If not, a surgeon from the pool of surgeons allowed to perform surgery n pool is
randomly assigned to the case n.
• For the anesthesiologist:
o If there is an anesthesiologist performing an induction for the case (𝑛 − 1), then
it is assigned to the case n.
o If not, if there are anesthesiologists present at the suite that day, one is randomly
assigned to the case n.
o If not, an anesthesiologist from the anesthesiologist pool is randomly assigned to
the case n.

2.2. Manual scheduling mode

In this section, we describe how the user can manually schedule non-elective cases. The principle
is the following: while running the simulation of a schedule execution, the user might not want to
rely on the strategies described in the previous subsection (FF, BF, WF), or might want to compare
its own scheduling strategy with the suggested ones. The environment can be either determinist
(we know the durations and the non-elective cases to be scheduled in advance), or stochastic. We
describe the process in a stochastic mode as it is slightly more complicated. Indeed, with a
stochastic environment, we have to save the description of the disruptions in order to be able to
replay them, while in a deterministic environment, we do not need to do so since neither the arrivals
nor the durations change from one scenario/replication to another.
To begin with, the user starts the schedule execution simulation. Since it is a stochastic
environment, every activity duration is saved so that the schedule execution can be replicated later
on.

129
Chapter VI. Modeling and scheduling non-elective cases

The simulation runs until the arrival of the first non-elective case to schedule (Disruption #1 Time
in Figure 55. This disruption is called the “triggering disruption”. The simulation resets and the
experimenter is launched.

Figure 55 - Illustration of manual non-elective scheduling (1/3). At the first disruption, the main simulation
resets, and the experimenter is automatically launched.

In the Experimenter, the DT-DSS tries all the scheduling solutions possible:
• Scheduling strategies proposed by the model (FF, BF, WF).
• Exhaustive list of all the (rank, OR) possible for the case to be scheduled. For instance, if
the case arrives when OR#1 is performing its 2nd surgery, then the tool will try the rank 3,
4, 5, etc. until the last possible rank in the OR#1.
In Figure 56, we give the example of 2 replications for 2 different scenarios. The durations before
the disruption arrival time are all the same because they are considered as having already happened;
they have been saved and replayed. The durations after the disruption are stochastic: there is the
same number of cases but the durations are different.
Scenario 1 is an example of FF scheduling. Scenario 2 is an example of scheduling the case in
OR#n at rank 5. Once presented with the results, the user can choose which scenario they wish to
implement. In our case, it is scenario 1 (yellow start).

Figure 56 - Illustration of manual non-elective scheduling (2/3). The experimenter launches n scenarios of
m replications to test all the possible scheduling solutions for the non-elective. The user chooses to
implement one of the scheduling scenarios (yellow star).

The Experimenter then closes automatically, and the simulation starts over from 𝑡 = 0. The
performed schedule is replayed until 𝑡 = “disruption #1 time”. The DT-DSS automatically
schedules the disruptions as previously stated by the user and runs until the next disruption. Note

130
 Chapter VI. Modeling and scheduling non-elective cases

that the durations of the case after the disruption are not the same as in the Experimenter: This is
due to the stochastic nature of the environment. We illustrate this in Figure 57.

Figure 57 - Illustration of manual non-elective scheduling (3/3). The scheduling scenarios chosen by the
user is implemented in the main simulation. The simulation restart from t=0 until the next new disruption.

131
Chapter VI. Modeling and scheduling non-elective cases

3. Chapter synthesis
In this Chapter, we have presented how we model non-elective arrivals and how we add them to
the ongoing schedule. The first fit strategy is what the OR managers usually do when an urgent
case arrives (i.e. a case which needs to be operated as soon as possible). Best fit and worst fit
strategies are simplified strategies compared to the ones used in (Van Riet and Demeulemeester
2015) (i.e. BF descending, BF ascending, WF ascending, and WF descending). We consider a
flexible scheduling where (1) both elective and non-elective cases can be scheduled in each OR,
and (2) non-electives can be scheduled at any point of the schedule (Van Riet and Demeulemeester
2015).

132
 Chapter VI. Modeling and scheduling non-elective cases

PART 3 - PROOF OF CONCEPT

133
 Chapter VII. Presentation of the study case

CHAPTER VII. PRESENTATION OF THE STUDY CASE

In Part 1, we have presented the context and the problematic of our research. In Part 2, we have
described our solution proposal. In this third and last part, we provide a proof of concept based
on a study case.
First, we describe our partners’ healthcare facilities. Second, we discuss the database they provided
us with and explain how we structured and treated the data. Third, we present the method and the
criterion set used to choose a study case. Fourth, we present the study case we use for this research.

135
Chapter VII. Presentation of the study case

1. Presentation of our partners and their facilities

The Vivalto Santé French group was created in 2009. It consists in 91 private healthcare facilities
distributed between 6 countries (France, Switzerland, Portugal, Spain, Czech Republic, and
Slovakia). The group developed a model called the “Third Way”, which is based on a medical and
capitalistic partnership.27 Our first partner, the Hôpital Privé de la Baie (HPB), situated in Avranches,
France is part of this group. HPB welcomes 18500 patients per year and hires 70 doctors and 200
paramedical staff. 78% (8054) of its patients admitted for surgery are outpatients.28
Our second partner, the Centre Hospitalier d’Albi (CHA), situated in Albi, France, is part of the
“Coeur d’Occitanie” regional hospital group, which regroups 7 healthcare facilities and serves a
330,000 inhabitants’ population. Each year, the CHA realizes 1,400 births, 25,000 inpatient
admission, 120,000 consultations and 4,000 surgeries. The staff consists of 2000 professionals and
includes 200 doctors. On top of that, the CHA is the headquarters of the emergency medical service
of the Tarn department.29
Lastly, our third partner is the Centre Hospitalier Intercommunal de Créteil (CHIC), situated in Créteil,
France. It is a supporting facility of the “Hôpitaux Confluence” regional hospital group. The CHIC
is approved as a university hospital for 9 disciplines. The CHIC’s staff consists in 2587
professionals (405 medical staff and 2182 non-medical staff). Each year, the CHIC realizes 3,000
births and 14,000 surgeries.30
We present key information on these 3 healthcare facilities in numbers in Table 37. It is interesting
to note that, these partners allow us to have access to facilities:
- From both the public and private sectors.
- From both regional and university hospitals.
- With different activity volumes.
- With outpatient surgery services.
- With maternity, Intensive care unit and emergency services.
- With different surgical suite sizes.
- With different emergency services patient population.

27 [Link]
28 [Link]
29 [Link]
30 [Link]

136
 Chapter VII. Presentation of the study case

Table 37 – Brief presentation of the facilities of our partners.

Centre Hospitalier
Hôpital privé de La Centre Hospitalier
Facility Intercommunal de
Baie (HPB) d’Albi (CHA)
Créteil (CHIC)
Avranches, Manche Créteil, Val-de-Marne
Location Albi, Tarn (81)
(50) (94)
Sector Private Public Public
Total: unknown Total: 2000 people Total: 2587 people
Staff Medical: 70 doctors Medical: 200 doctors Medical: 405 doctors
Paramedical: 200ppl Paramedical: unknown Paramedical: unknown
172 beds including 33 616 beds and places,
outpatient surgery including 12
Accommodations 563 beds
places and 16 outpatient surgery
chemotherapy places places
18,500 patients/year
6,500 emergency visits
146,000 hospitalized
3,504 medicine
days
admissions 25, 000
336,000 consultations
10,342 surgery hospitalizations,
104,000 emergency
Activity admissions 120,000 consultations
visits
8,054 outpatient 4,000 surgeries
14,000 surgeries
surgery admissions 1 400 births
73,000 imaging acts
1106 follow-up and
3,000 births
rehabilitation care
admissions
Emergency Monday to Friday
24/7 24/7
Service (8:30am-7pm)
Maternity Service No Yes Yes
Intensive Care
No Yes31 Yes
Unit
1 suite with 10 ORs 1 suite with 8 ORs:
6 – surgery 6 – surgery 3 suites with several
Surgical suites
2 – ophthalmology 1 – endoscopy ORs each.
2 – endoscopy 1 – C-section

31 [Link]

137
Chapter VII. Presentation of the study case

2. Proposed method for choosing a study case

The database provided by our partners were descriptions of the performed schedule. HPB and
CHIC have an OR software while CHA does not. The databases are organized as follows: each
row corresponds to a surgical case and each column corresponds to an information about that
surgical case. In Table 38, we briefly present these databases.
Table 38 – Brief presentation of the OR software database provided by our partners.

Facility HPB CHA CHIC

Number of cases 75,253 147 26,213

January 2016 to Nov 2018 to June

Extraction dates 2021
December 2021 2019

Extraction duration 6 years 1 week 8 months

We focus on the HPB database because it is the first we got access to and because that it is a rather
rich database. We use the other databases afterwards to understand better what we could expect to
extract from another type of database or OR software, and if our data needs could be aligned with
regular OR software functionalities.
The initial HPB database contains 75,253 cases that were performed over 6 years. Our objective is
to extract data from one relevant day for which we can apply our prospective and retrospective
methodology using our DT-DSS. In Figure 58, we describe the macro-steps we followed to
structure and correct the initial database before extracting a study case. Step #2 is detailed in Figure
59 and in Chapter V. The schedule of potential study cases must respect conditions described in
Table 39.

2 - Correct and
1 - Populate Complete 4 - Chose and
3 - Correct Shift
And Structure Timestamps Extract Study
Schedule
Initial DB and Durations Case
Values

Figure 58 – Database Treatment Steps.

Figure 59 – Steps to Correct and Complete Timestamps and Durations Values from the Database.

138
 Chapter VII. Presentation of the study case

Table 39 – Schedule constraints that must be respected in to be able to reach our study objectives

# Objective Constraint
Simulate the execution of the provisional All performed and provisional timestamps
1 and the performed schedules for an entire must be available, whether initially or after
day. correction (see Chapter V).

Test the robustness of the schedule by

2 simulating its execution in an environment Not relevant.
where durations are stochastic.

Test the resilience of the schedule by

simulating its execution in an environment There must be at least one urgent case in
3
where there are stochastic arrivals of non- the performed schedule.
elective cases.

5 Be representative of a regular operating day. The day must be a weekday.

139
Chapter VII. Presentation of the study case

3. Description of the study case

3.1. Overview
In Table 40, we provide an overview of the characteristics of the provisional and the performed
schedule.
Table 40 – Overview of the Provisional Schedule and the Performed Schedule.

Provisional Schedule Performed Schedule Difference

# Surgeons 9 9 0

# Anesthesiologist 5 5 0

# Open Operating Room 6 6 0

# Bed in the Waiting Area 5 5 0

# Bed for LRA 3 3 0

# Bed for OIP 3 3 0

# Bed in PACU 14 14 0

# Case 51 53 0

# Elective Case 51 51 0

# Semi-urgent Case 0 0 0

# Urgent Case 0 2 +2

# Surgery Type 27 29 +2

# Represented Specialty 6 6 0

# Inpatient 9 10 +1

# Outpatient 42 43 +1

# Cases with LRA 3 4 +1

# Cases with OIP 19 19 0

140
 Chapter VII. Presentation of the study case

To display and analyze both schedules, we simulate their execution in a deterministic environment
with flexible resource constraints and aggregated processes. This means that the surgical suite
process consists only in the patient pathway (1-step preoperative care, 3-step perioperative care,
and 2-step post-operative care) without including the OR cleanup, and that the only resources we
consider are the nominative ORs.

3.2. Provisional schedule

3.2.1. Provisional master surgery schedule

We display the provisional master surgery schedule (MSS) in Figure 60.
Each row corresponds to one of the nine ORs. The waiting line is a fictional OR that we do not
use in the prospective analysis. The color code is the following: OR shift (green), OR off schedule
(blue), and OR off schedule but if a case scheduled during this time window would be considered
overtime (yellow and red).
The provisional MSS has 6 ORs opened throughout the day; there are 4 morning shifts from 8am
to 1pm (OR#1, OR#2, OR#6, and OR#8) and 5 afternoon shifts from 2pm to 6pm (OR#1,
OR#5 OR#6, OR#7, and OR#8).
Note that: (1) an OR cannot be preempted before it opens for its first shift, but it can perform
overtime during the lunch break and after the provisional end of the last shift, (2) we consider the
staff schedule to be the same as the MSS. This is also true for the performed schedule execution.

Figure 60 –Provisional Master Surgery Schedule.

3.2.2. Surgery schedule

We display the provisional schedule in Figure 61. The color code is the following: setup with
anesthesiologist (yellow), setup without anesthesiologist (blue), surgical procedure (green), reversal
(purple), idle time (light gray), and off schedule (dark gray). We keep the same color code for the
rest of this Chapter.
We highlight in black the window time during which there is at least one OR in overtime: it happens
during the lunch break from 1pm to 2pm (OR#1 and OR#6 and OR#8) and until 7:20pm (OR#4
and OR#5).

141
Chapter VII. Presentation of the study case

Figure 61 - Simulation of the Provisional Schedule Execution in a Deterministic Environment with

Aggregated Processes and Flexible Constraints.

3.2.3. KPI analysis

We display in Table 41 the KPI related to the OR utilization, and in Table 42 the ones related to
patient waiting time (PWT).
In Table 41, the column “duration” is in minutes (rounded to the closest value), and the column
“rate” is a percentage (rounded to the closest value) which is calculated according to the following
formula:
𝐾𝑃𝐼𝑑𝑢𝑟𝑎𝑡𝑖𝑜𝑛
𝐾𝑃𝐼𝑟𝑎𝑡𝑒 = ⁄𝑆𝑢𝑚 𝑜𝑓 𝐴𝑙𝑙 𝑂𝑅 𝑆ℎ𝑖𝑓𝑡𝑠

The utilization rate is 781%, and the overtime is 145%. This can be explained by the fact that some
OR are overutilized (5, 6 and 7) and others are underutilized (1, 2 and 8). We note that none of the
KPIs reach their target (except the global overutilization rate, since the suite is underutilized).
Based on this provisional schedule, we can suggest that the activity is not well distributed between
the different rooms. If the constraints imposed by specialties allow it (each OR can only welcome
a certain range of specialties), it might be interesting during the schedule execution to move cases
from overutilized ORs to underutilized ones. Plus, since the suite is underutilized, it could be
interesting to schedule more cases (elective add-ons, semi-urgent cases, urgent cases).
Table 41 - KPIs Related to the Operating Room Utilization.

KPI Duration (minutes) Rate (%) Rate Target

Utilization 1874.3 78,1% ~85%

Overtime 353.6 14,7% ≤ 5%

Idle 525.7 21,9% ≤ 5%

Overutilization 0.0 0,0% ≤ 5%

Underutilization 172.6 7,2% ≤ 5%

In Table 42, the KPIs are all in minutes. The KPIs show that there is no patient waiting time.
This is because the arrival times are computed so that the patients are available for surgery when

142
 Chapter VII. Presentation of the study case

the staff and the OR are themselves available for the surgery. Moreover, this simulation considers
flexible constraints on resources, so the ORs are the only limiting resources.
Table 42 - KPIs Related to the Patient Waiting Time.

Patient Waiting Time Minimum Maximum Average

Total 0.0° 0.0° 0.0°

Material Resources 0.0° 0.0° 0.0°

3.3. Performed schedule

3.3.1. Master surgery scheduling

We display the performed master surgery schedule (MSS) in Figure 62. Each row corresponds to
one of the nine ORs. The color code is the following: OR shift (green), OR off schedule (blue),
and OR off schedule but if a case scheduled during this time window would be considered overtime
(yellow and red). We keep the same code for all the MSS displayed in this manuscript.
The performed MSS has 6 ORs opened throughout the day; there are 5 morning shifts from 8am
to 1pm (OR#1, #2, #6, #7 and #8) and 5 afternoon shifts from 2pm to 6pm (OR#1, #5 #6, #7,
and #8). All these shifts were present in the provisional MSS (Figure 60) except the OR#7
morning shift from 8am to 1pm.

Figure 62 – Retrospective Analysis (step #1): Performed Master Surgery Schedule.

3.3.2. Surgery Schedule

We display the performed schedule in Figure 63; we obtained it by running a single-replication
scenario of the performed schedule execution in the DT deterministic environment. The color
code is the following: setup with anesthesiologist (yellow), setup without anesthesiologist (blue),
surgical procedure (green), reversal (purple), idle time (light gray), and off schedule (dark grey).
Note that the patient pathway steps are opaque for the elective cases and see-through for the non-
elective ones, and that we used a red line to indicate the non-electives for better readability. We
keep the same color code for the rest of the Chapter.

143
Chapter VII. Presentation of the study case

We highlighted in black the window time during which there is at least one OR is in overtime: it
happens during the lunch break from 1pm to 2pm (OR #1) and until 8:03pm (OR #1, #5, #6 and
#7). The only resources considered are the nominative operating rooms (flexible resource
constraints).

Figure 63 - Retrospective Analysis (step #1): Performed Surgery Schedule.

There are two urgent cases in the performed schedule that were not in the provisional schedule.
We describe them in Table 43, show their surgeon and room allowed in Table 44, and display their
steps durations in Table 45.
Case#1 arrived at 7:08pm, had a 14-minute preoperative step and entered OR#1 at 7:23pm.
Case#2 arrived at 5:15pm, had a 1-hour 45-minute long preoperative step and entered OR#5 at
7:03pm. Both cases entered at the earliest time possible and were the last case in their respective
OR.
We do not know why the OR manager decided to schedule the cases in these ORs, but we can try
to logically backtrack the reasons behind those choices. We know that (1) the cases can be
scheduled in OR#1 to 5, (2) OR#2 has been closed for several hours, (3) OR#3 and #4 were not
open at all throughout the day. Consequently OR#1 and OR#5 seem like the best choice. Case #2
was scheduled in OR#5 because it was the last room with the latest end shift. Case #1 was
scheduled in OR#1 because it was the last before one room that had been opened and continuing
in OR#5 would have meant needing to wait for the cleanup to be over.
Table 43 - Description of the Urgent Cases in the Performed Schedule.

Suite Arrival OR Entry Preoperative Max Rank in

# #OR OR allowed Rank
Time Time Care Location the OR

1 19:08:39 19:23:44 1 Waiting Area 1, 2, 3, 4, 5 8 8

2 17:15:27 19:03:06 5 LRA 1, 2, 3, 4, 5 5 5

144
 Chapter VII. Presentation of the study case

Table 44 - Study Case Description: OR and Surgeons Allowed for the Urgent Cases

Case ID OR ID Allowed Surgeons ID Allowed

1 [1, 2, 3, 4, 5] [17, 18, 19]

2 [1, 2, 3, 4, 5] [48, 49, 50]

Table 45 - Description of the Urgent Cases Durations in the Performed Schedule (in minutes).

Preoperative Surgical Move To PACU

# Setup Reversal
Care Procedure PACU Monitoring

1 14.08 19.6 14.3 5.9 5.8 61.9

2 106.7 6.9 8.4 2.8 0.3 10.8

3.3.3. KPI Analysis

During the performed schedule execution, the utilization, the overtime, the idle time, and
underutilization rates did not reach their targets (see Table 46). The utilization and idle time rate
values can partly be explained by the fact that the OR cleanup in included in the utilization rate in
the prospective analysis, but not in the retrospective analysis. This is because the performed
cleanup timestamps are not recorded in the original database.
Table 46 - KPI Related to the Operating Room Utilization.

KPI Duration (minutes) Rate (%) Rate Target

Utilization 2087 77% ~85%

Overtime 259 10% ≤ 5%

Idle 613 23% ≤ 5%

Overutilization 0 0% ≤ 5%

Underutilization 355 13% ≤ 5%

Regarding PWT, the total PWT is null because the arrival time of the patients in the surgical suite
in the performed schedule is computed so that the is ready for entering the OR only when the OR
is available itself. Note that we do not consider human resources since we are simulating the
performed schedule execution (flexible resource constraints).

145
Chapter VII. Presentation of the study case

Table 47 - KPI Related to the Patient Waiting Time.

Patient Waiting Time Minimum Maximum Average

Total 0.0 0.0 0.0

Material Resources 0.0 0.0 0.0

3.4. Comparison of KPIs for both schedules (provisional and performed ones)
In Table 48, we display the utilization related KPIs for both schedules as well as the gap between
them. The utilization and idle time barely change and stay under the performance target. The
overtime decreases by 5%, which is good. The underutilization increases by 6%. The capacity
required for performing all the surgeries (utilization + overtime) decreases from 93% to 87%. Since
there are no cancelled cases from the provisional schedule (on the contrary there are 2 more urgent
cases), this could mean that the stochastic durations are overestimated.
Table 48 - Comparison of KPIs for both schedules.

Provisional Performed
KPI Rate Target Gap
Rate Rate

Utilization 78% 77% ~85% - 1%

Overtime 15% 10% ≤ 5% -5%

Idle 22% 23% ≤ 5% +1%

Overutilization 0% 0% ≤ 5% /

Underutilization 7% 13% ≤ 5% +6%

146
 Chapter VII. Presentation of the study case

4. Chapter synthesis
In this Chapter, we have described how we obtained the data for our study case. In the remaining
of this manuscript, we will use this study case to provide a proof of concept for the use of a DT-
DSS to apply our prospective and retrospective analysis.

147
 Chapter VIII. Prospective analysis and retrospective analysis

CHAPTER VIII. PROSPECTIVE ANALYSIS AND RETROSPECTIVE

ANALYSIS

In Chapter VII, we have first presented the facilities of our three partners: HPB, CHA and CHIC.
Then, we have proposed a method to choose a study case from the surgical suite’ databases we had
access to. Finaly, we have applied this method to extract a one-day study case from HPB’s database.
We described both the provisional and the performed schedules of our study case in terms of
master surgery schedule, surgery schedule and KPIs.
This Chapter is structured into four sections, aimed at showcasing the application of both
retrospective and prospective analyses on our selected case study. Firstly, we delve into the
experimental setup necessary to conduct both analyses, with a detailed exposition of parameter
configurations (Section 1). Following this, we present the findings of the prospective analysis
(Section 2) alongside those of the retrospective analysis (Section 3). In concluding this manuscript,
we offer a synthesis of the chapter, highlighting the interconnection between the analyses and
elucidating how they provides insights to take tangible real-world actions (Section 4).

149
Chapter VIII. Prospective analysis and retrospective analysis

1. Experimentations design

1.1. Table with all the experimentations

Table 49 presents how to configure the DT-DSS to be able to run the prospective analysis (rows
1 to 5), and the retrospective analysis (rows 6 to 10). Each step (rows) is described in Chapter 3,
and each parameter are defined in Appendix #4. Our simulations were run on a machine with an
Intel(R) Core(TM) 𝑖7 − 9750𝐻 processor running at 2.6 𝐺𝐻𝑧, with 32 𝐺𝐵 of RAM and an
integrated graphics card. All simulations are from midnight to midnight (𝑡 = 0, 𝑡 = 11: 59 𝑝𝑚).
A run takes about 1 minute.

1.1.1. A note on the number of possible scenarios and the studied KPI
Number of possible scenarios. In the last column of Table 49, we indicated the number of
scenarios to be run for each experimentation; it is either 1 or 60. The 60 scenarios allow to represent
the arrival and scheduling of non-elective cases that were either in the initial schedule or not (see
Chapter 6). These scenarios are determined by the number of arrivals and the arrival time window:
# 𝑝𝑜𝑠𝑠𝑖𝑏𝑙𝑒 𝑠𝑐𝑒𝑛𝑎𝑟𝑖𝑜𝑠 = (# 𝑎𝑟𝑟𝑖𝑣𝑎𝑙 𝑤𝑖𝑛𝑑𝑜𝑤𝑠) ∗ (# 𝑁𝐸 𝑎𝑟𝑟𝑖𝑣𝑎𝑙𝑠) ∗
(# 𝑠𝑐ℎ𝑒𝑑𝑢𝑙𝑖𝑛𝑔 𝑠𝑡𝑟𝑎𝑡𝑒𝑔𝑖𝑒𝑠) = 5 ∗ 4 ∗ 3 = 60 scenarios
The options for each of these parameters are the following:
• Arrival time window (5 options): midnight - midnight, 7am - 7pm, 7am - noon, noon -
2pm, 2pm - 7pm.
• Number of NE arrivals (4 options). The number of arrivals is randomly selected from a
set of number of arrivals per day that consists in all days with NE cases, 25% of the days
with the least number of arrival (Q1: lower quartile), 25% of the days with the maximum
number of arrivals (Q4: upper quartile) and the rest of the days (Q2 and Q3)
• Scheduling strategies (3 options): FIFO, BF, WF
KPI studied. Throughout the experimentations, we consider performance indicators related to
OR usage (utilization, idle time, overtime, overutilization, and underutilization). We compute both
their duration in minutes and their percentages (duration divided by the sum of OR shift length).
Note that: (1) the surgical suite utilization is the sum of all the ORs utilization, (2) the surgical suite
utilization rate is the sum of all ORs utilization divided by the sum of all of the ORs shifts, and (3)
a similar computation can be made for each OR utilization related KPI.
For 1-replication scenarios, we display the discrete value of each indicator. For n-replication
scenarios, we display the mean rate value (95% confidence interval), the sample standard deviation,
the minimum and the maximum values. The OR usage KPIs are described in Chapter I.
We also consider different indicators related to patient waiting time (PWT) in minutes such as total
PWT, total PWT for human resources, total PWT for material resources, etc. For both 1-replication
and n-replication scenarios, we display the average, minimum and maximum values of each
indicator.

150
 Chapter VIII. Prospective analysis and retrospective analysis

1.1.2. Description of the Prospective Analysis

The prospective analysis consists of 5 steps. To begin with, we simulate the provisional schedule
execution in a deterministic environment to assess resource synchronization (step #1) and schedule
performance (step #2). Then we consider the impact of stochastic durations on the performance
to assess schedule robustness (step #3). After, we add additional non-elective arrivals, and we
compute the schedule performance for different scheduling strategies (step #4). Finally, we
simulate the provisional schedule execution in a stochastic environment that has both stochastic
durations and stochastic non-elective arrivals to assess both the robustness and resilience of the
provisional schedule at the same time (step #5).
Throughout the prospective analysis, we use detailed surgical processes and apply strict resource
constraints. First, we consider a detailed patient pathway, meaning that the preoperative care phase
is different depending on which anesthesia the patient requires. Second, regarding human
resources, we consider the surgeons, the anesthesiologists, and the OR nurses to be nominative.
Third and finally, on top of keeping the nominative operating rooms, we add non-nominative
limited preoperative beds (dedicated to the waiting area, the LRA and the OIP), and non-
nominative post-operative beds in the PACU. Note that nominative resources are necessarily
limiting and can provoke PWT (Chapter 4).
It should be noted that as there are many scenarios simulated at each step, we will not present the
results of each of them, but only present the ones that seem the most important to use.

1.1.3. Description of the retrospective Analysis

The retrospective analysis (described in Chapter III) consists in 3 different steps: simulate the
performed schedule execution in a determinist environment and compute its performance (step
#1), try different scheduling solutions for the urgent cases that arrived during schedule execution
(step #2), and identify the performance gap between the performed and the provisional schedules
and assess whether they are a consequence of offline or online operational decisions (step #3).
The simulation of the performed schedule is systematically realized in a deterministic environment
with an aggregated surgical suite process and flexible constraints on resources (justified in Chapter
IV). This means that the patient pathway is modeled based on the existing timestamps of the
database (suite arrival, OR entry, incision...), and that the resources modeled consist in nominative
ORs. All simulations are from midnight to midnight (t=0, t=11:59pm). A run takes about 1 minute.
We have already displayed the results of step #1 in Chapter VII. As for step #3, we refer the reader
to other papers we have previously worked on (Rifi, Fontanili, and Jeanney 2020; Rifi, Martinez, et
al. 2022; Rifi, Fontanili, and Jeanney 2022). Consequently, we only present step #2.

151
Chapter VIII. Prospective analysis and retrospective analysis

Table 49 - Experimentations performed by the DT-DSS

Keep NE Cases Add NE Arrivals

Initial Process Constraints on Duration # # Possible
# Objective of Initial to Initial
Schedule Type Resources Type Replications Scenarios
Schedule Schedule

Assess Resource
1 Deterministic No No 1 1
Synchronization

2 Assess Performance Deterministic No No 1 1

3 Assess Robustness Provisional Detailed Strict Stochastic No No 30 1

4 Assess Resilience Deterministic No Yes 30 60

Simulate Provisional
5 Stochastic No Yes 30 60
Schedule Execution

6 Assess Performance Performed Aggregated Flexible Deterministic No No 1 1

Assess other Scheduling

7 Performed Aggregated Flexible Deterministic Yes No 1 1
Strategy

8 Provisional Detailed Strict Stochastic No Yes 30 60

Identify the Root Cause

9 Performed Aggregated Flexible Deterministic No No 1 1
of Performance Gaps

10 Performed Aggregated Flexible Deterministic Yes No 30 60

152
 Chapter VIII. Prospective analysis and retrospective analysis

2. Prospective analysis experimentations

2.1. Steps #1 and #2. Assess resource synchronization and schedule performance

2.1.1. Introduction
In Chapter VII, we have displayed and analyzed the provisional schedule as validated by the staff.
Our next goal is to assess the resource synchronization (step #1) and to analyze the schedule
performance (step #2).

2.1.2. Simulation of the provisional schedule execution

In Figure 64 we display the results of the provisional schedule execution. The color code remains
the same as for Figure 61, except that we add waiting for anesthesiologist (red) and waiting for
surgeon (orange). Note that on the figure, we cannot see any orange color as the duration during
which patients wait for their surgeon is non-existent in the simulation. However, we can notice
several times where the patient is waiting in the OR for the anesthesiologist. This is because one
anesthesiologist is responsible for several rooms. We highlighted in black the window time during
which there is at least one OR in overtime: it happens during the lunch break from 1pm to 2pm
(OR#1, #2, #6) and until 10:20pm (OR #1, #4, #5 and #7).

Figure 64 – Prospective Analysis (steps #1 and #2): Simulation of the Provisional Schedule Execution in a
deterministic environment with detailed processes and strict constraints on resources.

2.1.3. KPI analysis

[Link]. Assess resource synchronization using patient waiting time

We assess resource synchronization thanks to the indicators related to PWT (i.e. patient idle
time) that we display in Table 50 – Prospective Analysis (steps #1 and #2): Global patient waiting
time., (global PWT), Table 51 (PWT for material resources) and Table 52 (PWT for human
resources). All waiting times are in minutes.
Note that in Table 50, the first row corresponds to the PWT for the right rank. This is since each
case in the schedule has a specific rank. Thus, if a patient arrives too early at the surgical suite, they
will receive preoperative care and then wait for their turn to be received by their OR nurse. In a
nutshell: the PWT for the right rank is a PWT for the OR nurse that is due to the patient arriving
early; this explains why we put it in the PWT for the human resources dashboard.

153
Chapter VIII. Prospective analysis and retrospective analysis

We highlight the cells that we discuss in our analysis. On average, each patient waits 74 minutes
(minimum 0 minutes, maximum 212 minutes). We note that the PWT is mainly for human
resources (minimum 0 minutes, maximum 175 minutes, average 61 minutes). This PWT is divided
between the anesthesiologist (min 0 minutes, max 34 minutes, average 3 minutes) and the OR nurse
(minimum 0 minutes, maximum 175 minutes, average 74 minutes) (right rank). In other words,
patients are scheduled to arrive in the suite about 1-hour 15-minutes too early. Please note that this
is not an observation that holds for all patients, since there is also OR idle time – meaning that
some patients are scheduled to arrive in the suite too late.
Consequently, resource synchronization improvements could be made by adapting patient arrival
times in the surgical suite.
Table 50 – Prospective Analysis (steps #1 and #2): Global patient waiting time.

# Patient Waiting Time Indicators Minimum Maximum Average

1 All resources 0 212 74
2 Human Resources 0 175 61
3 Material Resources 0 95 13
Table 51 - Prospective Analysis (steps #1 and #2): Patient waiting time for material resources.

# Patient Waiting Time Indicators Minimum Maximum Average

1 Preoperative Beds 0 95 13
2 Operating Room 0 18 1
3 Postoperative Beds 0 0 0
Table 52 - Prospective Analysis (steps #1 and #2): Patient waiting time for human resources.

# Patient Waiting Time Indicators Minimum Maximum Average

1 OR Nurse (Right Rank) 0 175 57
s OR Nurse (Other) 0 0 0
3 Anesthesiologist 0 34 3
4 Surgeon 0 0 0

[Link]. Schedule performance

We assess schedule performance thanks to the KPI displayed in Table 53. The utilization, the
underutilization and the overutilization rates respect the targets.

154
 Chapter VIII. Prospective analysis and retrospective analysis

Table 53 - Prospective Analysis (Step #1 and #2): KPI Related to the Operating Room Utilization.

Initial Provisional Schedule Step #1 & #2

Rate Rate Rate
KPI Duration
Duration (minutes) (duration/ (duration/ Target
(minutes)
planned open) planned open)
Utilization 1874.3 78.1 2022.4 84.3 ~85%
Overtime 353.2 14.7 369,1 15.4 ≤ 5%
Idle 525.7 21.9 377.6 15.7 ≤ 5%
Overutilization 0 0 0 0 ≤ 5%
Underutilization 172.6 7.2 8.5 0.3 ≤ 5%
Now that we have computed the provisional schedule performance in a determinist environment,
we are going to assess whether the surgical suite organization is able to maintain this level of
performance in a stochastic environment with stochastic durations, non-elective arrivals, or both.

2.2. Step #3. Robustness analysis

2.2.1. Introduction
In the previous steps we have studied the provisional schedule execution in a determinist
environment with either aggregated processes and flexible resource constraints (Chapter 7), or
detailed processes and strict resource constraints (previous section).
For this third step, we keep the last configuration and add stochastic durations. Our goal is to
assess whether the schedule is robust, meaning, whether it stays performant in case of variable
durations. We proceed to run 30 replications.
In this section, we present an analysis of the aggregated KPIs for all replications of the scenario
and display two examples of surgical schedules.

2.2.2. KPI analysis

We display the summary of the KPI values across all the replications in Table 54. We compare the
average values of this scenario (second column) with the values from the previous step (last
column):
• The utilization rate and the underutilization time both drop by almost 15%.
• The overtime rate increases by 2%, and the idle rate by almost 15%.
• The total PWT decreases by about 5 minutes.
At first glance, we could say that the organization does not maintain its performance when dealing
with duration variability. However, the 15% underutilization drop can probably explain the
utilization drop, and the idle rate rise of the same value. Indeed, it seems like the duration
stochasticity provoked a drop of 15% in the volume of required OR time.

155
Chapter VIII. Prospective analysis and retrospective analysis

Table 54 - Prospective Analysis (Step #3): Summary of the KPI values across all the replications. The
PWT are in minutes and the PWT results from steps #1 and #2 are the average value.

Step
KPI Rate STD Min Max Target
#1-2

Utilization 70.6 ± 1.1 3 65.5 78.2 ~85% 84.3

Overtime 13.5 ± 0.6 1.6 11.5 17.3 ≤ 5% 15.4

Idle 29.4 ± 1.1 3 21.8 34.5 ≤ 5% 15.7

Overutilization 0.00 ± N/A 0.0 0.0 0.0 ≤ 5% 0.0

Underutilization 15.9 ± 1.5 3.9 7.2 21.9 ≤ 5% 0.4

≤ 20
Total PWT 68.6 ± 4.2 11.3 51 95.1
minutes 74*

PWT for human /

55.5 ± 1.9 5.1 46.1 65.3
resources 61*

PWT for material /

13.1 ± 2.6 7.00 3.1 32.5
resources 13*

2.2.3. Surgical schedules

In Figure 65, we display the box plot of the utilization rate across the 30 replications we ran. We
display two examples of provisional schedule executions in Figure 66 (replication with the highest
utilization rate), and in Figure 67 (replication with the lowest utilization rate). It seems like the high
utilization comes from: (1) the duration of the first case of OR#5, and (2) the PWT for an
anesthesiologist that resulted from it in OR#1.

156
 Chapter VIII. Prospective analysis and retrospective analysis

Figure 65 - Prospective Analysis (Step #4): Utilization Box Plot.

Figure 66 - Prospective Analysis (Step #4): Provisional Schedule Execution with detailed pathway, strict
constraints, and stochastic durations, of the replication with the highest utilization rate.

Figure 67 - Provisional Schedule Execution with detailed pathway, strict constraints, and stochastic
durations, of the replication with the lowest utilization rate.

157
Chapter VIII. Prospective analysis and retrospective analysis

2.3. Step #4. Resilience analysis

2.3.1. Introduction
In the previous steps we have studied the provisional schedule execution in a determinist
environment (Chapter 7 and this Chapter), and in a stochastic environment where durations are
variable. We now want to assess the resilience of the provisional schedule. To do so, we simulate
its execution with different non-elective arrival scenarios, and different non-elective scheduling
strategy scenarios.
For a better readability, we present a single non-elective schedule arrival scenario. The number of
arrivals is randomly extracted from a histogram with the number of non-elective arrivals per day,
and the time window of arrival is set from midnight to midnight (i.e. the entire day).

2.3.2. All scenario: KPI analysis

Upon simulation the provisional schedule execution degraded upon the introduction of non-
elective arrivals. We note that the BF and WF strategies proposed the same solutions. This is
because these scheduling strategies shift to “schedule non-elective case at the end of the provisional
schedule while minimizing overtime” when there is not enough room at the end of any allowed
OR for scheduling the non-elective case which is our case here. Therefore, from now on we only
present the FF and the BF solutions. We display them in Table 55. We note that the utilization and
the overtime are slightly higher for the FF solution than for the BF one. The patient waiting time
confidence interval for FF is larger than that for BF.
In Table 56, we compare the performance of the schedule execution in a determinist environment
(step #1-2) with the schedule execution disrupted by non-elective arrivals (step #4). Except for the
PWT that increases by about 10 minutes (no matter the scheduling solution), the performance gap
between step #1-2 and step #4 is small.

158
 Chapter VIII. Prospective analysis and retrospective analysis

Table 55 –Prospective Analysis (step #4): Resilience FF VS BF

Scenario #1 - FF Scenario #2 - BF
KPI Target
Mean STD Min Max Mean STD Min Max

84.4 ± 83.1 ±
Utilization 2.1 77.4 88 1.4 78.4 84.3 ~85%
0.8 0.5

16.5 ± 15.7 ±
Overtime 1.5 11.6 20.7 0.7 13.7 16.5 ≤ 5%
0.6 0.2

15.6 ± 16.9 ±
Idle 2.1 12 22.6 1.4 15.7 21.6 ≤ 5%
0.8 0.5

1.5 ± 0.0 ±
Overutilization 1.6 0 5.5 0 0 0.1 ≤ 5%
0.6 0.0

0.6 ± 1.2 ±
Underutilization 2.1 0 11 1.7 0 6.5 ≤ 5%
0.8 0.6

86.9 ± 85.6 ± ≤ 20
Total PWT 37.9 64.4 247.5 13 65.9 125.9
14.2 4.9 minutes

PWT for human 65.0 ±

12.4 52.3 107.7 72 ± 3.6 9.7 55.1 92.1 /
resources 4.6

PWT for
21.9 ± 13.6 ±
material 26.7 9 139.8 6.9 9 34 /
10 2.6
resources

159
Chapter VIII. Prospective analysis and retrospective analysis

Table 56 - Prospective Analysis (step #4). Comparison of the performance of schedule execution in a
determinist environment with schedule execution disrupted by non-elective arrivals.

Step Step #4 + Gap Step #4 + Gap

KPI Target
#1-2 FF BF

Utilization 84.3 84.4 ± 0.8 + 0.1 83.1 ± 0.5 - 1.1 ~85%

Overtime 15.4 16.5 ± 0.6 + 1.2 15.7 ± 0.2 + 0.3 ≤ 5%

Idle 15.7 15.6 ± 0.8 - 0.1 16.9 ± 0.5 + 1.1 ≤ 5%

Overutilization 0 1.5 ± 0.6 + 1.5 0±0 + 0.0 ≤ 5%

Underutilization 0.4 0.6 ± 0.8 + 0.3 1.2 ± 0.6 + 0.8 ≤ 5%

74 ≤ 20
Total PWT 86.9 ± 14.1 + 12.9 85.6 ± 4.9 + 11.6
minutes

PWT for human 61

65 ± 4.6 +4.0 72 ± 3.6 + 11.0 /
resources

PWT for 13
material 21.9 ± 10 +8.9 13.6 ± 2.6 + 0.6 /
resources

2.3.3. Example of surgical schedule.

In Figure 68 and Figure 69, we display the surgical schedule for the replication where the FF
strategy led to the maximum utilization rate of the scenario (88%). The random disruptions are the
same so this allows to compare two scheduling strategies for the same schedule execution.
The color code remains the same as for Figure 61, except that we add PWT for anesthesiologists
(red for elective cases, pink for non-elective cases) and PWT for surgeons (orange). In this
replication, there are three additional non-elective arrivals.

Figure 68 - Prospective Analysis (step #4). Example of a surgical schedule with the FF strategy.

160
 Chapter VIII. Prospective analysis and retrospective analysis

Figure 69 - Prospective Analysis (step #4). Example of a surgical schedule with the WF strategy.

161
 Chapter VIII. Prospective analysis and retrospective analysis

2.4. Step #5. Simulate the provisional schedule execution in a stochastic

environment
To conclude our prospective analysis, we simulate the prospective schedule execution in a
stochastic environment with both variable durations and non-elective arrivals. For a better
readability, we present a single non-elective schedule arrival scenario. The number of arrivals is
randomly extracted from a histogram with the number of non-elective arrivals per days, and the
time window of arrival is set from midnight to midnight (i.e. the entire day).

2.4.1. KPI analysis

We display the KPI related to the FF scheduling strategy and the BF one in Table 57. Neither the
FF scheduling strategy nor the BF scheduling strategy allows to reach the KPI targets. The
utilization rate drop probably comes from the way the stochastic durations are computed.
Table 57 –Prospective Analysis (step #5): All

Scenario #1 - FF Scenario #2 - BF
KPI Target
Mean STD Min Max Mean STD Min Max

73.7 ± 69.2 ±
Utilization 4.2 66.14 81.8 3.5 61.2 78 ~85%
1.6 1.3

14.4 ± 14.2 ±
Overtime 2.3 11.8 19.7 2.1 10.9 18.5 ≤ 5%
0.9 0.8

26.3 ± 30.8 ±
Idle 4.2 18.2 33.9 3.5 22 38.8 ≤ 5%
1.6 1.3

Overutilization 0±0 0.1 0 0.4 0 ± N/A 0 0 0 ≤ 5%

16.6 ±
Underutilization 12 ± 2.1 5.6 0 21.5 4.6 5.1 25.3 ≤ 5%
1.7

86.9 ± 86.6 ± ≤ 20
Total PWT 38.7 55.9 255.7 17.4 62.5 151
14.5 6.5 minutes

PWT for human 61.9 ±

13.6 46.6 112.9 69.1 ± 4 10.7 49.1 98.4 /
resources 5.1

PWT for
17.6 ±
material 25 ± 9.7 26 9.2 142.8 9.8 4.2 52.6 /
3.6
resources

162
 Chapter VIII. Prospective analysis and retrospective analysis

2.4.2. Example of surgery schedules

In Figure 70 and Figure 71, we display the surgical schedule for the replication where the FF
strategy led to the maximum utilization rate of the scenario (81.8%). The random disruptions are
the same, so this allows to compare two scheduling strategies for the same schedule execution. The
color code remains the same as for Figure 61, except that we add PWT for anesthesiologist (red
for elective cases, pink for non-elective cases) and PWT for surgeon (orange). In this replication,
there are three additional non-elective arrivals.

Figure 70 - Prospective Analysis (step #5). Example #1 of a surgical schedule with the FF strategy.

Figure 71 - Prospective Analysis (step #5). Example #1 of a surgical schedule with the WF strategy.

In this second example, we display a replication where both the FF (Figure 72) and the BF (Figure
73) strategy scheduling lead to the same utilization. We notice that the case durations in OR#1 are
quite short. The utilization is the same because the non-elective case arrives late during the day and
can only be scheduled after the end of the last OR shift.

Figure 72 - Prospective Analysis (step #5). Example #2 of a surgical schedule with the FF strategy.

163
Chapter VIII. Prospective analysis and retrospective analysis

Figure 73 - Prospective Analysis (step #5). Example #2 of a surgical schedule with the WF strategy.

164
 Chapter VIII. Prospective analysis and retrospective analysis

3. Retrospective analysis - step#2. Test other scheduling strategies on the

performed schedule non-elective cases

3.1. Display surgical schedule

During the real performed schedule execution, the OR manager dealt with two urgent cases and
scheduled them in OR#1 and #2. In this step, we simulate the performed schedule execution as it
happened in real-life, but we change the urgent case scheduling solution. We test FF, BF and WF
strategies. BF and WF strategies give the same scheduling solution because there are no shift end
durations available for any of the urgent cases.
In Table 58, we display the Suite Arrival Time and the OR Entry time for each strategy. In Figure
74 and Figure 75, we display visually the performed schedule with respectively the FF strategy and
the WF strategy. We note that the FF strategy leads to scheduling the case in an OR#2 that has
been closed for several hours, and that the BF strategy leads to scheduling both cases one after the
other.
Table 58 - Retrospective Analysis (step #1 & 2): Comparison of the scheduling strategy

Real NE Scheduling FF NE scheduling

BF NE scheduling
OR Entry Time
# Suite Arrival Time OR Entry Time OR OR Entry Time #OR #OR
1 19:08:39 19:23:44 1 19:23:44 1 19:23:44 1
2 17:15:27 19:03:06 5 19:03:06 2 19:03:06 1

Figure 74 – Retrospective Analysis (step #2): First Fit Scheduling

165
Chapter VIII. Prospective analysis and retrospective analysis

Figure 75 – Retrospective Analysis (step #2): Best Fit Scheduling

3.2. Result analysis

In Table 59, we display side to side the KPI related to OR usage for the execution in a determinist
environment for the performed schedule (column 2), the performed schedule with an FF
scheduling strategy (column 3), and the performed schedule with a WF scheduling strategy (column
4).
The KPI values related to the OR usage are the same no matter the scheduling strategy (OR
manager, FF, WF). This is due to the fact that the urgent cases had arrival times and preoperative
durations such that they could only be scheduled in overtime, after the last OR shift end.
Consequently, the FF strategy and the WF strategy analysis are the same as the step #1 of the
prospective analysis (Chapter V).
Table 59 – Retrospective Analysis (step #2). KPIs Related to the Operating Room Utilization.

Rate (%) –
Performed Rate (%) –
Rate (%) –
schedule + FF Performed Rate
KPI Performed
(step #2) schedule + Bf Target
schedule (step#1)
(step #2)

Utilization 77.3 77.3 77.3 ~85%

Overtime 9.6 9.6 9.6 ≤ 5%

Idle 22.7 22.7 22.7 ≤ 5%

Overutilization 0.0 0.0 0.0 ≤ 5%

Underutilization 13.1 13.1 13.1 ≤ 5%

166
 Chapter VIII. Prospective analysis and retrospective analysis

4. Chapter synthesis
In this chapter we have presented how we can use the DT-DSS to provide a prospective analysis
on the provisional schedule, and a retrospective analysis on the retrospective schedule in order to
improve OR management. In this synthesis, we would like to give the reader an idea of how these
analyses could be implemented in the daily life of the suite. In Table 60, we take the example of a
surgical suite which scheduling meeting is on Thursday.
The second column describes what the user does with the DT-DSS. At every scheduling meeting,
we especially suggest running a retrospective analysis on the days that passed since the last meeting
(row #1), and to run a prospective analysis on the future week (row #3). The third column
describes the information acquired after each step of the analyses. The fourth and last column
gives examples of decisions / actions to take based on this information.

167
 Chapter VIII. Prospective analysis and retrospective analysis

Table 60 – How does the DT-DSS can be used during the weekly scheduling meeting in order to facilitate schedule execution?

Allows to get the following

# Actions in the DT-DSS And take the following decisions/actions in the real world
information (see C3)

Step 1. Compute Performed The information related to the suite performance can help when taking decisions related to opening more OR
Schedule Performance shifts or closing existing ones (punctually or regularly).
Run a retrospective analysis of:
Step 2. Assess the impact of
Thursday and Friday of week(i-1), The information related to the impact of different non-elective case scheduling strategies can either serve to
1 implementing other decisions to
and Monday, Tuesday and encourage or discourage OR managers to implement specific strategies.
tackle the disruptions.
Wednesday of week(i)
Step 3. Identify performance Performance gat root causes can either be from the tactical and offline operational level, or from the online
gap root causes operational level. Identifying the root causes can help target the efforts of the staff

Compare the results of the

Update the prospective analysis Update the activity duration modeling and the non-elective patient arrival modeling and add new disruptions
2 retrospective and the prospective
parameters. modeling.
analyses of week(i-1) and week(i)

Step 1 to 4. Assess feasibility, Provide to the OR manager and the rest of the stakeholders’ information on the relevance of the current
performance, robustness, and provisional schedule. They can then modify the schedule (add or remove case, change case sequence, etc.) and
resilience run the prospective analysis again.
Run a prospective analysis of
3
week(i+1) Step 5. Assess the best non-
elective case scheduling strategy Provide to the OR manager information as to what NE case scheduling strategy might be the most relevant to
based on the provisional use during schedule execution.
schedule

168
 Conclusion

CONCLUSION

169
Conclusion

1. Reminder of the problematic

As we demonstrated in Part 1 of this PhD Thesis, the surgical suite is an inherently complex
environment subject to uncertainties. This is due to numerous human and technical resources that
are involved and must be synchronized for each operating case (Figure 1 and Figure 6). This
complexity implies mastering the surgical cases' planning and scheduling problem. The planning
and scheduling problem is divided into 4 decision levels: (1) strategic (capacity planning problem),
(2) tactical (master surgery scheduling problem), (3) offline operational (surgery scheduling
problem) and (4) online operational (schedule execution management problem). In the State of the
Art Chapter (Chapter II), we show that there are few research works for treating the problems
occurring at the operational level compared to all the contributions at the 2 top levels. Despite this,
observations, and interviews in 5 hospital surgical suites allowed us to highlight that there is not
any decision-support tool for helping the surgical suite staff each time a disruption occurs. This is
the starting point of this PhD that focuses on the operational level.
At the end of the offline operational level, a provisional schedule is created. This schedule is used
by the OR manager as a guide during the online operational level (schedule execution), although it
is known from the start of the day that it will not be fully respected. Indeed, part of the complexity
of the surgical suite lies in the stochastic nature of its activity. We focus on two types of
uncertainties: (1) duration variability and (2) non-elective arrivals. These uncertainties can disrupt
the schedule execution; consequently, the operational levels are strongly linked to the disruption
management problem.
In Chapter II, we show that the predictive disruption management can take place during both offline and
online levels; its goal is to anticipate and account for the disruptions before they even occur. On
the other hand, the reactive disruption management is related to the online level only, during which the
OR manager provides a remedy for a disruption only after it has happened. Disruption
management aims at maintaining – or even improving – the performance of the provisional
schedule throughout its execution. In this work, we assess the surgical suite performance based on
three KPIs: (1) the OR utilization, (2) the staff overtime and (3) the average patient waiting time.
This leads us to our general research question: “How can we promote and maintain the
performance of a surgical suite’s organization under uncertainties?”. We focus on the disruption
management issue at the operational levels.
In the remainder of this conclusion, we discuss our contributions, the limits of our study, and our
future research perspectives.

170
 Conclusion

2. Contributions

2.1. Prospective and retrospective methodologies to analyze the operating schedule

(contributions #1 and #2)

The disruption management is mostly handled by the OR manager. We make three observations:

1. The day-to-day regulation is dependent on the quality of the surgery scheduling: a non-
performant provisional schedule is likely to lead to a non-performant performed schedule.
2. Feedback on schedule execution is only provided in case of a serious undesirable event.
Most of the days, disruption management strategies are never revisited.
3. The disruption management strategies are usually empirical and can vary depending on the
OR manager personality. To our knowledge, there is no specific training designed for OR
managers, which could be a partial explanation for the previously mentioned statement.

Based on these observations; we propose 5 sub-research questions:

Q1. How can we anticipate disruptions before schedule execution?

Q2. How can we assess the quality of the predictive disruption management?
Q3. How can we assess the quality of the reactive disruptions management?
Q4. How can we identify whether performance lack stems from offline or online decisions?
Q5. How can we train OR managers on disruption management?

To answer these research questions, we propose three methodologies:

1. A prospective methodology to assess the quality of the predictive disruption management

(contribution #1).
2. A retrospective methodology to assess the quality of the reactive management
(contribution #2).
3. A methodology to train the OR manager on disruption management in a virtual
environment (contribution #3).

2.2. Prospective analysis: assessing schedule robustness using a digital twin

(contribution #2)
To assess the robustness of the provisional schedule, we simulated the provisional schedule while
considering duration stochasticity.
We faced two main issues: neither the provisional schedule durations nor the standardized surgery
types were recorded in the database. This issue was addressed through our contribution #2. We
estimated stochastic durations for our specific case study only. We manually assigned a
standardized surgery label to each type of surgery performed during our case study. Subsequently,
we attempted to categorize the remaining historical data based on these standardized labels.
For each duration type (ex: preoperative care duration, setup duration, etc.), we proposed a
computation method based on the number of available datapoints with (1) the same surgery type

171
Conclusion

and the same surgeon 𝑛𝑠𝑎𝑚𝑒 𝑠𝑢𝑟𝑔𝑒𝑜𝑛 , or (2) the same surgery type only 𝑛𝑎𝑙𝑙 𝑠𝑢𝑟𝑔𝑒𝑜𝑛𝑠 . If
𝑛𝑠𝑎𝑚𝑒 𝑠𝑢𝑟𝑔𝑒𝑜𝑛 > 100, the duration is based on the historical durations of cases with the same
surgeon and the same surgery type. Else, if 𝑛𝑎𝑙𝑙 𝑠𝑢𝑟𝑔𝑒𝑜𝑛𝑠 > 100 the duration was based on the
historical durations of cases with the same surgery type only. Otherwise, we employed a uniform
distribution spanning from the minimum to the maximum duration of historical duration cases
with the same surgery type only. By adhering to this approach, we successfully obtained discrete
and stochastic provisional durations to simulate the provisional schedule execution in our digital
twin.

2.3. Prospective analysis: assessing schedule resilience using a digital twin

(contribution #2)
To assess the robustness of the provisional schedule, we wanted to simulate the provisional
schedule while considering non elective arrivals. To do so, we needed to create non-elective arrival
scenarios and provide plausible scheduling solutions. We had already proposed some
contribution related to that subject in (Abdoune et al. 2023; Rifi et al. 2023).
The main issues we faced while trying to create our non-elective arrival scenarios was that our
database did not specify (1) the cases’ urgency levels, (2) the maximum delay allowed for a case to
be scheduled, (3) the surgeons allowed to perform the surgery, (4) the OR in which the case could
be performed.
Consequently, we hypothesized that the maximum delay allowed was always respected and
specified the case urgency level based on the time difference between the patient admission in the
hospital and the patient arrival in the surgical suite. Surgeries performed on their admission day
were considered urgent, surgeries performed 4 days or more after their admission day were
considered elective, and the rest were considered semi-urgent. We created arrival scenarios based
on the number of arrivals per day and the time window during which the arrival happened.
Regarding the surgeons and the OR assignment to non-elective cases, we studied the historical
database and proposed rules based on the number of times a surgeon had practice a surgery within
a certain specialty, or an OR had welcome a case from a certain specialty.

2.4. Methodology to train the OR manager to disruption management in a virtual

environment (contribution #3)
Our contributions #3 is a methodology to train the OR manager on disruption management in a
virtual environment. We previously worked on this topic in (Rifi, Fontanili, and Jeanney
2020; Rifi, Martinez, et al. 2022; Rifi, Fontanili, and Jeanney 2022; Rifi, Fontanili, et al.
2022)
Although our DT-DSS can be used offline to support both our prospective and retrospective
analysis, it can also be used as if it was an online tool to support schedule execution. The principle
is the following: the OR manager (1) simulates the schedule execution in the DT-DSS and pretends
the tool is synchronized with a real surgical suite – meaning that what happens in the tool happens
in real life, (2) tests different scheduling strategies – including their own – in case of an urgent case

172
 Conclusion

arrival, (3) implements the strategy of their choice, (4) continues until there are no more urgent
cases to schedule, and (5) studies the organization performance based on the choice they have
made throughout the simulation.
We developed our DT-DSS so that the training can take place on a performed schedule or a
provisional schedule. In the first case (performed schedule), the processes’ durations and the urgent
cases are identical to what happened in real life; in other words, the virtual environment is
determinist. In the second case, the processes’ durations and the urgent case arrivals are stochastic;
they can be configured and can vary from one training session to another. While training on a
performed schedule, the goal is to assess whether or not the schedule execution could have been
better. While training on a provisional schedule, the goal is to anticipate before the schedule
execution, what could be the best way to deal with disruptions based on that specific provisional
schedule. Either way, these training sessions can be used to help OR managers assess the impact
of disruptions and their decisions on the surgical suite, as well as promote a common approach to
disruption management in a facility.

2.5. Building a digital twin of the surgical suite (contributions #4, #5, #6 and #7)
We proposed and applied a standardized methodology to build a DT-DSS for any specific surgical
suite. To do so we used our methodology to conceive and build a digital twin-based decision
support system of a surgical suite (contribution #4). In this sub-section, we discuss how we
concretely built our DT of a surgical suite.

2.5.1. Modeling and simulation of the schedule execution (contribution #5)

In Chapter IV, we describe how we model and simulate the schedule execution using discrete event
simulation in our surgical suite DT. This is quite a complex task as the surgical suite processes
gather multiple human resources and material resources that must be synchronized for a smooth
schedule execution. Since the OR database does not have all the data required to model such a
heavy process, we combined our on-site observations, staff interviews and the OR database. We
propose different approaches for the provisional and the performed schedule due to the fact that
their simulations have different goals.

[Link]. Modeling and simulation of the performed schedule execution

When simulating the performed schedule execution, the objective is to have a simulated schedule
execution that is as close to the performed schedule execution as possible (i.e. the performed and
the real patient room entry time are identical), then to exhaustively model the activity (i.e. model the
different steps of the preoperative patient pathway). The objective of this simulation is to replay
the execution of a past schedule and to be able to have feedback on the decisions that have been
taken after each disruption. Then, we would like to analyze if another decision should have been
taken, to keep the target performance.
Consequently, we model the performed surgical suite processes based on the timestamps available
in the database. We obtain an “aggregated” surgical suite process with a one-step preoperative
phase, a 3-step perioperative phase, a 2-step post-operative phase, and no OR cleanup. Regarding
resources, we only model nominative ORs (we call this “flexible” constraints on resources).

173
Conclusion

The reasons are the following:

1. Material resources: there are no information regarding material resources different than
the OR in the database.
2. Human resources: the only IDs recorded are those of the surgeons and the
anesthesiologists.
3. Human resources: the anesthesiologist ID is not always recorded, even in case of an
induction in the OR.
4. Human resources. The times during which the human resources are present in the OR
are not recorded – even the anesthesiologist and the surgeon. When simulating their
presence according to best practices (from room entry to room exit), or according to our
observations (room entry to incision for the anesthesiologist, room entry to the suture for
the surgeon), the simulated patient pathway does not match the reality anymore.

[Link]. Modeling and simulation of the provisional schedule execution

On the other hand, the provisional schedule is naturally based on estimations and hypotheses. For
instance, timestamps and durations are estimated as they are not known in advance and the
resource synchronization is assessed in an empirical manner. While executing the provisional
schedule, we want to get a good idea of all the disruptions that could happen.
Consequently, for the provisional schedule execution, we used both timestamps available in the
database and timestamps estimated based on our observations and staff interviews (see Chapter
VIII). This allows us to propose a richer patient pathway with three preoperative care options (no
induction, LRA and OIP), and to add an OR cleanup step at the end of every reversal. By
opposition to the previous term “aggregated”, we call this process “detailed”. Concerning human
resources, we model nominative surgeons (from incision to suture), anesthesiologists (from room
entry to induction) and OR nurses (from at least room entry to PACU entry). This can cause patient
waiting times during schedule execution. As for material resources, we keep nominative ORs and
add beds for preoperative care and post-operative care. Since the number of beds has probably
evolved throughout the 6-year database we work with, we cannot know exactly how many beds
were available. We decide to choose a number that does not provoke irrational patient waiting
times during provisional schedule execution. By opposition to the previous term “flexible”, we call
this resource constraint type “strict”.
Our contribution here lies in the fact that we use different approach for the provisional and the
performed schedule execution, and that the provisional schedule execution relies on a very detailed
surgical suite process and that we consider resources throughout the entire surgical suite patient
pathway – which as far as we know is quite rare.

2.5.2. Methodology to clean and complete the patient pathway timestamps of a real-world
surgical suite database (contribution #6)
To build a surgical suite digital twin, we must have access to a reliable OR software database. This
raises several issues. First, recording timestamps of each surgical case in the operating room
software (suite arrival, OR entry, incision start, suture start, OR exit, PACU entry, PACU exit) is
usually done manually by the suite staff. Not only is it not the priority as they are more preoccupied
by providing care to the patient, but they also have little to no interest in the data once the surgery

174
 Conclusion

day is over. Second, the OR software is catered for the hospital staff. Consequently, information
that are interesting to us, but that are not required by the staff might not even be available in the
database. Third, some surgical suites still record part of all their data on paper, which make the use
of their data impossible. This is notably the case of induction related timestamps and anesthesia
details.
Multiple issues are encountered when dealing with our data: canceled cases do not appear in the
database, the urgency level is never specified, performed timestamps can be either incorrect or
missing, and the only information available for the provisional database are the patient provisional
room entry time, their provisional room, and their provisional rank in the room.
Based the data available in the database, and the study perimeter, we propose a standardized
methodology to clean and structure an OR database. This global methodology includes our
contribution #6: “Methodology to clean and complete the patient pathway timestamps of a real-
world surgical suite database”. This methodology includes defining rules to: (1) identify performed
incorrect timestamps, (2) compute missing performed timestamps based on existing ones, and (3)
compute discrete and stochastic provisional durations based on historical performed durations. We
discuss this methodology and its results in detail in Chapter V.

2.5.3. Prototype of a digital twin-based decision support system applied to a real-world case
study (contribution #7)
After applying our standardized methodology to build the digital twin of a surgical suite, we
obtained our own DT-DSS curated to the HPB surgical suite. We used it in Part 3 to provide a
proof of concept for the potential use of our prospective and retrospective analysis, and it consists
in itself as the POC of contribution #5.

175
Conclusion

3. Limits and possible improvements

In this section we draw a parallel between our study’s limits and the possible improvements that
we could implement in future research. We discuss our need for a richer and more reliable database,
our surgical suite DT, the functionalities of the decision support system, and the choice of our
study case.

3.1. Data
The input data we used to feed our digital twin included the description of the provisional and the
performed operating and resource schedules. We did not have access to the full provisional data,
so we had to make modelling hypothesis. Consequently, we can never know what the real initial
provisional schedule was. Plus, there are probably methods that provide better results to estimate
discrete and stochastic provisional durations for the surgical suite activities than the ones we
obtained.

3.2. Surgical suite DT

We discuss potential improvements in our digital twin in terms of resources, uncertainties, and
perimeter.
Resources. In Table 61, we display a synthesis of how each surgical suite human resource is
modeled in our DT for the provisional schedule execution. The columns indicate: whether or not
the resource appears in the model (column 2), whether or not the resource is limited and can thus
provoke patient waiting times (column 3), whether or not the resource is nominative (column 4),
and whether or not the simulated resource is on the same schedule as in real-life (column 5). We
do the same for material resources in Table 62. These two tables clearly highlight both the limits
and the improvement possibilities for the modeling and simulation of resources in our surgical
suite DT. Note that for the performed schedule, even more work is needed as we only model
nominative ORs.

176
 Conclusion

Table 61 - Human resources modeled in our DT during the provisional schedule execution.

Staff Profession Modeled? Limited? Nominative? Real Schedule?

Surgeon X X X

Anesthesiologist X X X

OR Nurse X X

Anesthesiologist Nurse

PACU Nurse X

Radiologist Operators

Nurse Assistants X

Table 62 - Human resources modeled in our DT during the provisional schedule execution.

Real
Type of Material Resources Modeled? Limited? Nominative?
Schedule?

Operating Rooms X X X X

Preoperative Beds X X

Post-Operative Beds X X

Surgical Supplies (disposable)

Surgical Instruments (reusable)

Imaging and Visualization

Equipment

Surgical Support Equipment

We complete these two tables with a word on the anesthetist nurse which was not modeled in our
study. Although it is not a common practice, anesthetist nurses are legally allowed to perform the
induction in the OR instead of the anesthesiologist. In our study we have considered only
nominative anesthesiologists, which created patient waiting times (PWT) when the anesthesiologist
was working in two parallels ORs. However, in real life, it is possible that there were no PWT if
the anesthetist nurse performed the induction themselves. Modeling the dynamics between the
anesthesiologist and the anesthetist nurse could be interesting but would require having access to
their presence time in the OR, which is data hard to come by.

177
Conclusion

Uncertainty. Another limit of our model is that we only focus on two types of uncertainties: non-
elective arrivals and duration variability. It would be interesting to include uncertainties on patient
cancellation, patient no-show, patient suite arrival time, human resource unavailability (ex: sick
leave), and material resource unavailability (ex: equipment breakdown).
Perimeter. Our DT focus on the surgical suite process from the patient suite entry to the patient
suite exit. However, the surgical suite activity is strongly impacted by the activity of external services
– and vice-versa. It would be interesting to extend the surgical suite DT to the other services that
interact with the surgical suite. This includes the inpatient wards, the outpatient wards, the ICU,
and the stretcher-bearer services.

3.3. Decision support system

Our DSS has an interesting marge of improvement in terms of KPI range and in terms of
disruption management strategy options. Although in this study we mostly focused on OR
utilization, staff overtime and patient waiting time, more indicators could be computed.
Prospective analysis indicators. First, we could include more indicators catered to assessing the
quality of the predictive disruption management. We could develop a set of indicators dedicated to
assessing a provisional schedule’s robustness and resilience instead of using the performance gap
only, and it could also be interesting to study the stability of the operating schedule (i.e. its ability
to remain despite disruptions) during the prospective analysis.
Resource utilization indicators. Today, we only focus on the OR utilization. However, it would
be interesting to add indicators related to the staff utilization or to the PACU and preoperative
beds’ utilization.
Reactive disruption management strategy. Our current DSS allows three strategies for
scheduling non-elective cases. However, as we have seen in Chapter II, the scope of schedule
disruptions and online decisions to tackle them is broader than that. We discussed adding more
uncertainties (and thus disruptions) in the previous section. As for the decisions, it would be
interesting to:
• Add more options for non-elective case scheduling:
o model block, open and mixed strategy.
o Arbitrate between scheduling a case in overtime or postponing it.
• Add decisions such as online rescheduling resources (human or material) to deal with the
evolution of the demand throughout the day (case cancelling, new case arrivals, shift early
end, etc.).
Predictive disruption management strategy. Today, we only assess the quality of the predictive
disruption management by evaluating the robustness and the resilience of the provisional schedule.
This means that if the provisional schedule performance is low, it will most likely impact the
performed schedule performance. Consequently, a future improvement could be to give the user
insight as to how to improve an already built provisional schedule.

178
 Conclusion

User Point of View. One of the weaknesses of our study is that, although we performed many
on-site observations and staff interviews, as well as discussed at length the goals of our research
with an expert committee, we did not have the opportunity to let a OR manager test our DT-DSS.

3.4. Study cases

In our current study case, urgent cases arrive at the end of the day: their earliest possible room
entry time is after the last OR shift end. This strongly reduces the diversity of potential scheduling
solutions proposed by our tool.
It would be interesting to have a panel of different one-day schedules to have a more exhaustive
view of what insights our DT-DSS can provide depending on the situation.
1-day study case VS 1-week study case. In our study we use a one-day study case. However,
real-life OR managers dealing with scheduling and rescheduling decisions consider more than a
one-day horizon. For instance, an elective case can be postponed to the next day, or to the next
week. Same could be said for semi-urgent cases. A strong improvement would be to consider
several day study cases (for instance, a 1-week study case) and to develop decisions such as
scheduling or rescheduling a case on a one-week horizon.
Throughout our research, we had to address the issue of the lack of data in the database (partial
recording of the provisional schedule, no recordings of the staff presence in the OR, etc.). It would
open more modeling and simulation opportunities if we had access to a richer database. This would
allow us to describe more accurately the activity within the surgical suite:
- Material resources: nominative
- Human resources: nominative staff with specific processes.

179
Conclusion

4. Future research perspectives

Throughout this research project, we have developed a surgical suite digital twin based on discrete
event simulation, and we have used it as an offline decision support system to improve predictive
and reactive disruption management. In this section, we discuss future research perspectives.

4.1. An online DT-DSS

We propose to transform our offline DT-DSS into an online DT-DSS. This would allow us to
cover both the offline and the online sides of the operational decision-level. The objective would
then be to apply the OR manager training methodology to online management. This would allow
us to have: (1) a DT for monitoring (DTM), (2) a DT for predicting (DTP), and (3) a DT for
optimizing (DTO). This future work perspective is closely related to the work done by (Abdallah
Karakra 2021). We describe these below.
The requirements to develop an online tool include sensors to update the locations of patients,
staff, and material resources in real-time. Although several methods exist (geolocation, RFID
chips…), their implementation is technically complex and could be slowed down – and even
prevented – by staff reluctant to be monitored. The offline analysis is similar to the one we could
perform online. Thus we have decided to first develop a functional offline tool, before
transforming it into an online tool in a future project.
First, we propose to create a DT for monitoring the surgical suite activity (DTM). To do so,
we would develop an online connection between our current DT and the surgical suite. This
would allow to simulate the schedule execution in the DT at the same time as it is happening in
real life; at any time the DT would be a virtual replication of the real-life surgical suite.
Second, we suggest developing a DT for predicting the surgical suite activity (DTP). To do
so, we propose to use the DTM and to couple it with the decision-support system present in our
current tool. By applying our third methodology (methodology to train an OR manager), we would
provide the OR manager with an online tool that could help them to anticipate future disruptions
(predictive disruption management), or to assess the impact of disruptions that have already
occurred and to test the impact of different solutions to deal with them (reactive disruption
management).
Third, we submit the idea of capitalizing on both the DTM and the DTP to create a DT to
optimize the disruption management (DTO). This would require modeling and simulating
more types of disruptions and more types of disruption management solutions. Instead of just
displaying possible disruption management solutions like in the DTP, the tool would suggest the
best solution for a specific set of KPIs.

4.2. A surgical suite DT generator

One perspective could be to work on how to automatically generate a surgical suite DT. This is a
part that took a long time in our research. This would require to: (1) Propose an ideal surgical suite
DT database, and (2) Identify which parameters would be necessary.

180
 Table of contents

TABLE OF CONTENTS

Abstract ................................................................................................................................. i

Résumé................................................................................................................................. iii

Résumé long en français ..................................................................................................... v

1. Partie 1 - Contexte et problématiques .......................................................................... vi
1.1. Chapitre 1. Contexte de l’étude : le bloc opératoire et ses salles opératoires........ vi
1.2. Chapitre 2. Revue de littérature et travaux similaires ......................................... viii
2. Partie 2 – Proposition d’une méthode et d’un outil ....................................................... x
2.1. Chapitre 3. Proposition de solution ........................................................................ x
2.2. Chapitre 4. Modélisation et simulation du déroulement du programme opératoire
...................................................................................................................................... xi
2.3. Chapitre 5. Calculer les durées ............................................................................. xii
2.4. Chapitre 6. Modéliser et programmer les interventions non-électives ................ xiii
3. Partie 3 – Preuve de concept ....................................................................................... xv
3.1. Chapitre 7. Présentation du cas d’étude................................................................ xv
3.2. Chapitre 8. Application des analyses prospective et rétrospective. ...................... xv
4. Conclusion ................................................................................................................. xvii

Acknowledgements ........................................................................................................... xix

List of abbreviations .......................................................................................................... xx

Summary ........................................................................................................................... xxi

INTRODUCTION ........................................................................................... 1

PART 1 - CONTEXT AND PROBLEMATICS ........................................ 11

Chapter I. The context of the surgical suite / operating room ...................................... 13
1. Overview of the working of a surgical suite................................................................ 14
1.1. The inner workings of a surgical suite.................................................................. 14
1.2. The relationship between the surgical suite and outside units.............................. 15
2. Patient pathways involving the surgical suite.............................................................. 17
2.1. Patient features that influence the patient pathway .............................................. 17
2.2. From taking a surgeon appointment to being ready for hospital admission ......... 21

181
Table of contents
2.3. From the hospital admission to entering the surgical suite. ................................. 23
2.4. From entering to exiting the surgical suite: the patient pathway in the surgical
suite ............................................................................................................................. 25
2.5. After the surgical suite: from exiting the surgical suite to leaving the hospital ... 29
3. Surgical suite resources ............................................................................................... 30
3.1. Material resources ................................................................................................ 30
3.2. Human resources .................................................................................................. 32
4. Performance evaluation in the OR .............................................................................. 37
4.1. Brief review on performance in the OR ............................................................... 37
4.2. KPI used to assess schedule performance, robustness and resilience .................. 37
5. Chapter synthesis: selected on-site problematics ........................................................ 39

Chapter II. Background and related works.................................................................... 41

1. Planning and scheduling problem ............................................................................... 42
1.1. Strategic level ....................................................................................................... 42
1.2. Tactical level: master surgery scheduling problem .............................................. 43
1.3. Offline operational level: the surgery scheduling problem .................................. 44
1.4. Online operating level: the schedule execution problem ..................................... 45
2. The predictive and reactive disruption management problem .................................... 47
2.1. Introduction .......................................................................................................... 47
2.2. Uncertainties......................................................................................................... 47
2.3. The disruption management problem at the operational level ............................. 48
3. A digital twin for the surgical suite based on discrete event simulation ..................... 54
3.1. Discrete-event simulation ..................................................................................... 54
3.2. Digital twin ........................................................................................................... 56
4. Chapter synthesis......................................................................................................... 58

PART 2 - PROPOSED METHODOLOGY AND TOOL ......................... 59

Chapter III. Solution proposal ........................................................................................ 60
1. Prospective analysis of the provisional schedule ........................................................ 62
2. Retrospective analysis of the performed schedule ...................................................... 64
3. A training environment for the OR manager .............................................................. 65
4. Proposition of a standardized method to build a surgical suite digital twin ............... 66
5. Chapter synthesis......................................................................................................... 68

182
 Table of contents
Chapter IV. Modeling and simulation of the schedule execution ................................. 69
1. Proposal of different modelling approaches depending to available data ................... 71
1.1. Surgical suite process ........................................................................................... 71
1.2. Resource constraints ............................................................................................. 72
2. Application to our study case ...................................................................................... 74
2.1. Introduction .......................................................................................................... 74
2.2. Flexible and strict resource constraints ................................................................ 74
2.3. Aggregated surgical suite process with flexible constraints on resources............ 79
2.4. Detailed surgical suite process with strict constraints on resources ..................... 81
3. Additional modeling hypotheses applied to our simulation tool ................................. 85
3.1. Anesthesia type label and preoperative care option ............................................. 85
3.2. Patient movements modeling ............................................................................... 87
3.3. Modeling OR schedules ....................................................................................... 87
4. Chapter synthesis ......................................................................................................... 89

Chapter V. Computing durations .................................................................................... 91

1. Correcting the initial timestamps................................................................................. 93
1.1. Introduction .......................................................................................................... 93
1.2. Finding incoherent timestamps using the patient point of view ........................... 95
1.3. Finding incoherent timestamps using the operating room point of view ............. 95
1.4. Correction ............................................................................................................. 96
2. Computing durations ................................................................................................... 98
2.1. Introduction .......................................................................................................... 98
2.2. Computing initially available durations ............................................................... 98
2.3. Impactful criterion of the operating duration ....................................................... 98
2.4. Method for each correction................................................................................... 99
2.5. Estimate missing durations ................................................................................. 100
3. Computing missing timestamps ................................................................................ 101
3.1. Introduction ........................................................................................................ 101
3.2. Correct pathway extremities: suite entry and PACU exit................................... 102
3.3. Correct each patient pathway independently: incision, suture, PACU entry ..... 102
3.4. Correct patient pathway by considering them dependent on each other: room
entry, room exit ......................................................................................................... 105

183
Table of contents
3.5. Results after trying to compute the missing timestamps .................................... 108
4. Compute timestamps and durations for the provisional schedule ............................. 111
4.1. Introduction ........................................................................................................ 111
4.2. Step description .................................................................................................. 112
5. Chapter synthesis....................................................................................................... 120

Chapter VI. Modeling and scheduling non-elective cases ........................................... 121

1. Solution proposal....................................................................................................... 124
1.1. Modeling additional non-elective (ANE) arrivals scenarios .............................. 125
1.2. Creating ANE arrival scenarios.......................................................................... 126
1.3. Synthesis............................................................................................................. 127
2. Modeling non-elective scheduling scenarios ............................................................ 128
2.1. Automatic scheduling mode ............................................................................... 128
2.2. Manual scheduling mode ................................................................................... 129
3. Chapter synthesis....................................................................................................... 132

PART 3 - PROOF OF CONCEPT ............................................................ 133

Chapter VII. Presentation of the study case ................................................................. 135
1. Presentation of our partners and their facilities ......................................................... 136
2. Proposed method for choosing a study case.............................................................. 138
3. Description of the study case .................................................................................... 140
3.1. Overview ............................................................................................................ 140
3.2. Provisional schedule ........................................................................................... 141
3.3. Performed schedule ............................................................................................ 143
3.4. Comparison of KPIs for both schedules (provisional and performed ones) ...... 146
4. Chapter synthesis....................................................................................................... 147

Chapter VIII. Prospective analysis and retrospective analysis................................... 149

1. Experimentations design ........................................................................................... 150
1.1. Table with all the experimentations ................................................................... 150
2. Prospective analysis experimentations ...................................................................... 153
2.1. Steps #1 and #2. Assess resource synchronization and schedule performance . 153
2.2. Step #3. Robustness analysis .............................................................................. 155
2.3. Step #4. Resilience analysis ............................................................................... 158

184
 Table of contents
2.4. Step #5. Simulate the provisional schedule execution in a stochastic environment
................................................................................................................................... 162
2.4.2. Example of surgery schedules ......................................................................... 163
3. Retrospective analysis - step#2. Test other scheduling strategies on the performed
schedule non-elective cases ........................................................................................... 165
3.1. Display surgical schedule ................................................................................... 165
3.2. Result analysis .................................................................................................... 166
4. Chapter synthesis ....................................................................................................... 167

CONCLUSION ............................................................................................ 169

1. Reminder of the problematic ..................................................................................... 170
2. Contributions ............................................................................................................. 171
2.1. Prospective and retrospective methodologies to analyze the operating schedule
(contributions #1 and #2) ........................................................................................... 171
2.2. Prospective analysis: assessing schedule robustness using a digital twin
(contribution #2) ........................................................................................................ 171
2.3. Prospective analysis: assessing schedule resilience using a digital twin
(contribution #2) ........................................................................................................ 172
2.4. Methodology to train the OR manager to disruption management in a virtual
environment (contribution #3) ................................................................................... 172
2.5. Building a digital twin of the surgical suite (contributions #4, #5, #6 and #7) .. 173
3. Limits and possible improvements ............................................................................ 176
3.1. Data..................................................................................................................... 176
3.2. Surgical suite DT ................................................................................................ 176
3.3. Decision support system ..................................................................................... 178
3.4. Study cases ......................................................................................................... 179
4. Future research perspectives ...................................................................................... 180
4.1. An online DT-DSS ............................................................................................. 180
4.2. A surgical suite DT generator ............................................................................. 180

Table of contents .............................................................................................................. 181

References......................................................................................................................... 187

List of figures ................................................................................................................... 198

List of tables ..................................................................................................................... 203

Appendices ....................................................................................................................... 206

185
Table of contents
1. Appendix #1 – description of the study case............................................................. 207
2. Appendix #2 – visual illustration of our surgical suite digital twin .......................... 210
2.1. Introduction ........................................................................................................ 210
2.2. Illustration of The Schedule Execution in our Surgical Suite Digital Twin ...... 210
2.3. Another dashboard example ............................................................................... 212
3. Appendix #3 – description of publications ............................................................... 216
3.1. List of publications ............................................................................................. 216
3.2. Summary ............................................................................................................ 217
4. Appendix #4 – description of the dt-dss.................................................................... 220
4.1. Introduction ........................................................................................................ 220
4.2. Parameters .......................................................................................................... 222
4.3. Description of the patient pathway..................................................................... 224

Outil d’aide à la décision à base de jumeau numérique pour l’analyse prospective et

rétrospective d’un programme de bloc opératoire soumis à des incertitudes ........... 233

Digital Twin-Based Decision Support System for the Prospective and the
Retrospective Analysis of an Operating Room under Uncertainties .......................... 234

186
 References

REFERENCES

Abdelrasol, Zakaria, Nermine Harraz, and Amr Eltawil. 2014. “Operating Room Scheduling
Problems: A Survey and a Proposed Solution Framework.” In Transactions on Engineering
Technologies, edited by Haeng Kon Kim, Sio-Iong Ao, and Mahyar A. Amouzegar, 717–
31. Dordrecht: Springer Netherlands. [Link]
Abdoune, Farah, Leah Rifi, Franck Fontanili, and Olivier Cardin. 2023. “Handling Uncertainties
with and Within Digital Twins.” In Service Oriented, Holonic and Multi-Agent Manufacturing
Systems for Industry of the Future, edited by Theodor Borangiu, Damien Trentesaux, and
Paulo Leitão, 1083:118–29. Studies in Computational Intelligence. Cham: Springer
International Publishing. [Link]
Abramovici, Michael, Jens Christian Göbel, and Hoang Bao Dang. 2016. “Semantic Data
Management for the Development and Continuous Reconfiguration of Smart Products
and Systems.” CIRP Annals 65 (1): 185–88. [Link]
Addis, Bernardetta, Giuliana Carello, Andrea Grosso, and Elena Tànfani. 2016. “Operating
Room Scheduling and Rescheduling: A Rolling Horizon Approach.” Flexible Services and
Manufacturing Journal 28 (1): 206–32. [Link]
Aivaliotis, P., K. Georgoulias, Z. Arkouli, and S. Makris. 2019. “Methodology for Enabling
Digital Twin Using Advanced Physics-Based Modelling in Predictive Maintenance.”
Procedia CIRP, 52nd CIRP Conference on Manufacturing Systems (CMS), Ljubljana,
Slovenia, June 12-14, 2019, 81 (January): 417–22.
[Link]
Allen, Robert W., Kevin M. Taaffe, and Gilbert Ritchie. 2014. “Surgery Rescheduling Using
Discrete Event Simulation: A Case Study.” In Proceedings of the Winter Simulation Conference
2014, 1365–76. [Link]
Augusto, Vincent, Xiaolan Xie, and Viviana Perdomo. 2008. “Operating Theatre Scheduling
Using Lagrangian Relaxation.” European Journal of Industrial Engineering 2 (2): 172.
[Link]
———. 2010. “Operating Theatre Scheduling with Patient Recovery in Both Operating Rooms
and Recovery Beds.” Computers & Industrial Engineering / Computers and Industrial
Engineering 58 (March). [Link]
Baesler, Felipe, J. Gatica Fuentes, and Rodrigo Correa. 2015. “Simulation Optimisation for
Operating Room Scheduling.” [Link]
Ballestín, Francisco, Ángeles Pérez, and Sacramento Quintanilla. 2019. “Scheduling and
Rescheduling Elective Patients in Operating Rooms to Minimise the Percentage of
Tardy Patients.” Journal of Scheduling 22: 107–18.
Bam, Maya, Brian T. Denton, Mark P. Van Oyen, and Mark E. Cowen. 2017. “Surgery
Scheduling with Recovery Resources.” IISE Transactions 49 (10): 942–55.
[Link]
Banks, Jerry. n.d. HANDBOOK OF SIMULATION Principles, Methodology, Advances, Applications,
and Practice.
Boschert, Stefan, and Roland Rosen. 2016. “Digital Twin—The Simulation Aspect.” In
Mechatronic Futures: Challenges and Solutions for Mechatronic Systems and Their Designers, edited
by Peter Hehenberger and David Bradley, 59–74. Cham: Springer International
Publishing. [Link]

187
References
Bovim, Thomas Reiten, Marielle Christiansen, Anders N. Gullhav, Troels Martin Range, and
Lars Hellemo. 2020. “Stochastic Master Surgery Scheduling.” European Journal of
Operational Research 285 (2): 695–711. [Link]
Bruni, M. E., P. Beraldi, and D. Conforti. 2015. “A Stochastic Programming Approach for
Operating Theatre Scheduling under Uncertainty.” IMA Journal of Management
Mathematics 26 (1): 99–119. [Link]
Caputo, F., A. Greco, M. Fera, and R. Macchiaroli. 2019. “Digital Twins to Enhance the
Integration of Ergonomics in the Workplace Design.” International Journal of Industrial
Ergonomics 71 (May): 20–31. [Link]
Cardoen, Brecht, Erik Demeulemeester, and Jeroen Beliën. 2010. “Operating Room Planning
and Scheduling: A Literature Review.” European Journal of Operational Research 201 (3):
921–32. [Link]
Castro, Pedro M., and Inês Marques. 2015. “Operating Room Scheduling with Generalized
Disjunctive Programming.” Computers & Operations Research 64 (December): 262–73.
[Link]
Ceschia, Sara, and Andrea Schaerf. 2016. “Dynamic Patient Admission Scheduling with
Operating Room Constraints, Flexible Horizons, and Patient Delays.” Journal of
Scheduling 19 (4): 377–89. [Link]
Chaabane, Sondes, Nadine Meskens, Alain Guinet, and Marius Laurent. 2006. “Comparison of
Two Methods of Operating Theatre Planning: Application in Belgian Hospital.” In 2006
International Conference on Service Systems and Service Management, 1:386–92. IEEE.
Childers, Christopher P., and Melinda Maggard-Gibbons. 2018. “Understanding Costs of Care
in the Operating Room.” JAMA Surgery 153 (4): e176233–e176233.
[Link]
Choi, Sangdo, and Wilbert E. Wilhelm. 2014. “On Capacity Allocation for Operating Rooms.”
Computers & Operations Research 44 (April): 174–84.
[Link]
Cima, Robert R., Michael J. Brown, James R. Hebl, Robin Moore, James C. Rogers, Anantha
Kollengode, Gwendolyn J. Amstutz, Cheryl A. Weisbrod, Bradly J. Narr, and Claude
Deschamps. 2011. “Use of Lean and Six Sigma Methodology to Improve Operating
Room Efficiency in a High-Volume Tertiary-Care Academic Medical Center.” Journal of
the American College of Surgeons 213 (1): 83–92.
[Link]
Demeulemeester, Erik, Jeroen Beliën, Brecht Cardoen, and Michael Samudra. 2013. “Operating
Room Planning and Scheduling.” In Handbook of Healthcare Operations Management:
Methods and Applications, edited by Brian T. Denton, 121–52. International Series in
Operations Research & Management Science. New York, NY: Springer.
[Link]
Dexter, F. 2000. “A Strategy to Decide Whether to Move the Last Case of the Day in an
Operating Room to Another Empty Operating Room to Decrease Overtime Labor
Costs.” Anesthesia and Analgesia 91 (4): 925–28. [Link]
200010000-00029.
Dexter, F., A. Macario, and L. O’Neill. 1999. “A Strategy for Deciding Operating Room
Assignments for Second-Shift Anesthetists.” Anesthesia and Analgesia 89 (4): 920–24.
[Link]
Dexter, F., A. Macario, and R. D. Traub. 1999. “Which Algorithm for Scheduling Add-on
Elective Cases Maximizes Operating Room Utilization? Use of Bin Packing Algorithms

188
 References
and Fuzzy Constraints in Operating Room Management.” Anesthesiology 91 (5): 1491–
1500. [Link]
Dexter, F., and R. D. Traub. 2000a. “Sequencing Cases in the Operating Room: Predicting
Whether One Surgical Case Will Last Longer than Another.” Anesthesia and Analgesia 90
(4): 975–79. [Link]
———. 2000b. “Statistical Method for Predicting When Patients Should Be Ready on the Day
of Surgery.” Anesthesiology 93 (4): 1107–14. [Link]
200010000-00036.
Dexter, F., R. D. Traub, and P. Lebowitz. 2001. “Scheduling a Delay between Different
Surgeons’ Cases in the Same Operating Room on the Same Day Using Upper Prediction
Bounds for Case Durations.” Anesthesia and Analgesia 92 (4): 943–46.
[Link]
Dexter, Franklin, Richard H. Epstein, Rodney D. Traub, Yan Xiao, and David C. Warltier.
2004. “Making Management Decisions on the Day of Surgery Based on Operating
Room Efficiency and Patient Waiting Times.” Anesthesiology 101 (6): 1444–53.
[Link]
Dexter, Franklin, Alex Macario, and Rodney D. Traub. 1999. “Optimal Sequencing of Urgent
Surgical Cases.” Journal of Clinical Monitoring and Computing 15 (3): 153–62.
[Link]
Dexter, Franklin, Thomas C Smith, David J Tatman, and Alex Macario. 2003. “Physicians’
Perceptions of Minimum Time That Should Be Saved to Move a Surgical Case from
One Operating Room to Another: Internet−based Survey of the Membership of the
Association of Anesthesia Clinical Directors (Aacd).” Journal of Clinical Anesthesia 15 (3):
206–10. [Link]
Dexter, Franklin, and Rodney D. Traub. 2002. “How to Schedule Elective Surgical Cases into
Specific Operating Rooms to Maximize the Efficiency of Use of Operating Room
Time.” Anesthesia & Analgesia 94 (4): 933–42. [Link]
200204000-00030.
Dios, Manuel, Jose M. Molina-Pariente, Victor Fernandez-Viagas, Jose L. Andrade-Pineda, and
Jose M. Framinan. 2015. “A Decision Support System for Operating Room
Scheduling.” Computers & Industrial Engineering 88 (October): 430–43.
[Link]
Duma, Davide, and Roberto Aringhieri. 2015. “An Online Optimization Approach for the Real
Time Management of Operating Rooms.” Operations Research for Health Care, ORAHS
2014 - The 40th international conference of the EURO working group on Operational
Research Applied to Health Services, 7 (December): 40–51.
[Link]
Elayan, Haya, Moayad Aloqaily, and Mohsen Guizani. 2021. “Digital Twin for Intelligent
Context-Aware IoT Healthcare Systems.” IEEE Internet of Things Journal 8 (23): 16749–
57. [Link]
Erdem, Ergin, Xiuli Qu, and Jing Shi. 2012. “Rescheduling of Elective Patients upon the Arrival
of Emergency Patients.” Decision Support Systems 54 (1): 551–63.
Erol, Tolga, Arif Furkan Mendi, and Dilara Doğan. 2020. “The Digital Twin Revolution in
Healthcare.” In 2020 4th International Symposium on Multidisciplinary Studies and Innovative
Technologies (ISMSIT), 1–7. [Link]
Evans, G.W., E. Unger, and T.B. Gor. 1996. “A Simulation Model for Evaluating Personnel
Schedules in a Hospital Emergency Department.” In Proceedings Winter Simulation
Conference, 1205–9. [Link]

189
References
Ewen, Hanna, and Lars Mönch. 2014. “A Simulation-Based Framework to Schedule Surgeries
in an Eye Hospital.” IIE Transactions on Healthcare Systems Engineering 4 (4): 191–208.
[Link]
Forbus, John J., and Daniel Berleant. 2022. “Discrete-Event Simulation in Healthcare Settings:
A Review.” Modelling 3 (4): 417–33.
Fügener, Andreas, Sebastian Schiffels, and Rainer Kolisch. 2017. “Overutilization and
Underutilization of Operating Rooms - Insights from Behavioral Health Care
Operations Management.” Health Care Management Science 20 (1): 115–28.
[Link]
Grieves, Michael, and John Vickers. 2017. “Digital Twin: Mitigating Unpredictable, Undesirable
Emergent Behavior in Complex Systems.” In Transdisciplinary Perspectives on Complex
Systems: New Findings and Approaches, edited by Franz-Josef Kahlen, Shannon Flumerfelt,
and Anabela Alves, 85–113. Cham: Springer International Publishing.
[Link]
Guda, Harish, Milind Dawande, Ganesh Janakiraman, and Kyung Sung Jung. 2016. “Optimal
Policy for a Stochastic Scheduling Problem with Applications to Surgical Scheduling.”
Production and Operations Management 25 (7): 1194–1202.
[Link]
Guerriero, Francesca, and Rosita Guido. 2011. “Operational Research in the Management of
the Operating Theatre: A Survey.” Health Care Management Science 14 (1): 89–114.
[Link]
Guido, Rosita, and Domenico Conforti. 2017. “A Hybrid Genetic Approach for Solving an
Integrated Multi-Objective Operating Room Planning and Scheduling Problem.”
Computers & Operations Research 87 (November): 270–82.
[Link]
Guinet, Alain, and Sondes Chaabane. 2003. “Operating Theatre Planning.” International Journal
of Production Economics 85 (1): 69–81.
Gul, Serhat, Brian T. Denton, John W. Fowler, and Todd Huschka. 2011. “Bi-Criteria
Scheduling of Surgical Services for an Outpatient Procedure Center.” Production and
Operations Management 20 (3): 406–17. [Link]
5956.2011.01232.x.
Günal, M M, and M. Pidd. 2010. “Discrete Event Simulation for Performance Modelling in
Health Care: A Review of the Literature.” Journal of Simulation 4 (1): 42–51.
[Link]
Hans, Erwin W., and Peter T. Vanberkel. 2012. “Operating Theatre Planning and Scheduling.”
In Handbook of Healthcare System Scheduling, edited by Randolph Hall, 105–30.
International Series in Operations Research & Management Science. Boston, MA:
Springer US. [Link]
Harris, Sean, and David Claudio. 2022. “Current Trends in Operating Room Scheduling 2015
to 2020: A Literature Review.” Operations Research Forum 3 (1): 21.
[Link]
Hashemi Doulabi, Seyed Hossein, Louis-Martin Rousseau, and Gilles Pesant. 2016. “A
Constraint-Programming-Based Branch-and-Price-and-Cut Approach for Operating
Room Planning and Scheduling.” INFORMS Journal on Computing 28 (3): 432–48.
[Link]
He, Tian-yong, and Wei Xiang. 2013. “Surgery Rescheduling Based on Pareto Solution Set
Under Uncertain Resource.” In Proceedings of 20th International Conference on Industrial

190
 References
Engineering and Engineering Management, edited by Ershi Qi, Jiang Shen, and Runliang Dou,
389–97. Berlin, Heidelberg: Springer. [Link]
Heydari, Mehdi, and Asie Soudi. 2016. “Predictive / Reactive Planning and Scheduling of a
Surgical Suite with Emergency Patient Arrival.” Journal of Medical Systems 40 (1): 30.
[Link]
J. Brown, Michael, Arun Subramanian, Timothy B. Curry, Daryl J. Kor, Steven L. Moran, and
Thomas R. Rohleder. 2014. “Improving Operating Room Productivity via Parallel
Anesthesia Processing.” International Journal of Health Care Quality Assurance 27 (8): 697–
706. [Link]
Jones, David, Chris Snider, Aydin Nassehi, Jason Yon, and Ben Hicks. 2020. “Characterising
the Digital Twin: A Systematic Literature Review.” CIRP Journal of Manufacturing Science
and Technology 29 (May): 36–52. [Link]
Kamran, Mehdi A., Behrooz Karimi, and Nico Dellaert. 2020. “A Column-Generation-
Heuristic-Based Benders’ Decomposition for Solving Adaptive Allocation Scheduling
of Patients in Operating Rooms.” Computers & Industrial Engineering 148 (October):
106698. [Link]
Karakra, A., E. Lamine, F. Fontanili, and J. Lamothe. 2020. “HospiT’Win: A Digital Twin
Framework for Patients’ Pathways Real-Time Monitoring and Hospital Organizational
Resilience Capacity Enhancement.” In Proceedings of the 9th International Workshop on
Innovative Simulation for Healthcare (IWISH 2020), 62–71. CAL-TEK srl.
[Link]
Karakra, Abdallah. 2021. “HospiT’Win : Designing a Discrete Event Simulation-Based Digital
Twin for Real-Time Monitoring and near-Future Prediction of Patient Pathways in the
Hospital.” Phdthesis, Ecole des Mines d’Albi-Carmaux. [Link]
[Link]/tel-03437096.
Karakra, Abdallah, Franck Fontanili, Elyes Lamine, and Jacques Lamothe. 2019. “HospiT’Win:
A Predictive Simulation-Based Digital Twin for Patients Pathways in Hospital.” In 2019
IEEE EMBS International Conference on Biomedical & Health Informatics (BHI), 1–4.
Chicago, IL, USA: IEEE. [Link]
———. 2022. “A Discrete Event Simulation-Based Methodology for Building a Digital Twin
of Patient Pathways in the Hospital for near Real-Time Monitoring and Predictive
Simulation.” Digital Twin 2: 1.
Karakra, Abdallah, Franck Fontanili, Elyes Lamine, Jacques Lamothe, and Adel Taweel. 2018.
“Pervasive Computing Integrated Discrete Event Simulation for a Hospital Digital
Twin.” In 2018 IEEE/ACS 15th International Conference on Computer Systems and
Applications (AICCSA), 1–6. [Link]
Kaye, Deborah R, Amy N Luckenbaugh, Mary Oerline, Brent K Hollenbeck, Lindsey A Herrel,
Justin B Dimick, and John M Hollingsworth. 2020. “Understanding the Costs
Associated With Surgical Care Delivery in the Medicare Population.” Annals of Surgery
271 (1): 23–28. [Link]
Kharraja, Said, Pascal Albert, and Sondes Chaabane. 2006. “Block Scheduling: Toward a Master
Surgical Schedule.” In 2006 International Conference on Service Systems and Service Management,
1:429–35. IEEE.
Knapp, G. L., T. Mukherjee, J. S. Zuback, H. L. Wei, T. A. Palmer, A. De, and T. DebRoy.
2017. “Building Blocks for a Digital Twin of Additive Manufacturing.” Acta Materialia
135 (August): 390–99. [Link]

191
References
Kooij, Rimmert van der, Martijn R.K. Mes, and Erwin W. Hans. 2014. “Simulation Framework
to Analyze Operating Room Release Mechanisms.” In Proceedings of the Winter Simulation
Conference 2014, 1144–55. [Link]
Koppka, Lisa, Lara Wiesche, Matthias Schacht, and Brigitte Werners. 2018. “Optimal
Distribution of Operating Hours over Operating Rooms Using Probabilities.” European
Journal of Operational Research 267 (3): 1156–71.
[Link]
Kroer, Line Ravnskjær, Karoline Foverskov, Charlotte Vilhelmsen, Aske Skouboe Hansen, and
Jesper Larsen. 2018. “Planning and Scheduling Operating Rooms for Elective and
Emergency Surgeries with Uncertain Duration.” Operations Research for Health Care 19
(December): 107–19. [Link]
Lamiri, Mehdi, Xiaolan Xie, Alexandre Dolgui, and Frédéric Grimaud. 2008. “A Stochastic
Model for Operating Room Planning with Elective and Emergency Demand for
Surgery.” European Journal of Operational Research 185 (3): 1026–37.
[Link]
Latorre-Núñez, Guillermo, Armin Lüer-Villagra, Vladimir Marianov, Carlos Obreque,
Francisco Ramis, and Liliana Neriz. 2016. “Scheduling Operating Rooms with
Consideration of All Resources, Post Anesthesia Beds and Emergency Surgeries.”
Computers & Industrial Engineering 97 (July): 248–57.
[Link]
Lebowitz, Philip. 2003. “Schedule the Short Procedure First to Improve OR Efficiency.”
AORN Journal 78 (4): 651–54, 657–59. [Link]
6.
Lehtonen, Juha‐Matti, Paulus Torkki, Antti Peltokorpi, and Teemu Moilanen. 2013. “Increasing
Operating Room Productivity by Duration Categories and a Newsvendor Model.”
International Journal of Health Care Quality Assurance 26 (2): 80–92.
[Link]
Leng, Jiewu, Hao Zhang, Douxi Yan, Qiang Liu, Xin Chen, and Ding Zhang. 2019. “Digital
Twin-Driven Manufacturing Cyber-Physical System for Parallel Controlling of Smart
Workshop.” Journal of Ambient Intelligence and Humanized Computing 10 (3): 1155–66.
[Link]
Liang, Feng, Yuanyuan Guo, and Richard Y. K. Fung. 2015. “Simulation-Based Optimization
for Surgery Scheduling in Operation Theatre Management Using Response Surface
Method.” Journal of Medical Systems 39 (11): 159. [Link]
0349-5.
Liu, Mengnan, Shuiliang Fang, Huiyue Dong, and Cunzhi Xu. 2021. “Review of Digital Twin
about Concepts, Technologies, and Industrial Applications.” Journal of Manufacturing
Systems, Digital Twin towards Smart Manufacturing and Industry 4.0, 58 (January): 346–
61. [Link]
Liu, Shiyong, Yan Li, Konstantinos P. Triantis, Hong Xue, and Youfa Wang. 2020. “The
Diffusion of Discrete Event Simulation Approaches in Health Care Management in the
Past Four Decades: A Comprehensive Review.” MDM Policy & Practice 5 (1):
238146832091524. [Link]
Liu, Ying, Lin Zhang, Yuan Yang, Longfei Zhou, Lei Ren, Fei Wang, Rong Liu, Zhibo Pang,
and M. Jamal Deen. 2019. “A Novel Cloud-Based Framework for the Elderly Healthcare
Services Using Digital Twin.” IEEE Access 7: 49088–101.
[Link]

192
 References
Lutters, Eric. 2018. “PILOT PRODUCTION ENVIRONMENTS DRIVEN BY DIGITAL
TWINS.” The South African Journal of Industrial Engineering 29 (3): 40–53.
[Link]
Ma, Guoxuan, and Erik Demeulemeester. 2013. “A Multilevel Integrative Approach to Hospital
Case Mix and Capacity Planning.” Computers & Operations Research, Operations research
for health care delivery, 40 (9): 2198–2207. [Link]
Macario, Alex, Terry S. Vitez, Brian Dunn, Tom McDonald, and Byron Brown. 1997. “Hospital
Costs and Severity of Illness in Three Types of Elective Surgery.” Anesthesiology: The
Journal of the American Society of Anesthesiologists 86 (1): 92–100.
Madubuike, Obinna C., and Chimay J. Anumba. 2022. “Digital Twin Application in Healthcare
Facilities Management,” May, 366–73. [Link]
———. 2023. “Digital Twin–Based Health Care Facilities Management.” Journal of Computing in
Civil Engineering 37 (2): 04022057. [Link]
Madubuike, Obinna Chimezie, Chinemelu J. Anumba, and Evangelia Agapaki. 2023. “Scenarios
for Digital Twin Deployment in Healthcare Facilities Management.” Journal of Facilities
Management ahead-of-print (ahead-of-print). [Link]
0107.
Makboul, Salma, Said Kharraja, Abderrahman Abbassi, and Ahmed El Hilali Alaoui. 2022. “A
Two-Stage Robust Optimization Approach for the Master Surgical Schedule Problem
under Uncertainty Considering Downstream Resources.” Health Care Management Science
25 (1): 63–88. [Link]
Marcon, Eric, and Franklin Dexter. 2006. “Impact of Surgical Sequencing on Post Anesthesia
Care Unit Staffing.” Health Care Management Science 9 (1): 87–98.
[Link]
Marjamaa, R., A. Vakkuri, and O. Kirvelä. 2008. “Operating Room Management: Why, How
and by Whom?” Acta Anaesthesiologica Scandinavica 52 (5): 596–600.
[Link]
May, Jerrold H., William E. Spangler, David P. Strum, and Luis G. Vargas. 2011. “The Surgical
Scheduling Problem: Current Research and Future Opportunities.” Production and
Operations Management 20 (3): 392–405. [Link]
5956.2011.01221.x.
Mercier, Gregoire, and Gerald Naro. 2014. “Costing Hospital Surgery Services: The Method
Matters.” PLOS ONE 9 (5): e97290. [Link]
M’Hallah, R., and A. H. Al-Roomi. 2014. “The Planning and Scheduling of Operating Rooms:
A Simulation Approach.” Computers & Industrial Engineering 78 (December): 235–48.
[Link]
Molina-Pariente, Jose M., Victor Fernandez-Viagas, and Jose M. Framinan. 2015. “Integrated
Operating Room Planning and Scheduling Problem with Assistant Surgeon Dependent
Surgery Durations.” Computers & Industrial Engineering 82 (April): 8–20.
[Link]
Ng, Nathan, Rodney A. Gabriel, Julian McAuley, Charles Elkan, and Zachary C. Lipton. 2017.
“Predicting Surgery Duration with Neural Heteroscedastic Regression.”
arXiv:1702.05386 [Cs, Stat], July. [Link]
Niu, Qing, Qingjin Peng, Tarek El Mekkawy, Yin Yin Tan, Helga Bruant, and Leanne Bernaerdt.
2007. “PERFORMANCE ANALYSIS OF THE OPERATING ROOM USING
SIMULATION.” Proceedings of the Canadian Engineering Education Association (CEEA).
[Link]

193
References
Peng, Yidong, Xiuli Qu, and Jing Shi. 2014. “A Hybrid Simulation and Genetic Algorithm
Approach to Determine the Optimal Scheduling Templates for Open Access Clinics
Admitting Walk-in Patients.” Computers & Industrial Engineering 72 (June): 282–96.
[Link]
Perdomo, Viviana, Vincent Augusto, and Xiaolan Xie. 2006. “Operating Theatre Scheduling
Using Lagrangian Relaxation.” 2006 International Conference on Service Systems and Service
Management, October, 1234–39. [Link]
Persson, Marie, Helena Hvitfeldt-Forsberg, Maria Unbeck, Olof Gustaf Sköldenberg, Andreas
Stark, Paula Kelly-Pettersson, and Pamela Mazzocato. 2017. “Operational Strategies to
Manage Non-Elective Orthopaedic Surgical Flows: A Simulation Modelling Study.”
BMJ Open 7 (4): e013303. [Link]
Przasnyski, Zbigniew H. 1986. “Operating Room Scheduling: A Literature Review.” AORN
Journal 44 (1): 67–82.
Rachuba, Sebastian, Lisa Imhoff, and Brigitte Werners. 2022. “Tactical Blueprints for Surgical
Weeks – An Integrated Approach for Operating Rooms and Intensive Care Units.”
European Journal of Operational Research 298 (1): 243–60.
[Link]
Rahimi, Iman, and Amir H. Gandomi. 2021. “A Comprehensive Review and Analysis of
Operating Room and Surgery Scheduling.” Archives of Computational Methods in Engineering
28 (3): 1667–88. [Link]
“Result of the Implementation of a Quality Management System Based on the ISO 9001:2015
Standard in a Surgical Intensive Care Unit.” 2023. Revista Española de Anestesiología y
Reanimación (English Edition) 70 (1): 26–36.
[Link]
Rifi, Leah, Franck Fontanili, and Michel Jeanney. 2020. “Proposition d’une Démarche Outillée
d’analyse Rétrospective Du Déroulement Du Programme Au Bloc Opératoire :
Application à La Régulation.” In GISEH 2020 - 10ème Conférence Francophone En Gestion
et Ingénierie Des Systèmes Hospitaliers. Valenciennes, France. [Link]
03229530.
———. 2022. “A Tool-Based Approach to Analyze Operating Room Schedule Execution:
Application to Online Management.” In Healthcare Systems, 179–93. John Wiley & Sons,
Ltd. [Link]
Rifi, Leah, Franck Fontanili, Cléa Martinez, Maria Di Mascolo, and Virginie Fortineau. 2023.
“A Simulation-Based Approach for Assessing the Impact of Uncertainty on Patient
Waiting Time in the Operating Room.” In Proceedings of the Winter Simulation Conference,
1057–68. WSC ’22. Singapore, Singapore: IEEE Press.
Rifi, Leah, Franck Fontanili, Maria Di Mascolo, and Cléa Martinez. 2022. “Framework for a
Retrospective Analysis of Operating Room Schedule Execution.” International Journal of
Healthcare Technology and Management 19 (1): 37.
[Link]
Rifi, Leah, Clea Martinez, Maria Di Mascolo, and Franck Fontanili. 2022. “Proposition d’ un
outil d’aide à la décision pour la régulation des blocs opératoires.” In GISEH 2022 - 11e
Conférence Francophone en Gestion et Ingénierie des Systèmes Hospitaliers, Saint-Etienne, France.
[Link]
Robinson, Stewart. 2004. Simulation: The Practice of Model Development and Use. Chichester: Wiley.
Rosen, Roland, Georg von Wichert, George Lo, and Kurt D. Bettenhausen. 2015. “About The
Importance of Autonomy and Digital Twins for the Future of Manufacturing.” IFAC-

194
 References
PapersOnLine, 15th IFAC Symposium onInformation Control Problems
inManufacturing, 48 (3): 567–72. [Link]
Roshanaei, Vahid, Curtiss Luong, Dionne M. Aleman, and David R. Urbach. 2017a.
“Collaborative Operating Room Planning and Scheduling.” INFORMS Journal on
Computing 29 (3): 558–80.
———. 2017b. “Collaborative Operating Room Planning and Scheduling.” INFORMS Journal
on Computing 29 (3): 558–80. [Link]
Saadouli, Hadhemi, Badreddine Jerbi, Abdelaziz Dammak, Lotfi Masmoudi, and Abir Bouaziz.
2015. “A Stochastic Optimization and Simulation Approach for Scheduling Operating
Rooms and Recovery Beds in an Orthopedic Surgery Department.” Computers &
Industrial Engineering 80 (February): 72–79. [Link]
Samudra, M., E. Demeulemeester, and B. Cardoen. 2013. “A Closer View at the Patient Surgery
Planning and Scheduling Problem: A Literature Review.” Review of Business and Economic
Literature 58 (2): 115–40.
Samudra, Michael, Carla Van Riet, Erik Demeulemeester, Brecht Cardoen, Nancy
Vansteenkiste, and Frank E. Rademakers. 2016. “Scheduling Operating Rooms:
Achievements, Challenges and Pitfalls.” Journal of Scheduling 19 (5): 493–525.
[Link]
Saremi, Alireza, Payman Jula, Tarek ElMekkawy, and G. Gary Wang. 2013. “Appointment
Scheduling of Outpatient Surgical Services in a Multistage Operating Room
Department.” International Journal of Production Economics, Special Issue on Service Science,
141 (2): 646–58. [Link]
Schluse, Michael, and Juergen Rossmann. 2016. “From Simulation to Experimentable Digital
Twins: Simulation-Based Development and Operation of Complex Technical Systems.”
In 2016 IEEE International Symposium on Systems Engineering (ISSE), 1–6.
[Link]
Schoenfelder, Jan, Sebastian Kohl, Manuel Glaser, Sebastian McRae, Jens O. Brunner, and
Thomas Koperna. 2021. “Simulation-Based Evaluation of Operating Room
Management Policies.” BMC Health Services Research 21 (1): 271.
[Link]
Schroeder, Greyce N., Charles Steinmetz, Carlos E. Pereira, and Danubia B. Espindola. 2016.
“Digital Twin Data Modeling with AutomationML and a Communication Methodology
for Data Exchange.” IFAC-PapersOnLine, 4th IFAC Symposium on Telematics
Applications TA 2016, 49 (30): 12–17. [Link]
Schultz, Jamie, and David Claudio. 2014. “Variability Based Surgical Scheduling: A Simulation
Approach.” In Proceedings of the Winter Simulation Conference 2014, 1353–64.
[Link]
Semeraro, Concetta, Mario Lezoche, Hervé Panetto, and Michele Dassisti. 2021. “Digital Twin
Paradigm: A Systematic Literature Review.” Computers in Industry 130 (September):
103469. [Link]
Shao, Guodong, Simon Frechette, and Vijay Srinivasan. 2023. “An Analysis of the New ISO
23247 Series of Standards on Digital Twin Framework for Manufacturing.” In .
American Society of Mechanical Engineers Digital Collection.
[Link]
Shao, Guodong, and Moneer Helu. 2020. “Framework for a Digital Twin in Manufacturing:
Scope and Requirements.” Manufacturing Letters 24 (April): 105–7.
[Link]

195
References
Sokovic, M, D Pavletic, and K Kern Pipan. 2010. “Quality Improvement Methodologies –
PDCA Cycle, RADAR Matrix, DMAIC and DFSS.” Journal of Achievements in Materials
and Manufacturing Engineering 43 (1).
Song, Ying, and Yongkui Li. 2022. “Digital Twin Aided Healthcare Facility Management: A
Case Study of Shanghai Tongji Hospital,” March, 1145–55.
[Link]
Stuart, Kari, and Erhan Kozan. 2012. “Reactive Scheduling Model for the Operating Theatre.”
Flexible Services and Manufacturing Journal 24 (4): 400–421.
[Link]
Stuart, Kari, Erhan Kozan, Michael Sinnott, and James Collier. 2010. “An Innovative Robust
Reactive Surgery Assignment Model.” ASOR Bulletin 29 (3): 48–59.
Thorwarth, Michael, and Amr Arisha. 2009. “Application of Discrete-Event Simulation in
Health Care: A Review.” [Link]
Turhan, Aykut Melih, and Bilge Bilgen. 2017. “Mixed Integer Programming Based Heuristics
for the Patient Admission Scheduling Problem.” Computers & Operations Research 80
(April): 38–49. [Link]
Van Huele, C., and M. Vanhoucke. 2015. “Operating Theatre Modelling: Integrating Social
Measures.” Journal of Simulation 9 (2): 121–28. [Link]
Van Riet, Carla, and Erik Demeulemeester. 2015. “Trade-Offs in Operating Room Planning for
Electives and Emergencies: A Review.” Operations Research for Health Care, ORAHS 2014
- The 40th international conference of the EURO working group on Operational
Research Applied to Health Services, 7 (December): 52–69.
[Link]
Vancroonenburg, Wim, Pieter Smet, and Greet Vanden Berghe. 2015. “A Two-Phase Heuristic
Approach to Multi-Day Surgical Case Scheduling Considering Generalized Resource
Constraints.” Operations Research for Health Care, ORAHS 2014 - The 40th international
conference of the EURO working group on Operational Research Applied to Health
Services, 7 (December): 27–39. [Link]
VanDerHorn, Eric, and Sankaran Mahadevan. 2021. “Digital Twin: Generalization,
Characterization and Implementation.” Decision Support Systems 145 (June): 113524.
[Link]
Vázquez-Serrano, Jesús, Rodrigo Peimbert-García, and Leopoldo Cárdenas-Barrón. 2021.
“Discrete-Event Simulation Modeling in Healthcare: A Comprehensive Review.”
International Journal of Environmental Research and Public Health 18 (November): 12262.
[Link]
Wang, Lien, Erik Demeulemeester, Nancy Vansteenkiste, and Frank E. Rademakers. 2021.
“Operating Room Planning and Scheduling for Outpatients and Inpatients: A Review
and Future Research.” Operations Research for Health Care 31 (December): 100323.
[Link]
Wang, Shuo, Vahid Roshanaei, Dionne Aleman, and David Urbach. 2016. “A Discrete Event
Simulation Evaluation of Distributed Operating Room Scheduling.” IIE Transactions on
Healthcare Systems Engineering 6 (4): 236–45.
Wang, Xi Vincent, and Lihui Wang. 2019. “Digital Twin-Based WEEE Recycling, Recovery and
Remanufacturing in the Background of Industry 4.0.” International Journal of Production
Research 57 (12): 3892–3902. [Link]
Xiang, Feng, Zhi Zhang, Ying Zuo, and Fei Tao. 2019. “Digital Twin Driven Green Material
Optimal-Selection towards Sustainable Manufacturing.” Procedia CIRP, 52nd CIRP

196
 References
Conference on Manufacturing Systems (CMS), Ljubljana, Slovenia, June 12-14, 2019, 81
(January): 1290–94. [Link]
Xiang, Wei, Jiao Yin, and Gino Lim. 2015a. “A Short-Term Operating Room Surgery
Scheduling Problem Integrating Multiple Nurses Roster Constraints.” Artificial Intelligence
in Medicine 63 (2): 91–106. [Link]
———. 2015b. “An Ant Colony Optimization Approach for Solving an Operating Room
Surgery Scheduling Problem.” Computers & Industrial Engineering 85 (July): 335–45.
[Link]
Xie, Jiacheng, Xuewen Wang, Zhaojian Yang, and Shangqing Hao. 2019. “Virtual Monitoring
Method for Hydraulic Supports Based on Digital Twin Theory.” Mining Technology 128
(2): 77–87. [Link]
Yahia, Zakaria, Junichi Iijima, Nermine A Harraz, and Amr B Eltawil. 2017. “A Design and
Engineering Methodology for Organization-Based Simulation Model for Operating
Room Scheduling Problems.” SIMULATION 93 (5): 363–78.
[Link]
Zhang, Xiange. 2018. “Application of Discrete Event Simulation in Health Care: A Systematic
Review.” BMC Health Services Research 18 (1): 687. [Link]
3456-4.
Zhang, Zheng, and Xiaolan Xie. 2015. “Simulation-Based Optimization for Surgery
Appointment Scheduling of Multiple Operating Rooms.” IIE Transactions 47 (9): 998–
1012. [Link]
Zhou, J., and F. Dexter. 1998. “Method to Assist in the Scheduling of Add-on Surgical Cases--
Upper Prediction Bounds for Surgical Case Durations Based on the Log-Normal
Distribution.” Anesthesiology 89 (5): 1228–32. [Link]
199811000-00024.
Zhu, Shuwan, Wenjuan Fan, Shanlin Yang, Jun Pei, and Panos M. Pardalos. 2019. “Operating
Room Planning and Surgical Case Scheduling: A Review of Literature.” Journal of
Combinatorial Optimization 37 (3): 757–805. [Link]
6.
Zhu, Zexuan, Chao Liu, and Xun Xu. 2019. “Visualisation of the Digital Twin Data in
Manufacturing by Using Augmented Reality.” Procedia CIRP, 52nd CIRP Conference on
Manufacturing Systems (CMS), Ljubljana, Slovenia, June 12-14, 2019, 81 (January): 898–
903. [Link]
Zonderland, Maartje E., and Richard J. Boucherie. 2021. “A Survey of Literature Reviews on
Patient Planning and Scheduling in Healthcare.” In Handbook of Healthcare Logistics:
Bridging the Gap between Theory and Practice, edited by Maartje E. Zonderland, Richard J.
Boucherie, Erwin W. Hans, and Nikky Kortbeek, 17–23. International Series in
Operations Research & Management Science. Cham: Springer International Publishing.
[Link]

197
List of figures

LIST OF FIGURES

Figure 1 –Interactions between the surgical suite and the other services. ........................................... 2
Figure 2 - Example of a provisional operating schedule for a 4-OR surgical suite. ........................... 3
Figure 3 – The 4 decision levels of planning and scheduling activity................................................... 5
Figure 4 – Research Positioning ................................................................................................................. 6
Figure 5 - High level process mapping for the surgical suite ............................................................... 14
Figure 6 – Interactions between the surgical suite and external services ........................................... 16
Figure 7 – Patient pathway from surgeon consultation to being ready for the hospital admission
for elective inpatients and outpatients............................................................................................ 23
Figure 8 - Non-elective patient pathway in the emergency wards. ..................................................... 25
Figure 9 – Illustration of a patient pathway in the surgical suite. ........................................................ 26
Figure 10 – Surgical suite patient pathway with in-OR induction. ..................................................... 27
Figure 11 – Preoperative care patient pathway options........................................................................ 28
Figure 12 - The different types of material resources required in a surgical suite. ........................... 30
Figure 13 – Commented layout of a real-world surgical suite ............................................................. 31
Figure 14 – Surgical suite staff categories and the different occupations they consist of. .............. 32
Figure 15 – Brief description of the missions of the suite medical staff. ........................................... 33
Figure 16 - Brief description of the missions of the suite paramedical staff (1/2). .......................... 34
Figure 17 - Brief description of the missions of the suite paramedical staff (2/2). .......................... 35
Figure 18 – Representation of the surgical team members during a surgery. .................................. 36
Figure 19 – Description of our OR usage related KPIs. ...................................................................... 38
Figure 20 - Complete framework ............................................................................................................. 65
Figure 21 - Illustration of the steps of a standardized method to build to build a surgical suite digital
twin. ..................................................................................................................................................... 66
Figure 22 – Illustration of the difference between an aggregated process (top) and a detailed process
(bottom) for the preoperative care of a surgery requiring an LRA............................................ 71
Figure 23 – Illustration of the difference between the theoretical best situation (left) and an example
of real situation (right) of the human resource presence during the perioperative phase of a
patient requiring induction in the operating room. ...................................................................... 73
Figure 24 - Example of the Impact of the Anesthesiologist Being a Limited Resource ................. 76
Figure 25 - Which resources can be modeled as infinite resources?................................................... 77

198
 List of figures

Figure 26 - Description of the aggregated surgical suite process with flexible resource constraints:
timestamps, steps and required resources of the patient pathway. ............................................ 80
Figure 27 - Description of the detailed surgical suite process with strict resource constraints:
timestamps, steps and required resources of the patient pathway (perioperative and post-
operative phases) ............................................................................................................................... 82
Figure 28 - Description of the detailed surgical suite process with strict resource constraints:
timestamps, steps and required resources of the patient pathway (preoperative phase without
induction)............................................................................................................................................ 83
Figure 29 - Description of the detailed surgical suite process with strict resource constraints:
timestamps, steps and required resources of the patient pathway (preoperative phase for LRA
induction)............................................................................................................................................ 83
Figure 30 - Description of the detailed surgical suite process with strict resource constraints:
timestamps, steps and required resources of the patient pathway (preoperative phase with
ophthalmology sedation induction) ................................................................................................ 84
Figure 31 – Representation of our DT-DSS so far. .............................................................................. 90
Figure 32 - Detailed Patient Pathway Timeline ..................................................................................... 92
Figure 33 – Steps followed to correct and compute timestamps and durations ............................... 92
Figure 34 – Timestamps and Steps of an Aggregated Patient Pathway ............................................. 93
Figure 35 – Illustration of the surgical suite processes from the patient POV (top) and the operating
room POV (bottom) ......................................................................................................................... 94
Figure 36 – Illustration of incoherence timestamps in the light of operating room usage (OR POV).
.............................................................................................................................................................. 95
Figure 37 - Comparison of the Number of Timestamps after Correction from the Patient POV
(orange) and the OR POV (grey). ................................................................................................... 96
Figure 38 – We compute “Suite Entry Time” based on “Room Entry Time” and “Preoperative
Care Duration”. ...............................................................................................................................102
Figure 39 - We compute “PACU Exit Time” based on “PACU Entry Time” and “PACU
Monitoring Duration”. ...................................................................................................................102
Figure 40 – Illustration of cases where 𝑡𝑖 + 1– 𝑑𝑡𝑖, 𝑡𝑖 + 1 > 𝑡𝑖 − 1 is respected (top timeline) or
not respected (bottom timeline). ...................................................................................................103
Figure 41 – Illustration of cases where 𝑡𝑖 − 1 + 𝑑𝑡𝑖 − 1, 𝑡𝑖 < 𝑡𝑖 + 1 is respected (top timeline)
or not respected (bottom timeline). ..............................................................................................103
Figure 42 – We compute “Incision Time” based on “Suture Time” and “Procedure Duration”.
............................................................................................................................................................104
Figure 43 - We compute “Suture Time” based on “Room Exit Time” and “Reversal Duration”.
......................................................................................................................................................... 104h

199
List of figures

Figure 44 - We compute “PACU Entry Time” based on “PACU Exit” and “PACU Monitoring
Duration”. ........................................................................................................................................104
Figure 45 – We compute “Room Exit Time” based on “Suture Time”, “PACU Entry Time” and
“Move to PACU Duration” of the same patient, as well as “Room Entry Time” of the next
patient................................................................................................................................................105
Figure 46 - We compute “Room Exit Time” based on “Suture Time”, “PACU Entry Time” and
“Reversal Duration” of the same patient, as well as “Room Entry Time” of the next patient.
............................................................................................................................................................106
Figure 47 - We compute “Room Entry Time” based on “Suite Entry Time”, “Incision Time” and
“Setup Duration” of the same patient, as well as “Room Entry Time” of the previous patient.
............................................................................................................................................................107
Figure 48 - We compute “Room Entry Time” based on “Suite Entry Time”, “Incision Time” and
“Preoperative care Duration” of the same patient, as well as “Room Entry Time” of the
previous patient. ..............................................................................................................................108
Figure 49 – Number of timestamps recorded in the database: before correction (orange), after the
correction from the patient POV (yellow), after correction from the OR POV (grey), and after
computation of the missing performed timestamps using performed durations (blue). ......109
Figure 50 - Compute Timestamps and Durations for the Provisional schedule ............................111
Figure 51 – How do we chose the computation method for the durations of the provisional
schedule? ...........................................................................................................................................113
Figure 52 – Illustration of how the schedule evolves from the weekly staff meeting to the end of
the execution day. ............................................................................................................................122
Figure 53 - When and why do we perform non-elective case scheduling? ......................................123
Figure 54 - Illustration of The Three Non-Elective Scheduling Strategies Allowed in our DT-DSS.
............................................................................................................................................................128
Figure 55 - Illustration of manual non-elective scheduling (1/3). At the first disruption, the main
simulation resets, and the experimenter is automatically launched..........................................130
Figure 56 - Illustration of manual non-elective scheduling (2/3). The experimenter launches n
scenarios of m replications to test all the possible scheduling solutions for the non-elective.
The user chooses to implement one of the scheduling scenarios (yellow star). ....................130
Figure 57 - Illustration of manual non-elective scheduling (3/3). The scheduling scenarios chosen
by the user is implemented in the main simulation. The simulation restart from t=0 until the
next new disruption. .......................................................................................................................131
Figure 58 – Database Treatment Steps. ................................................................................................138
Figure 59 – Steps to Correct and Complete Timestamps and Durations Values from the Database.
............................................................................................................................................................138
Figure 60 –Provisional Master Surgery Schedule. ...............................................................................141

200
 List of figures

Figure 61 - Simulation of the Provisional Schedule Execution in a Deterministic Environment with

Aggregated Processes and Flexible Constraints..........................................................................142
Figure 62 – Retrospective Analysis (step #1): Performed Master Surgery Schedule. ....................143
Figure 63 - Retrospective Analysis (step #1): Performed Surgery Schedule. ..................................144
Figure 64 – Prospective Analysis (steps #1 and #2): Simulation of the Provisional Schedule
Execution in a deterministic environment with detailed processes and strict constraints on
resources. ..........................................................................................................................................153
Figure 65 - Prospective Analysis (Step #4): Utilization Box Plot. ....................................................157
Figure 66 - Prospective Analysis (Step #4): Provisional Schedule Execution with detailed pathway,
strict constraints, and stochastic durations, of the replication with the highest utilization
rate. ...................................................................................................................................................157
Figure 67 - Provisional Schedule Execution with detailed pathway, strict constraints, and stochastic
durations, of the replication with the lowest utilization rate. ................................................157
Figure 68 - Prospective Analysis (step #4). Example of a surgical schedule with the FF strategy.
............................................................................................................................................................160
Figure 69 - Prospective Analysis (step #4). Example of a surgical schedule with the WF strategy.
............................................................................................................................................................161
Figure 70 - Prospective Analysis (step #5). Example #1 of a surgical schedule with the FF strategy.
............................................................................................................................................................163
Figure 71 - Prospective Analysis (step #5). Example #1 of a surgical schedule with the WF strategy.
............................................................................................................................................................163
Figure 72 - Prospective Analysis (step #5). Example #2 of a surgical schedule with the FF strategy.
............................................................................................................................................................163
Figure 73 - Prospective Analysis (step #5). Example #2 of a surgical schedule with the WF strategy.
............................................................................................................................................................164
Figure 74 – Retrospective Analysis (step #2): First Fit Scheduling ..................................................165
Figure 75 – Retrospective Analysis (step #2): Best Fit Scheduling ..................................................166
Figure 76 – Surgical suite layout we used to create our digital twin. This figure was presented in
Chapter I of our manuscript. .........................................................................................................211
Figure 77 – 3D view of our surgical suite digital twin. .......................................................................211
Figure 78 – Screenshot from the second video. ..................................................................................212
Figure 79 - Dashboard example that can be made in our surgical suite digital twin. .....................215
Figure 80 – Bird-Eye View of the DT-DSS with input, output, parameters and tool description.
............................................................................................................................................................221
Figure 81 - Estimation of the number of replications using a graphical method. ..........................222

201
List of figures

Figure 82 – Patient Pathway Flowchart as Found in the DT-DSS ...................................................225

Figure 83 - Process #1: Escort Patient to its Preoperative Location ...............................................226
Figure 84 – Process #2: Preoperative Care (Part 2/2) .......................................................................228
Figure 85 – Process #3: Operative Care (Part 1/2) ............................................................................229
Figure 86 - Process #3: Operative Care (Part 2/2) .............................................................................230
Figure 87 – Process #4. Patient Recovery ............................................................................................231
Figure 88 – Process #5: Escort Patient to Exit ...................................................................................231
Figure 89 – Process #6: Room Cleaning ..............................................................................................231

202
 List of tables

LIST OF TABLES

Table 1 – Proposed definitions .................................................................................................................. 3

Table 2 – Proposed methodology .............................................................................................................. 7
Table 3 - Suggested contributions.............................................................................................................. 8
Table 4 - Main differences between inpatients and outpatients .......................................................... 18
Table 5 – Examples of categorization of non-elective patients found in the scientific literature.
Taken from (Van Riet and Demeulemeester 2015)) .................................................................... 19
Table 6 - Proposition of categories based on target for our study...................................................... 19
Table 7 - Brief description of the four main types of anesthesia provided in the surgical suite. ... 21
Table 8 – Example of dysfunctions coming from a lack of communication. ................................... 39
Table 9 - Review of Predictive and Reactive Disruption Management for the surgery scheduling
and the disruption management problems. Taken from (Kamran, Karimi, and Dellaert 2020).
The abbreviations stand for planning (P), scheduling (S), replanning (RP), rescheduling (RS),
patient booking strategy (PBP), DM (disruption management), reactive disruption
management (RDM), predictive and reactive disruption management (PRDM). ................... 53
Table 10 – Prospective analysis steps ...................................................................................................... 62
Table 11 - Retrospective analysis steps ................................................................................................... 64
Table 12 – Difference between the required and the available information in our study case....... 70
Table 13 – Description of the flexible and strict constraints on material resources ........................ 78
Table 14 – Description of the flexible and strict constraints on human resources .......................... 78
Table 15 – Description of the Steps of the Aggregated patient pathway .......................................... 81
Table 16 – Anesthesia Type Grouping ................................................................................................... 86
Table 17 – Patient Pathway Options based on Anesthesia Type. ....................................................... 86
Table 18 – Description of process type and resources for the performed and the provisional
schedule execution. ........................................................................................................................... 89
Table 19 – Who records the surgical suite processes’ timestamps? .................................................... 92
Table 20 - Example of coherent and incoherent timestamps. ............................................................. 94
Table 21 – Example of coherent and incoherent values two consecutive timestamps in the same
patient pathway .................................................................................................................................. 95
Table 22 – Number of timestamps before correction, after correction using patient POV, and after
correction using OR POV. .............................................................................................................. 97

203
List of tables

Table 23 - Percentage of timestamps before correction, after correction using patient POV, and
after correction using OR POV. ..................................................................................................... 97
Table 24 – Number and percentage of computed durations per duration type. .............................. 98
Table 25 – Proposition of grouping criteria. .......................................................................................... 99
Table 26 – Values of the WAPE for each duration type (column) and each estimation method
(row). The worse the WAPE is the more the colors tend to be red; the better the WAPE is
the more the colors tend to be green. ..........................................................................................100
Table 27 - Number of timestamps recorded in the database.............................................................109
Table 28 - Percentage of timestamps recorded in the database. .......................................................110
Table 29 – Example of Proposed Standardized Label for surgical procedures ..............................112
Table 30 – Computing cleanup duration. .............................................................................................116
Table 31 – Synthesis of the methods used to compute the deterministic durations for the
provisional schedule. .......................................................................................................................118
Table 32 - Synthesis of the methods used to compute the stochastic durations for the provisional
schedule. ...........................................................................................................................................119
Table 33 - When and why do we perform non-elective case scheduling? .......................................122
Table 34 – Description of the DT-DSS configuration for each time we must perform non-elective
case scheduling. ...............................................................................................................................123
Table 35 - Description of non-elective modeling and scheduling parameters for the DT-DSS
configuration for each time we must perform non-elective case scheduling. ........................125
Table 36 – Description of the parameters options to create ANE arrivals scenarios....................127
Table 37 – Brief presentation of the facilities of our partners. ..........................................................137
Table 38 – Brief presentation of the OR software database provided by our partners. ................138
Table 39 – Schedule constraints that must be respected in to be able to reach our study objectives
............................................................................................................................................................139
Table 40 – Overview of the Provisional Schedule and the Performed Schedule. ..........................140
Table 41 - KPIs Related to the Operating Room Utilization. ...........................................................142
Table 42 - KPIs Related to the Patient Waiting Time. .......................................................................143
Table 43 - Description of the Urgent Cases in the Performed Schedule. ........................................144
Table 44 - Study Case Description: OR and Surgeons Allowed for the Urgent Cases..................145
Table 45 - Description of the Urgent Cases Durations in the Performed Schedule (in minutes).
............................................................................................................................................................145
Table 46 - KPI Related to the Operating Room Utilization. .............................................................145
Table 47 - KPI Related to the Patient Waiting Time. .........................................................................146

204
 List of tables

Table 48 - Comparison of KPIs for both schedules. ..........................................................................146

Table 49 - Experimentations performed by the DT-DSS ..................................................................152
Table 50 – Prospective Analysis (steps #1 and #2): Global patient waiting time. .........................154
Table 51 - Prospective Analysis (steps #1 and #2): Patient waiting time for material resources.154
Table 52 - Prospective Analysis (steps #1 and #2): Patient waiting time for human resources. .154
Table 53 - Prospective Analysis (Step #1 and #2): KPI Related to the Operating Room Utilization.
............................................................................................................................................................155
Table 54 - Prospective Analysis (Step #3): Summary of the KPI values across all the replications.
The PWT are in minutes and the PWT results from steps #1 and #2 are the average value.
............................................................................................................................................................156
Table 55 –Prospective Analysis (step #4): Resilience FF VS BF ......................................................159
Table 56 - Prospective Analysis (step #4). Comparison of the performance of schedule execution
in a determinist environment with schedule execution disrupted by non-elective arrivals. .160
Table 57 –Prospective Analysis (step #5): All .....................................................................................162
Table 58 - Retrospective Analysis (step #1 & 2): Comparison of the scheduling strategy ...........165
Table 59 – Retrospective Analysis (step #2). KPIs Related to the Operating Room Utilization.
............................................................................................................................................................166
Table 60 – How does the DT-DSS can be used during the weekly scheduling meeting in order to
facilitate schedule execution? .........................................................................................................168
Table 61 - Human resources modeled in our DT during the provisional schedule execution. ....177
Table 62 - Human resources modeled in our DT during the provisional schedule execution. ....177
Table 63 - Case Study Description: Pathway and Resources. ............................................................207
Table 64 - Case Study Description: Provisional and Performed Arrival Time, OR ID, Case Rank
and Maximum Rank. .......................................................................................................................208
Table 65 - Description and Possible Values of the DT-DSS Parameters ........................................223

205
Appendices

APPENDICES

206
 Appendices

1. Appendix #1 – description of the study case

Table 63 - Case Study Description: Pathway and Resources.

Case Admissi Surge Urgency Preoperative OR Surgeon Anesthesiolog

ID on ry Level Care Type Induction ID ist ID
1 OUT 0 Elective WA 1 60 12
2 IN 1 Elective WA 1 60 12
3 IN 2 Elective WA 1 60 12
4 IN 3 Elective WA 1 60 12
5 OUT 4 Elective WA 1 40 4
6 IN 5 Elective WA 1 40 4
7 OUT 6 Elective WA 0 40 4
8 IN 7 Urgent WA 0 18 4
9 OUT 8 Elective WA 1 18 12
10 IN 9 Elective WA 1 18 12
11 IN 10 Elective WA 1 18 12
12 OUT 11 Elective WA 1 48 4
13 IN 12 Elective LRA 0 48 4
14 OUT 13 Elective LRA 0 48 4
15 OUT 14 Elective LRA 0 48 4
16 OUT 15 Urgent LRA 0 48 4
17 OUT 16 Elective OIP 1 34 5
18 OUT 16 Elective OIP 1 34 5
19 OUT 16 Elective OIP 1 34 5
20 OUT 16 Elective OIP 1 34 5
21 OUT 16 Elective OIP 1 34 5
22 OUT 16 Elective OIP 1 34 5
23 OUT 17 Elective OIP 1 32 4
24 OUT 18 Elective OIP 1 32 4
25 OUT 19 Elective OIP 1 32 4
26 OUT 20 Elective WA 1 32 4
27 OUT 20 Elective OIP 1 32 4
28 OUT 20 Elective OIP 1 32 4
29 OUT 16 Elective OIP 1 34 5
30 OUT 16 Elective OIP 1 34 5
31 OUT 16 Elective OIP 1 34 5
32 OUT 16 Elective OIP 1 34 5
33 OUT 16 Elective OIP 1 34 5
34 OUT 21 Elective OIP 1 37 42
35 OUT 22 Elective OIP 1 37 42
36 OUT 22 Elective WA 0 37 42
37 OUT 23 Elective OIP 1 37 42
38 OUT 24 Elective WA 0 37 42
39 OUT 25 Elective WA 1 23 3
40 OUT 26 Elective WA 1 23 3
41 OUT 25 Elective WA 1 23 3

207
Appendices

42 OUT 27 Elective WA 1 23 3
43 IN 26 Elective WA 1 23 3
44 OUT 27 Elective WA 1 23 3
45 IN 25 Elective WA 1 23 3
46 OUT 28 Elective WA 1 22 5
47 OUT 28 Elective WA 1 22 5
48 OUT 29 Elective WA 1 22 5
49 OUT 28 Elective WA 1 22 5
50 OUT 28 Elective WA 1 22 5
51 OUT 29 Elective WA 1 22 5
52 OUT 28 Elective WA 1 22 5
53 OUT 28 Elective WA 1 22 5

Table 64 - Case Study Description: Provisional and Performed Arrival Time, OR ID, Case Rank and
Maximum Rank.

Performed Provisional Performe Provisiona

Ca Perfor Provisi Perform Provisio
Suite Suite d l
se med onal ed Case nal Case
Arrival Arrival Maximum Maximum
ID OR ID OR ID Rank Rank
Time Time Rank Rank
1 7:49 AM 7:59 AM 1 1 1 8 1 7
2 8:43 AM 8:59 AM 1 1 2 8 2 7
3 9:42 AM 9:53 AM 1 1 3 8 3 7
4 10:58 AM 11:38 AM 1 1 4 8 4 7
5 1:25 PM 1:59 PM 1 1 5 8 5 7
6 3:59 PM 4:59 PM 1 1 6 8 6 7
7 5:53 PM 5:57 PM 1 1 7 8 7 7
8 7:08 PM 7:29 PM 1 0 8 8 2 2
9 7:48 AM 7:59 AM 2 2 1 3 1 3
10 9:29 AM 9:41 AM 2 2 2 3 2 3
11 11:32 AM 12:03 PM 2 2 3 3 3 3
12 1:30 PM 1:59 PM 5 5 1 5 1 4
13 1:54 PM 2:40 PM 5 5 2 5 2 4
14 4:28 PM 5:06 PM 5 5 3 5 3 4
15 4:28 PM 6:14 PM 5 5 4 5 4 4
16 5:15 PM 5:29 PM 5 0 5 5 1 2
17 7:37 AM 7:59 AM 6 6 1 12 1 17
18 8:14 AM 8:39 AM 6 6 2 12 3 17
19 8:58 AM 9:19 AM 6 6 3 12 5 17
20 9:58 AM 11:05 AM 6 6 4 12 7 17
21 10:36 AM 11:45 AM 6 6 5 12 9 17
22 11:41 AM 12:25 PM 6 6 6 12 11 17
23 2:06 PM 1:59 PM 6 6 7 12 12 17
24 2:08 PM 3:28 PM 6 6 8 12 13 17
25 3:48 PM 3:56 PM 6 6 9 12 14 17
26 4:09 PM 5:00 PM 6 6 10 12 15 17

208
 Appendices

27 5:15 PM 5:30 PM 6 6 11 12 16 17
28 5:41 PM 5:56 PM 6 6 12 12 17 17
29 7:41 AM 8:19 AM 7 6 1 10 2 17
30 8:33 AM 8:59 AM 7 6 2 10 4 17
31 9:26 AM 9:39 AM 7 6 3 10 6 17
32 10:08 AM 11:25 AM 7 6 4 10 8 17
33 11:27 AM 12:05 PM 7 6 5 10 10 17
34 1:30 PM 1:59 PM 7 7 6 10 1 5
35 2:08 PM 2:51 PM 7 7 7 10 2 5
36 3:37 PM 3:34 PM 7 7 8 10 3 5
37 3:37 PM 4:20 PM 7 7 9 10 4 5
38 4:24 PM 4:57 PM 7 7 10 10 5 5
39 7:43 AM 7:59 AM 8 8 1 15 1 15
40 8:21 AM 8:34 AM 8 8 2 15 2 15
41 8:54 AM 8:59 AM 8 8 3 15 3 15
42 9:07 AM 9:29 AM 8 8 4 15 4 15
43 9:43 AM 10:15 AM 8 8 5 15 6 15
44 9:55 AM 10:13 AM 8 8 6 15 5 15
45 10:12 AM 10:31 AM 8 8 7 15 7 15
46 1:32 PM 1:59 PM 8 8 8 15 8 15
47 2:04 PM 2:22 PM 8 8 9 15 9 15
48 2:55 PM 2:54 PM 8 8 10 15 10 15
49 3:37 PM 3:28 PM 8 8 11 15 11 15
50 3:37 PM 3:58 PM 8 8 12 15 12 15
51 4:24 PM 4:25 PM 8 8 13 15 13 15
52 4:44 PM 4:52 PM 8 8 14 15 14 15
53 5:05 PM 5:11 PM 8 8 15 15 15 15

209
Appendices

2. Appendix #2 – visual illustration of our surgical suite digital twin

2.1. Introduction
This appendix clarifies both part II and part III of our manuscript. Part II describes how we
have built our surgical suite digital twin (SS-DT), and part III discusses how we have used this
digital twin as a proof of concept for our prospective and retrospective analysis of the operating
schedule execution. In this annex, we provide videos and screenshots of our digital twin
decision support system. We hope this will help the reader to better visualize what our tool looks
like and how it can be used.
The first video is a slow recording of a schedule execution in our SS-DT 3D view. It displays
the different virtual rooms, patients and resources modeled in the SS-DT; one can see the patients
moving around the surgical suite, the human resources caring for them, and the material resources
being used. The link for the video is the following:
[Link]
The second video displays a fast recording of a schedule execution in our SS-DT 3D view
and the real-time computation of a part of the dashboard. This video highlights how the KPI
are computed while the simulation is running. This could be especially useful for an online
implementation of the SS-DT. This link for the video is the following:
[Link]
Finally, we describe the screenshot of an example of a dashboard that can be found in the
simulation model.

2.2. Illustration of The Schedule Execution in our Surgical Suite Digital Twin

2.2.1. Brief description of the 3D Model

For this research, we have worked with a specific surgical suite (Hôpital Privé de La Baie for the
French group Vivalto Santé). Thus, we have based our digital twin on the real layout of their service.
In Figure 76, we display the real-world surgical layout we have used to create our digital twin, and
in Figure 77, we display the 3D view of our model. In our model, we have only represented the
parts of the real-world surgical suite related to our study (the operating rooms, the patient waiting
area, etc.).
For an easier visualization, each operating room (OR) has a specific color, and patients’ shirts
match their assigned OR. Each staff has a specific shirt color too: green for surgeons, grey for
anesthesiologists, clear blue for the OR nurses, purple for the anesthesiologist nurses, yellow for
the PACU nurses and the OIP nurses, dark blue for the nurse assistants, and white for the stretcher-
bearers. This is particularly helpful when we perform visual verification to ensure that the patient
pathway is properly executed.

210
 Appendices

Note that visual changes can be done to make the model more user-friendly. This includes adding
walls, masking the staff that is not involved in the schedule execution, changing the appearance of
the patients depending on how long they have waited or on their urgency level, etc.

Figure 76 – Surgical suite layout we used to create our digital twin. This figure was presented in Chapter I
of our manuscript.

Figure 77 – 3D view of our surgical suite digital twin.

211
Appendices

2.2.2. First video: illustration of the patient pathway during schedule execution.
In this video, we display the 3D view of our surgical suite digital twin during the execution of an
operating schedule. We display an overview of the suite (00:00 and 00:46), the suite entrance/exit
(00:12), the patient waiting area (00:22), the operating rooms (00:30), and the PACU (00:38).
In the videos, the processes are slightly different than the ones presented in the manuscript as we
also consider the anesthesiologist nurse, the stretcher-bearers, and the stretchers. However the
three phases remain the same: preoperative, perioperative, and post-operative. We discuss the
construction of the surgical suite processes we have modeled in our SS-DT in Chapter IV.

2.2.3. Second video: KPI computation during schedule execution.

In this video, we display parts of the dashboard side to side with the 3D view to highlight how the
indicators and the schedules are updated as the schedule execution goes. Figure 78 is a screenshot
taken from that video. The top table (#1) computes the KPI related to OR usage and the Gantt
diagram below describe the performed schedule of each OR (#2), and the provisional master
surgery schedule (#3).
We discuss the KPIs from table #1 in Chapter I. Other examples of the surgical schedule (#2) and
of the master surgery schedule (#3) are displayed in Chapters VII and VIII.

3
Figure 78 – Screenshot from the second video.

2.3. Another dashboard example

In Figure 79, we display an example of dashboard that can be built in our surgical suite digital twin.
We do not focus on the values displayed; this is to illustrate what kind of information a user can
have access while running a schedule execution in our SS-DT.
We differentiate between three types of elements: parameters (green), tables (blue), and graphs
(orange). We briefly describe each of them below.

212
 Appendices

2.3.1. Parameters (green)

We use parameters to configurate the environment in which the schedule is executed. All these
parameters are available in the experimenter; we use them to create the different experimentations
needed to apply our prospective and retrospective analysis.
Parameters in table #1 configurate the number of beds available for the waiting area, the PACU,
the loco-regional anesthesia, and the ophthalmologic sedation. In our study case, we set their value
so that they would not create unreasonable patient waiting times.
Parameters in table #2 are used in Chapter VIII to configurate the DT environment to perform
the experimentations required to apply our prospective and retrospective analysis – which are
describe in Chapter III. They can also be used to configurate the DT environment in case of OR
manager training.
We used the following color code:
• Black: Training environment modalities (methodology in Chapter III and theoretical
application in Chapter VI).
• Blue: Uncertainties on duration variability (Chapter V).
• Orange: Choice of either provisional or performed schedule.
• Green: Resource constraints and process modeling (Chapter IV).
• Yellow: Non-elective cases arrival and scheduling strategies (Chapter VI).

2.3.2. Tables (blue)

Table #1 gives some overall information about the current schedule being executed.
Table #2 describes KPI related to OR usage. Each row corresponds to a KPI (either a rate or a
value in minute), and each column corresponds to a OR, the entire surgical suite, or the waiting
line. We did not discuss the use of the waiting line in the manuscript: it can be used to visualize the
KPIs related to the cases that have arrived during schedule execution and which were not scheduled
in any OR.
Table #3 displays KPI related to the patient pathway and the patient waiting time (PWT). Each
row corresponds to a KPI (total PWT, PWT for all or specific human resources, PWT for all or
specific material resources, duration of each patient step). The KPI are all displayed in minutes.
The columns indicate the minimum, maximum, and average values of these KPI across all the
patient pathway executed, as well as the sum of the values for all the patients of the day.
We discuss the choice and definition of most KPIs from table #2 and table #3 in Chapter I, and
we use them for analyzing our study cases in Chapters VII and VIII.

2.3.3. Graphs (orange)

The graphs in #1 display (from left to right) information related to the OR throughput, the patient
stay time in the OR, and the patient waiting time for each OR.
Graph #2 describes the patient pathway steps for each patient; each row corresponds to a patient.

213
Appendices

Graph #3 does the same for the OR; each row corresponds to an OR.
Graph #4 is the master surgery schedule: it describes the planned shifts for each OR; each row
corresponds to an OR.
We use graphs #3 and #4 in the Part III of our manuscript (Chapters VII and VIII) for the study
case analysis.

214
 Appendices

Figure 79 - Dashboard example that can be made in our surgical suite digital twin.

215
Appendices

3. Appendix #3 – description of publications

In this document, we a list of our publications and their respective abstract.

3.1. List of publications

3.1.1. Journal
Rifi, Leah, Franck Fontanili, Maria Di Mascolo, and Cléa Martinez. 2022. “Framework for a
Retrospective Analysis of Operating Room Schedule Execution.” International Journal of
Healthcare Technology and Management 19 (1): 37.
[Link]

3.1.2. Conferences with proceedings

Rifi, Leah, Franck Fontanili, Cléa Martinez, Maria Di Mascolo, and Virginie Fortineau. 2023. “A
Simulation-Based Approach for Assessing the Impact of Uncertainty on Patient Waiting
Time in the Operating Room.” In Proceedings of the Winter Simulation Conference, 1057–68.
WSC ’22. Singapore, Singapore: IEEE Press.
Rifi, Leah, Clea Martinez, Maria Di Mascolo, and Franck Fontanili. 2022. “Proposition d’un outil
d’aide à la décision pour la régulation des blocs opératoires.” In GISEH 2022 - 11e Conférence
Francophone en Gestion et Ingénierie des Systèmes Hospitaliers, Saint-Etienne, France. [Link]
[Link]/hal-03763576.
Rifi, Leah, Franck Fontanili, and Michel Jeanney. 2020. “Proposition d’une Démarche Outillée
d’analyse Rétrospective Du Déroulement Du Programme Au Bloc Opératoire : Application
à La Régulation.” In GISEH 2020 - 10ème Conférence Francophone En Gestion et Ingénierie Des
Systèmes Hospitaliers. Valenciennes, France. [Link]

3.1.3. Book Chapter

Abdoune, Farah, Leah Rifi, Franck Fontanili, and Olivier Cardin. 2023. “Handling Uncertainties
with and Within Digital Twins.” In Service Oriented, Holonic and Multi-Agent Manufacturing
Systems for Industry of the Future, edited by Theodor Borangiu, Damien Trentesaux, and Paulo
Leitão, 118–29. Studies in Computational Intelligence. Cham: Springer International
Publishing. [Link] This was first presented in
the 2022 SOHOMA workshop.
Rifi, Leah, Franck Fontanili, and Michel Jeanney. 2022. “A Tool-Based Approach to Analyze
Operating Room Schedule Execution: Application to Online Management.” In Healthcare
Systems, 179–93. John Wiley & Sons, Ltd. [Link]
Rifi, Leah, Franck Fontanili, and Michel Jeanney. 2022. “Démarche outillée d'analyse du
déroulement du programme au bloc opératoire : application à la régulation.” In Systèmes de
Soins et de Santé : défis d’aujourd’hui et opportunités de demain, pp191-204. ISTE Group, ISBN
9781784068677. [Link]

216
 Appendices

3.2. Summary

3.2.1. Journal
2022 - Framework for a Retrospective Analysis of Operating Room Schedule Execution.
The execution of an operating room schedule is constantly disrupted, which can decrease the
initially targeted performance. Online operational management (OnOM), which oversees daily
activity, can reduce the deviations caused by disruptions between the initial schedule and the
performed schedule. To support this process and encourage continuous improvement, we suggest
a framework for analysing schedule execution in retrospect. The objectives are twofold: 1) to
identify deviations and determine their root causes; and 2) to assess the relevance of the decisions
made to reduce these deviations. This approach relies on a logbook to gather qualitative data on
disruptions, and a dashboard to objectify the situation with computed indicators. We present an
example of a schedule execution analysis in an anonymised French General Hospital.

3.2.2. Conferences with proceedings

2022 - A simulation-based approach for assessing the impact of uncertainty on patient
waiting time in the operating room.
Demand for surgical care is rising worldwide, making the organization of the operating room (OR)
a topic of strong interest. During the last two decades, the number of papers on methods for OR
planning and scheduling under uncertainty has increased significantly. However, most hospitals
neglect this aspect, and use deterministic approaches to schedule their surgical interventions. This
leads us to the following research question: "How can discrete-event simulation help assess the
impact of uncertainty on patient waiting time in the OR?" To answer this question, we suggest a 3-
step methodology: (1) building the deterministic model of the studied OR, (2) implementing
uncertainties on activity durations, patient arrival times and patient care requirements, and (3)
experimenting with different uncertainty-related scenarios and analyzing the results. We have
applied this methodology to a use-case inspired from our partner's OR: Hôpital Privé de La Baie,
from the Vivalto Santé French health group.

2022 - Proposition d’un outil d’aide à la décision pour la régulation des blocs opératoires.
La performance d’un hôpital est fortement liée à la performance de son bloc opératoire. La
communauté scientifique s’intéresse particulièrement à la construction du programme opératoire
(PO). L’étape d’exécution du programme est cependant moins étudiée. Durant cette étape, la
qualité du PO peut être dégradée par des perturbations aléatoires (ex : arrivée d’urgences). Notre
question de recherche est la suivante : « Comment gérer les perturbations aléatoires durant le
déroulement du programme opératoire afin de maintenir le niveau de performance visé ? ». Pour y
répondre, nous proposons de construire un jumeau numérique asynchrone du bloc opératoire et
de l’utiliser comme outil d’aide à la décision pour la régulation. Cet outil permettra d’étudier a
posteriori les dysfonctionnements d’une journée opératoire, de tester l’impact de modifications du

217
Appendices

PO en cours d’exécution sur la performance du BO, et d’étudier la robustesse d’un PO

prévisionnel. Nous proposons une démarche générique pour construire un modèle de simulation
permettant de simuler l’exécution d’un programme opératoire prévisionnel ou réalisé dans un
environnement déterministe ou stochastique. Nous avons appliqué cette démarche dans le bloc
opératoire de l’Hôpital Privé de La Baie (Avranches, Normandie) du groupe Vivalto Santé.

2020 - Proposition d’une démarche outillée d’analyse rétrospective du déroulement du

programme au bloc opératoire : application à la régulation.
Au bloc opératoire, la réalisation du programme prévisionnel est ponctuée de perturbations
aléatoires qui peuvent dégrader notablement la performance initialement visée. La régulation a pour
mission de piloter l'activité du bloc au quotidien et de réduire les écarts de performance entre le
prévisionnel et le réalisé. Pour y parvenir, nous proposons une démarche outillée d'amélioration
continue basée sur une analyse rétrospective du déroulement du programme opératoire. Cette
démarche a pour but (1) de mettre en évidence les écarts entre le prévisionnel et le réalisé et de
déterminer leur(s) cause(s) racine(s) et (2) d'évaluer la pertinence des décisions prises par la
régulation pour corriger les écarts. Elle se base sur deux outils : un journal de bord qui permet le
recueil de données qualitatives sur la gestion des anomalies et un tableau de bord dont les
indicateurs permettent d'objectiver la situation. La méthodologie de conception de cette démarche
s'appuie d'abord sur une observation approfondie du terrain. Ensuite, un cycle itératif en trois
phases : concevoir, construire et tester permet de construire progressivement la démarche et ses
deux outils. Un exemple de l'analyse du déroulement du programme opératoire dans une salle
interventionnelle du CH de Narbonne illustrera l'applicabilité de cette démarche outillée.

3.2.3. Book Chapter

2022 - Handling uncertainties with and within digital twins.
The Digital Twin (DT) is often used in environments characterized by uncertainty and complexity,
where operating conditions are prone to variability based on external and internal factors. Thus,
the literature about DT emphasizes the importance, limitations, and absence of uncertainty
quantification. However, there is no explicit review discussing uncertainty in complex systems and
within the digital twin model. Such an explicit review could improve the conception, construction,
and utilization of DT in environments that are both dynamic and stochastic. Thus, this article aims
to (1) describe how a DT can help manage uncertainties in a dynamic system, and (2) explain how
DT should deal with uncertainties inside the model.

2022 - Démarche outillée d'analyse du déroulement du programme au bloc opératoire :

application à la régulation.
Dans ce chapitre, les auteurs se concentrent sur la régulation, une fonction généralement assumée
par un.e infirmier.ère confirmé.e ou un médecin expérimenté. Les auteurs présentent l'analyse du
croisement des informations entre les déviations attribuables, ou potentiellement attribuables, à la

218
 Appendices

gestion en ligne et les informations sur les perturbations décrites dans le journal de bord. Ils
détaillent la méthodologie qui leur permet, par itérations successives, de générer leur démarche
outillée. Les auteurs détaillent également les outils sur lesquels cette démarche s'appuie. Enfin, ils
illustrent son applicabilité à travers un exemple concret de gestion en ligne au Centre Hospitalier
de Narbonne.

2022 - A Tool-based Approach to Analyze Operating Room Schedule Execution:

Application to Online Management.
In this chapter, the authors define operating room (OR) as the surgical room where staff perform
surgery on patients, and operating suite as the surgical department that houses all the ORs. They
focus on online management, a function generally performed by an experienced paramedic or
medical officer. The authors present the analysis of the intersection of information between
deviations attributable, or potentially attributable, to online management and information on
disruptions described in the Logbook. They detail the methodology which allows them, through
successive iterations, to generate their tool-based approach. The authors also detail the tools upon
which it is based. Finally, they illustrate its applicability through a concrete example of online
management at the Centre Hospitalier de Narbonne.

219
Appendices

4. Appendix #4 – description of the dt-dss

4.1. Introduction
In this appendix, we provide a comprehensive overview of the DT-DSS, including its input, output,
parameters, and tools. Subsection 1 is dedicated to the description of the parameters, while
subsection 2 is for the surgical suite processes.
Figure 80 offers a bird's-eye view of the DT-DSS. The input data is a relational database made of
three tables; it describes the master surgery schedule as well as each case attributes and durations.
The output is a dashboard that consists in performance KPIs (i.e. OR overtime and OR
utilization), patient waiting time indicators (i.e. for resources for or material) and of a Gantt Chart
Diagram of the state of each OR (idle time, waiting for resources, setup, procedure, etc.).The
parameters are the following: the initial schedule type, the process type, the constraints on
resources, the duration type, whether to keep or not the NE cases of the initial schedule, whether
to add or not NE arrivals to the Initial schedule, and the number of replications. The DT-DSS is
built using a modeling and simulation tool: Flexsim Healthcare ®. It allows to model material
resources, human resources, and surgical suite processes, and to simulate schedule execution.

220
 Appendices

Figure 80 – Bird-Eye View of the DT-DSS with input, output, parameters and tool description.

221
Appendices

4.2. Parameters
In Chapter VIII, we describe the 7 different parameters we use to configurate the DT-DSS: the
initial schedule type, the process type, the constraints on resources, the duration type, whether to
keep or not the NE cases of the initial schedule, whether to add or not NE arrivals to the Initial
schedule, and the number of replications. Below, we propose a justification as to why the
replication number can be set to either 1 or 30.
Due to the stochastic nature of the simulated environment, it is of great importance to run a certain
number of replications in order to acquire a confidence interval with respect to the obtained results.
The number of replications was estimated using the graphical method of Figure 81, in which a
value of ~30 replications would result in a confidence interval deviation of the cumulative mean
average of %𝑜𝑣𝑒𝑟𝑡𝑖𝑚𝑒 of less than 7.5%. The focus on the %𝑜𝑣𝑒𝑟𝑡𝑖𝑚𝑒 was due to the fact of it
being the most sensitive KPI to parameter uncertainty (see Chapter 2 on uncertainty duration
modelling).

Figure 81 - Estimation of the number of replications using a graphical method.

222
 Appendices

Table 65 - Description and Possible Values of the DT-DSS Parameters

# Name Description Possible Values

The initial schedule type can be either a provisional schedule or a performed schedule. The provisional
Initial Schedule schedule is a staff-validated schedule built along the strategic, tactical, and offline operational scheduling and Provisional
1
Type planning decisions, while the performed schedule is the description of how the schedule was really executed Performed
during the online operational step (real-life surgery day).

The process can be modeled as either an aggregated or a detailed process (see Chapter IV). The aggregated Aggregated
2 Process Type process is based on the available timestamped data in the OR software, while the detailed process is a model
based on the available timestamped data in the OR software and our on-site observations. Detailed

The constraints on resources can either be flexible or strict (see Chapter IV). Strict resource constraints are Flexible
Constraints on
3 the (possibly incomplete) constraints described by the OR database timestamps, while the flexible resource
Resources Strict
constraints are the constraints on the operating room only.

The simulated durations can either be determinist or stochastic (see Chapter V). In our mode, a determinist Determinist
4 Duration Type duration is a fixed value known in advance while a stochastic duration in a value extracted from a statistical or
an empirical law based on historical values. Stochastic

Keep NE Cases of Yes

5 The simulated schedule can either contain the NE cases present in the initial schedule or not.
Initial Schedule No

Add NE Arrivals The simulated schedule can either contain additional NE cases that were not present in the initial schedule or Yes
6
to Initial Schedule not. No

7 # Replications The number of replications is set to 1 when the scenario is determinist and to 30 when the scenario is stochastic. 1 or 30

223
Appendices

4.3. Description of the patient pathway

Once a patient is admitted in the surgical suite, it follows the flowchart proposed in the Figure 82.
Each box within the flowchart represents a process, while diamonds denote intersections.
To enhance readability, the color inside the box indicates the type of constraints on resources
(blue for flexible, yellow for strict), while the color of the box's contour indicates the pathway
type (blue for detailed, yellow for aggregated, green for both). It's noteworthy that certain
processes remain identical whether they are aggregated (based on the database timestamps) or
detailed (based on the database timestamps plus observations).
The outcome of the intersections "Pathway Type?" and "Resource Type?" is determined by the
model parameters (respectively “process type” and “Constraints on Resources”).
Processes #1 to #6 represent the patient pathway used for the prospective analysis (detailed
processes with strict constraints on resources) and processes #7 to #11 represent the patient
pathway mostly used for the retrospective analysis (aggregated processes with flexible constraints
on resources). The rest of the processes are either not used or are used during the training
processes.
In the subsequent subsection, we propose a brief presentation of modeling processes #1 to #6
with Flexsim Healthcare ® software – in other words: patient flows. Note that the processes #7
to #11 are similar but simpler: they only require operating rooms as resources and have less detailed
processes. Thus, we provide screenshots of the flowcharts we modeled in Flexsim healthcare. The
color code is the following:
• Light Orange: resources that the patient can acquire (e.g. OR, surgeon, nurse…).
• Green: actions to change the state of the OR used by the patient. Examples of OR state
are: “idle”, “setup”, “procedure”. The evolution of the OR state is then displayed in the
dashboard’ Gantt Diagram.
• Orange: actions to change the state of the patient. Examples of patient’ states are: “idle”,
“waiting for OR”, “waiting for surgeon” or “procedure”. This especially allows us to
compute the KPI related to patient waiting time.
• Deep blue: physical action in the surgical suite. Examples of physical actions are: “setup
before incision”, “procedure” and “reversal”.
• Red: actions that support the automatic scheduling of non-elective cases in the ongoing
schedule. We will not develop how they work together in this manuscript.
• Yellow: actions to send notifications to the user in the Flexsim console. An example of
notification can be “At t = 9:30, patient #3 entered the operating room.”.
• Light Blue: other actions that make up the patient pathway.

224
 Appendices

Flexible + Start
Flexible Strict
Strict

Flexible Strict
Detailed + Resource Type?
Detailed Aggregated
Aggregated #7 #1
Escort to Escort to
preoperative preoperative
care location care location

#2
Preoperative Aggregated Preoperative
Detailed Pathway Type? Pathway Type? Detailed
Care Care
#8
Preoperatory
Care

#9 #3
Operative Care Flexible Resource Type? Strict Operative Care

#10 #11 #5 #4
Aggregated Aggregated
Patient Patient
Pathway Type? Escort to Exit Escort to Exit Pathway Type?
Recovery Recovery

Detailed
Detailed
#6
Detailed
Room Cleaning End Room Cleaning Detailed

Figure 82 – Patient Pathway Flowchart as Found in the DT-DSS

225
FlexSim HC Patient Flow (process) #1: Escort Patient to its Preoperative Location
(Figure 83). This first process starts with the arrival of the patient in the surgical suite and ends
with the patient arriving at the preoperative location. The patient can be canceled on arrival
based on its own attributes (e.g. the patient has a rash that has gone unnoticed), or based on a
cancellation percentage that is defined by the user. In this study we have used neither.

Figure 83 - Process #1: Escort Patient to its Preoperative Location

A note on this model: acquiring resources. We highlighted with a red circle the action
of acquiring the stretcher-bearer (“brancardier” in French) and the stretcher resource
itself. We did the same in green with the preoperative location. This is how Flexsim
allows us to model assigning a resource to a patient. When the patient needs to acquire
both a human resource and a material resource at the same time, we have decided to
model it as the following: first we acquire the resources, then we acquire the location.
This allows us to (1) model the patient waiting time for the human resource and for the
material resource, (2) block a human resource rather than a material resource in case
both are not available at the same time, which is what happens in real-life most often.
A note on this model: modeling walking time. We highlighted with a purple circle
the actions “split” and “join”. After the “split” action, both downstream activities will

226
 Appendices
start at the same time. The activities after the “join” action will only start once both
downstream activities after the “split” action are done.

FlexSim HC Patient Flow (process) #2: Preoperative Care (Figure 84). We display the
flowchart of the preoperative care in Figure x and Figure x. The red rectangle corresponds to
the part of the process that is common to both Figures.
The process starts with an action “stop preoperative” which allows to model the patient exiting
the surgical suite just before starting the preoperative phase based on its attributes. Then, the
process is divided into three strands based on the patient anesthesia type: (1) general, local or
spinal anesthesia, (2) locoregional anesthesia, or (3) ophthalmologic induction.
A note on this model: patient queues rules. In our modeling, process #1 and process
#2 are FIFO based: the first patient to arrive in the surgical suite is the first one to be
escorted to their preoperative location, and the first one to receive preoperative care.
However, (1) the sequencing of patients receiving perioperative care is based on the
operating schedule, and (2) perioperative care resources can be required during the
preoperative care. Thus, using the process highlighted by a green rectangle, we sort the
patients so that only the one scheduled to be next can move on to the next phase of the
patient pathway.
FlexSim HC Patient Flow (process) #3: Operative Care (Figure 85 and Figure 86). We
display the flowchart of the operative care in Figure x and Figure x. The process models two
situations:
(1) Left-side process: the patient type anesthesia does not require the anesthesiologist to
provide care in the OR, or there is not anesthesiologist assigned to the patient in the
database.
(2) Right-side process: the patient type anesthesia requires the anesthesiologist to provide
care in the OR and there is an anesthesiologist assigned to the patient in the database
The rest of the operative steps are common to all patients.

227
 Figure 84 – Process #2: Preoperative Care (Part 2/2)

228
 Appendices

Figure 85 – Process #3: Operative Care (Part 1/2)

229
 Figure 86 - Process #3: Operative Care (Part 2/2)

FlexSim HC Patient Flow (process) #4: Patient Recovery (Figure 87) & Process #5:
Escort Patient to Exit (Figure 88). Once the patient’s surgery is over, they are moved to the
recovery room, where they stay until their state is stabilized. After this, the stretcher bearer
comes get the patient and escort them outside of the surgical suite.
FlexSim HC Patient Flow (process) #6: Room Cleaning (Figure 89). As soon as the
patient exits the OR, an assistant nurse comes in and cleans it.

230
 Appendices

Figure 87 – Process #4. Patient Recovery Figure 88 – Process #5: Escort Patient to Exit

Figure 89 – Process #6: Room Cleaning

231
232
OUTIL D’AIDE A LA DECISION A BASE DE JUMEAU NUMERIQUE
POUR L’ANALYSE PROSPECTIVE ET RETROSPECTIVE D’UN
PROGRAMME DE BLOC OPERATOIRE SOUMIS A DES INCERTITUDES

Résumé :
Avec l'augmentation de la demande de soins dans le monde, les services hospitaliers sont de
plus en plus sollicités. Leur performance est étroitement liée à la performance de leur bloc
opératoire. En effet, le bloc opératoire est un important centre de revenus et de dépenses
puisqu'il représente 40% du budget de l'hôpital (Macario et al. 1997), et que 60% des patients
viennent à l'hôpital pour une intervention chirurgicale (Fugener et al. 2017). Il est donc
nécessaire que les blocs opératoires soient efficients.
Cependant, cela est rendu difficile par la complexité de leur organisation due à la diversité des
parcours patients, la multiplicité des métiers, les liens étroits avec les services amont et aval, la
synchronisation de plusieurs ressources et flux logistiques (personnels, médicaments et
dispositifs médicaux), etc. D'autre part, la variabilité des durées et les perturbations inhérentes
à la pratique médicale, comme les cas d'urgence, sont les principaux facteurs et événements qui
dégradent le programme opératoire et impliquent que le personnel prenne de fréquentes
décisions pour maintenir l'activité du bloc opératoire de manière optimale. Par conséquent, les
activités de planification et d'ordonnancement du bloc opératoire intéressent de plus en plus la
communauté scientifique.
Dans cette thèse de doctorat, nous nous concentrons sur les niveaux opérationnels hors ligne
et en ligne (Hans et Vanberkel 2012). Ceci nous amène aux questions de recherche suivantes :
(1) Comment évaluer la robustesse et la résilience du programme opératoire avant son exécution
(dimension prospective) ? (2) Comment rejouer le programme opératoire pour avoir un retour
d'expérience et évaluer les décisions prises lors de son exécution (dimension rétrospective) ?
La contribution de ce manuscrit est triple : (1) Nous proposons un système d'aide à la décision
basé sur un jumeau numérique pour la simulation et l'analyse prospectives et rétrospectives de
l'exécution du programme opératoire. (2) Nous décrivons une méthodologie standardisée pour
concevoir, construire et mettre en œuvre cet outil dans n'importe quel bloc opératoire. (3) Cette
méthodologie est appliquée à un bloc opératoire inspiré de l'Hôpital Privé de La Baie (groupe
Vivalto Santé), afin de disposer d'une preuve de concept permettant de simuler un programme
opératoire de façon prospective et rétrospective.
Mots clés : Bloc opératoire, Jumeau numérique, Régulation, Outil d’aide à la décision,
Modélisation et simulation, Incertitudes

233
DIGITAL TWIN-BASED DECISION SUPPORT SYSTEM FOR THE
PROSPECTIVE AND THE RETROSPECTIVE ANALYSIS OF AN
OPERATING ROOM UNDER UNCERTAINTIES

Abstract:
With healthcare demand rising worldwide, hospital services are increasingly needed. Hospitals’
performance is tightly linked to their surgical suite performance. Indeed, the surgical suite is an
important revenue and expense center with over 40% of the hospital’s budget dedicated to it
(Macario et al. 1997) and 60% of the patient coming into the hospital for surgical intervention
(Fugener et al. 2017). This makes it necessary for surgical suites to be efficient.
However, running a profitable surgical suite is quite hard and requires a methodological
approach due to the complexity of its functioning: the diversity of patient pathways, the
multiplicity of professions, the tight link with upstream and downstream wards, the
synchronization of several resources and logistic flows (drug and medical devices), etc. On the
other hand, durations variability and disruptions inherent in medical care like emergency cases
are the main factors and events that degrade the scheduled execution and involve the staff
making decisions frequently to preserve the surgical suite activity in an optimal way. Therefore,
OR planning and scheduling activities are of increasing interest to the scientific community.
In this PhD thesis, we focus on offline operational and online operational levels (Hans and
Vanberkel 2012). This leads us to the following research questions: (1) How can we assess the
robustness and the resilience of the schedule before its execution (prospective way)? (2) How
can we replay the schedule to have feedback and assess the decisions made during its execution
(retrospective way)?
The contribution of this manuscript is threefold: (1) we propose a digital twin-based decision
support system for the prospective and retrospective simulation and analysis of the operating
room schedule execution, (2) we describe a standardized methodology to conceive, build and
implement this tool in any surgical suite, (3) This methodology is applied to an operating room
inspired by the Private Hospital of La Baie (Vivalto Santé group, France), in order to have a
proof of concept allowing to simulate an operating program prospectively and retrospectively.
Keywords: Operating room, Digital twin, Operating room management, Decision support
system, Modeling and simulation, Uncertainties.