CHAPTER 6

EVIDENCE BASED TREATMENT SUGGESTION MODEL FOR

PREDICTED AILMENTS

6.1 OVERVIEW

As per the survey statistics referred in World Health Organization (WHO) World Health
Organization [2020], India has a high weight of transmittable, including Neglected Tropical
Diseases (NTDs), maternal and neonatal conditions and a rising weight of Non-transferable
diseases, psychological wellness and wounds Priya et al., [2020b] Priya et al., [2020a] as
specified in the report. Untimely mortality from NCDs represents 57% of absolute mortality
particularly among 30-69 ages gathering of individuals. Modified TB rate fluctuates from
2.2 to 2.8 million/year and Tuberculosis (TB) mortality differs from 220000 to 480000 every
year. It obviously shows that India has 27% of worldwide Tuberculosis (TB) weight and 34%
of TB mortality Gamberger et al., [2000] as defined in the survey report.
Electronic Medical Records are the digital equivalent of paper records that are maintained
at the hospital to keep track of the patient’s health condition. EMR contains the treatment
and medical history of a patient as it is collected by the individual medical practice. By
implementing EMR, patient data can be tracked over an extended period by multiple
healthcare providers. It can help identify those who are due for preventive checkups
and screenings and monitor how each patient measures up to certain requirements such as
vaccinations and blood pressure readings. EMRs are designed to help organizations provide
efficient and precise care. Using EMR systems, patients can quickly and easily receive
electronic copies of their healthcare information upon request. In the event they choose to
change healthcare providers, the universal detailed medical record will help give the patients’
new healthcare provider a comprehensive understanding of their medical history and current
state of health. Healthcare facilities will still be able to exchange patient information if the
patient does not request it.
Most of the information and contents that are available on the internet such as,
websites, blogs, videos are created by humans. However, the Internet of Things (IoT)
has made a revolution in the internet technology by making micro-devices to collect
information from any environment and allows to transfer them to the server. Nowadays, IoT
devices are used in various environments such as hospitals, agriculture fields, retail shops,

112
manufacturing industries, autonomous vehicles, forests, and even in the ocean to collect
real-time information. IoT is a network of lightweight embedded devices with multiple
sensors, connectivity devices, and software programs. IoT devices sense the information in
real-time from the environment and transfer them to cloud servers through the internet. Some
of the popular IoT devices that are available in the market are smartphones, smartwatches,
smart fire alarms, medical sensors, fitness trackers, smart security systems. IoT makes
virtually everything "smart," by improving the power of data collection, AI algorithms, and
networks. The thing in IoT can also be a person with a diabetes monitor implant, an animal
with tracking devices, etc.
Professionals belonging to the healthcare industry are at most provoked due to asymptotic
conditions which are more common in complicated or mixed comorbid factors where the
traits differ pertaining to distinct ethnic etiquettes. There may be a difference in the chest
pain caused to the patients with different food intake too. People who don’t have the basic
knowledge of English get checked or treated by a foreign doctor who doesn’t know the
[Link] et al., [2012]. Even Though the variation in pain differs as defined by Suo
et al., [2018], with respect to Indians pertaining to chest pain which would be a tough task
for high risk assessment. Sufferers from the village areas feel extremely uncomfortable
specifically as stated by Simacek et al., [2019] due to low knowledge and lack of family
support and terror of being shifted to an unknown new urban surroundings.
Severe causes of chest pain include high thickening of the heart muscle, jaw or hand strain
or breathe inactivity has to be assessed within a prior time thereby rushing to the nearest
health care centres or medical institution. After the assessment on diagnosing reasons
causing the acute symptoms pertaining to life threatening situations has to be evaluated and
reported or educated to the patients and to the caregivers. Recent advancement with a 12-lead
electrocardiograph (ECG) and blood tests for troponin enables the risk threat identification
and treatment path thereby ensuring the lives of the sufferers. If the caregivers are educated
enough by interpreting the ECG reports via internet care can be improved effectively. As
described by Johnson et al., [2016b], if specificity about the threat becomes low then the
sensitivity of the tests would definitely yield a broader understanding of the requirement for
serial troponins in accordance with interpretable ECGs for the accurate diagnosis of AMI.
In remote areas, Point-of-care testing as specified by Goldberger et al., [2000] for
troponins has become easier in this decade where a single test troponin has high sensitivity
cardiac troponin assay for ruling out AMI after a single blood test. Overlooking of patients

113
who have been detected by severity of heart pain can cause death due to arrhythmias. In
urban areas coronary care centres are high in number whereas in rural areas this facility
would be rare. Hence detecting and monitoring is needed till the patient gets transferred
to a large medical institution. In rural areas the directions pertaining to medical treatment
congruent to other medications such as aspirin, platelet P2Y12 receptor antagonists (such as
clopidogrel, ticagrelor, and prasugrel), beta blockers, and ailments pertaining to increased
threat of ischemic conditions pertaining to embolus or bleeding conditions.
In the machine-driven lifestyle, maintaining accurate care for individual health has
become a challenging factor. Due to the technological improvements in the healthcare sector,
real-time health monitoring has become an eminent system. Especially after the outspread of
COVID-19 and Remote Patient Monitoring (RPM), the usage of IoT-based healthcare devices
has increased exponentially. There are numerous categories of health monitoring systems
available in the market. Mostly, wearable devices with the concept of the Internet of Things
(IoT) in the health domain have a great impact on real-time monitoring systems for patients
of different aged groups. To create a smart system in the healthcare domain, rather than a
standalone wearable gadget, it is essential to have a complete ecosystem.
In the twenty-first century, automation in every sector has been highly adopted. To
make human life easier, comfortable, and healthier, innovative technologies are applied in
healthcare. Even so, attaining an accurate and efficient solution on healthcare monitoring
remains challenging. Machine learning algorithms are widely used to extract useful and
informative patterns from the information collected by IoT healthcare devices. Analyzing
these data using machine learning algorithms allows the physicians to predict the condition
of the patients and effectively prevents untimely or early deaths.
Provision of medical care in both metropolitan and rural districts differ largely and has
an impact on the quantity of people living over there. Also, the transport system availability
impacts the surge and decline in serving the people in the nearby arenas. Actions and
directions pertaining to severe heart disease lies in acquiring facilities related to cardiac
catheterisation and expert specific assistance that are unavailable in case of village areas.
This poses a severe heart pain sufferers to be pathetic as mentioned by Alam et al., [2011],
if subsequent course of action was not undertaken.
Bypass Surgery With Grafting (CABG ) enables a patient’s heart to repair and to remove
the blockage which are uncommon in case of severe;y affected patients. People whose
arteries and blood vessels are affected require this kind of treatment especially people

114
who have got cardiovascular heart disease due to high blood glucose concentration levels.
of Statistics and Programme Implementation [2016]; India-HelpAge [2014]. Following
these surgeries patients may suffer from other complications such as wounds and limitations
posing to transport in flight and bus pertaining to village environments. Issues such as deep
vein thrombosis need compression stockings, ingress to sufficient acquirance for systematic
secondary prevention, and critical care, patient recuperation and discharge.
Machine learning in healthcare is becoming more widely used and is helping patients
and clinicians in many different ways. The most common healthcare use cases for machine
learning are clinical decision support and the development of clinical care guidelines. There
are many notable examples of machine learning and healthcare concepts being applied in
medicine. Using machine learning algorithms in healthcare identifies the complex patterns
from the prior collected data and helps the physicians to make intelligent decisions.
Extraction of unknown patterns from the available EHR data helps to define the correlation
factors of diseases as well as for the computation of similarity score which exists between
the currently treated and new patients. The secondary attributes of health data such as
laboratory test reports, treatment and prescription notes are used directly for training model
using conventional or modern ML algorithms. To extract more relevant information of
EMR, doctors clinical prescriptional notes are further processed using an appropriate NLP
techniques. In certain clinical cases, patients similarity scores are calculated on the basis
of patients survival and improvement rates. This mechanism may be appropriate only for
a specific disease. Another way to compute the distance metric of patient’s similarity
is to incorporate feedback mechanism of multiple health professionals. The Personalized
healthcare systems are designed using deep learning network models for measuring the
similarity scores of patients.
The proposed idea is to diminish the pace of mortality because of either transmittable
or Non-transferable infections using evidence-based treatment suggestion models. The main
objective of the work is to make predictive analysis using various intelligent algorithms based
on comparison study of clinical decisions stored in patients Electronic Health Record (EHR)
along with patient’s health history, disease symptoms and demographics. The patient’s
similarity score is computed to improve the clinical results. It is useful in a variety
of situations, including new patient therapy or drug recommendations, clinical diagnosis,
therapeutic decision assistance, and many more. As a result, quick disease diagnosis and
appropriate treatment can help to lower the death rate in our country.

115
 In the proposed work, web application considers the medical signs and its levels of
severity expressed by the patient and afterward proposes corresponded manifestations to
the patient. In case, the client neglected to make reference to similar prior health issues
additionally at least five indications are taken into consideration. Finally, the evidence based
on similarity score as per the predicted ailment, a suitable treatment with health expert
opinion is recommended to the patient.

6.2 CONCEPTUAL DESIGN

The proposed conceptual framework is created to analyze different sicknesses dependent on

the likeness lattice of loads of specific indications for any predefined infection. Likewise, the
paper assessed the presentation of illness determination utilizing different ML calculations
and correlation study is completed as shown in Figure 6.1. Among these, SVM ensured
better exactness in infection determination.

Figure 6.1: Conceptual Flow of Evidence Based Treatment Suggestion Model

The enduser or patients logs into the system and enters the current symptoms he or
she is experiencing, as well as their severity degree. The system then generates a new list

116
of associated symptoms based on these parameters, from which the user should select the
symptoms they have and their severity. The model compares the symptoms and their severity
to an illness, and if the condition is identified, it calculates the appropriate treatment. Finally,
it is examined by a physician, and the patient is informed of the ailment and its treatment.

6.3 METHODOLOGY

The entire flow of the ailment diagnosis and treatment recommendation is depicted in Figure
6.2. The model’s input consists of a set of symptoms encountered by users, along with their
severity. Following the input of a set of symptoms, the user is presented with a list of related
symptoms (intelligently identified by proposed system based on the symptoms provided as
input) and asked whether these symptoms also occurred. This is done to ensure that the
user conveys all symptom-related information and to improve the accuracy of the result. The
severity of the symptoms is used to assign values to them. These symptoms are then fed into
the model as input.

Figure 6.2: Flow Diagram for Ailment Diagnosis and Treatment Recommendation. PA* -
Predicted Ailment, RT* - Recommended Treatment

117
6.3.1 Data Acquiring

The proposed model uses the Kaggle dataset and PubMed data Vaani [2013]; Bhardwaj
[2022]; Johnson et al., [2016a], which consists of 5597 datasets with 560 attributes, each
attribute representing a unique symptom, the datasets are classified into different diseases as
well as by stages of those diseases. e.g., records of cardiomyopathy are classified as dilated
cardiomyopathy; Hypertrophic cardiomyopathy etc. The system used 80 % of the data for
training and 20 % for testing. For treatment recommendation model, the system uses PubMed
dataset comprises of over 5000 ailments and 900 symptoms having approximately 2 Lakhs
patient records. In addition, MIMIC-III dataset comprising nearly 40,000 intensive care unit
patients record with various treatment prescribed to the patients along with exact dosage of
drugs, duration and the method of application.

a) Sample Dataset for Predicting ailments

Figure 6.3: b) Sample Dataset for Predicting ailments

118
 a) Sample Dataset for Treatment Suggestion Model

Figure 6.4: b) Sample Dataset for Treatment Suggestion Model

6.3.2 Data Preprocessing

For any prediction model, a series of steps such as data formatting, missing values by mean
method and data normalization are carried out for cleaning the medical datasets. Once the
cleaned data is retrieved, the same is fed into training model to perform classification and
association analysis. The Pearson correlation coefficient is calculated between each of the
symptoms entered by the user and the rest of the symptoms. The top 20 symptoms with
the highest correlation coefficient value are considered related symptoms. Given a pair of
random variables (X, Y), the Pearson correlation coefficient is calculated using the equation
(6. 11) as follows
cov(X, Y )
ρX, Y = (6.1)
σX σY

Where, cov is the covariance X is the standard deviation of X and Y is the standard
deviation of Y.

119
Figure 6.5: Flow Chart of the Treatment Suggestion Model for Predicted Ailments

a. Use case diagram for Patient Case Simliarity

120
 The proposed model is aimed for designing an efficient diagnosis model based on
similarity measures of various symptoms using ML techniques. It also provides the best
treatment suggestion for a disease, based on patients EHR records. The use case diagram
for the disease diagnosis model designed on the basis of patients similarity score concepts is
described in Figure 6.4. According to the given Figure 6.6, end-user is logged into the system
to provide set of symptoms as system input along with its varied levels of symptom severity.
Symptom severity is considered as any one of the values such as low, mild, moderate and
high depending on the symptom’s threshold value. Based on the chosen factors, the system
subsequently identifies a new set of related symptoms. The model with the symptoms
and their severity are compared with a label of diagnosed disease and once the ailment is
predicted, it calculates the treatment that should be undertaken.

Figure 6.6: Use case Diagram for Disease Diagnosis based on Patients Similarity Score

121
6.3.3 Data Flow Diagram (DFD)

The DFD for the proposed model to predict ailment using patient similarity score is
represented in the Figure 6.7 as follows:

Figure 6.7: DFD Level 1 for Ailment Prediction

6.3.4 Predicted Ailments Scheme (PAS)

The scheme to define ailment prediction uses detailed diagnosis of patient’s medical data
descriptions specified by doctors’ clinical notes. In addition, it considers the demographic
nature of patients rather than any disease simple definition. Based on the symptoms and
severity clusters in association with similarity scores, ailments dataset (A1, A2,..., An) is
forwarded as an input to treatment recommendation module.

6.3.5 Ailment Prediction

Disease prediction is a classification downside during which diseases are the categories
and symptoms act as features. The ultimate set of symptoms generated when computing
connected symptoms is provided as input to proposed model. These symptoms work as
weighted inputs whereas severity is defined as weight. This happens, as a result of some
diseases can have constant set of symptoms however with variable severity for every symptom.

122
Hence, the proposed model considers symptom severity and similarity score for ailment
prediction.

6.3.6 Treatment Recommendation

At this stage, treatment dataset scheme is considered as RTs = (T1, T2,..., Tn). The association
analysis is performed between the ailment prediction datasets scheme and treatment datasets
scheme using Apriori algorithm. A suitable associated item between predicted ailment
sets and recommended treatment sets are defined. The achieved confidence rate of the
treatment recommendation model is 82 %. Once a disorder is predicted, the end result is
despatched over to a registered health practitioner for verification. After the health practitioner
approves the end result, the version shows suitable clinical remedy to the user. This remedy
might also additionally encompass drug prescriptions, hospitalization, etc. The treatment is
recommended on the basis of association with the predicted ailment.

6.4 Result and Analysis

The proposed system aims to locate similarity between patients’ medical data using a
similarity matrix of weights of particular symptoms for a given illness. The work also
suggests an appropriate treatment based on various health records of patients and their
demographics.

6.4.1 User Interface

Once the user logged in, users are allowed to select the set of symptoms and its severity levels
as shown in Figure 6.8.

Figure 6.8: User Interface to Select Symptoms and Severity Levels

123
 The related symptom determination module is depicted in Figure 6.9, based on the
previous set of chosen symptoms.

Figure 6.9: User Interface for Related Symptom Determination

After the submissions of symptom and its severity input, the similarity score is computed
and the predicted ailment and treatment recommendation for the same is shown as in Figure
6.10.

Figure 6.10: User Interface for Ailment Prediction and Treatment Recommendation

124
 Additionally, patients past health history are displayed for the health expert to view and
either approve or reject the recommended treatment for the predicted ailment. Finally, system
accepts doctor’s final decision and the same information is updated as clinical notes in the
proposed system for further references.

6.4.2 Performance Measures

To increase system accuracy different machine learning algorithms were initially

implemented, giving accuracy as 80 %, 98 %, 97 %, 97 %, from Naive Bayes, SVM,
Random Forest and KNN respectively as depicted in Figure 6.11.

Table 6.1: Performance Metrics Comparison Using Various ML Algorithms

ML Models Accuracy (%) Precision (%) Recall (%) F1-Score (%)

SVM 98 97.6 97.5 97.9
Naive Bayes 80 80.1 79.2 80.5
K-NN 97 96.7 96.3 96.9
Random Forest 97 96.4 95.6 96.6

As indicated in table 6.1, the performance metrics comparison of various ML algorithms

as when the algorithms are tested for 20 % of the unknown samples of medical data. Among
the different ML models, SVM achieved support of 98 %, precision of 97.6 %, Recall value as
97.5 and F1-score as 97.9 % respectively. Whereas, Random Forest and K-NN yields nearly
97 % of various evaluation metrics. SVM performs better when evaluated for the above four
performance metrics. From the above numbers, the system selected Support vector machine
as the best classifier model.

6.5 SUMMARY

Disease prognosis at an early level allows in lowering the death rate. Classifying primarily
based totally on affected person demographics and keeping the document for prescriptions
and scientific facts for similarly analysis. Diagnosis version is skilled the usage of diverse
ML strategies and an assessment observe become experimented. Other than NB algorithm,
SVM, KNN and Random Forest executed higher accuracy to diagnose any disorder primarily
depends completely on a given set of symptoms. Integration with actual health data can
assist in validating the predictions produced through the models, for this reason this could

125
yield to greater correct and dependable results. The future plan is to consider a few admitted
patients’ cases with respect to treatments provided for any specific ailment and recommend
suitable drugs as per patient’s health data.

126