J Antimicrob Chemother 2018; 73: 564–568

doi:10.1093/jac/dkx427 Advance Access publication 5 December 2017

MIC-based dose adjustment: facts and fables

Johan W. Mouton1*, Anouk E. Muller1,2, Rafael Canton3, Christian G. Giske4, Gunnar Kahlmeter5 and
John Turnidge6

1
Department of Medical Microbiology and Infectious Diseases, Erasmus MC, Rotterdam, The Netherlands; 2Department of Medical
Microbiology, Haaglanden Medical Centre, The Hague, The Netherlands; 3Servicio de Microbiologı́a, Hospital Universitario Ramón y
Cajal and Instituto Ramón y Cajal de Investigación Sanitaria (IRYIS), Madrid, Spain; 4Department of Laboratory Medicine, Division of
Clinical Microbiology, Karolinska Institute and Karolinska University Hospital, Stockholm, Sweden; 5Department of Clinical Microbiology,

Downloaded from [Link] by guest on 29 April 2023

Central Hospital, 351 85, Växjö, Sweden; 6Adelaide Medical School, University of Adelaide, Adelaide, Australia

*Corresponding author. E-mail: jwmouton@[Link]

Over recent decades, several publications have described optimization procedures for antibiotic therapy in the
individual patient based on antimicrobial MIC values. Most methods include therapeutic drug monitoring and
use a single MIC determination plus the relevant pharmacokinetics/pharmacodynamics to adjust the dose to
optimize antimicrobial drug exposure and antibacterial effects. However, the use of an MIC obtained by a single
MIC determination is inappropriate. First, routine clinical laboratories cannot determine MICs with sufficient
accuracy to guide dosage owing to the inherent assay variation in the MIC test. Second, the variation in any MIC
determination, whatever method is used, must be accounted for. If dose adjustments are made based on thera-
peutic drug monitoring and include MIC determinations, MIC variation must be considered to prevent potential
underdosing of patients. We present the problems and some approaches that could be used in clinical practice.

Introduction Here we briefly provide the scientific basis for this issue and
present a number of examples to underscore our point of view
In recent years, several papers have described optimization pro- based on current knowledge of MIC measurement characteristics
cedures for antibiotic therapy based on microbial MIC values.1–4 and pharmacokinetic profiling.
The basis for these is usually that the MICs of relevant antimicro-
bials are determined and reported to the clinician. It is proposed
that concurrently determined antimicrobial serum concentra- What does an MIC represent?
tions are used to determine the individual pharmacokinetic pro- An MIC is obtained using an assay that determines in vitro the con-
file of the drug, usually with the help of a computerized centration that prevents visible growth (significant cell multiplica-
population pharmacokinetic program. The program is expected tion as determined by the unaided eye) in or on a standardized
to help the clinician to adjust the dosing regimen to optimize medium over 18!2 h, at 34 C–37 C using a standardized inocu-
exposure, taking into account pharmacokinetic/pharmacody-
lum.5 The concentrations used are 2-fold dilutions up and down
namic relationships, patient-specific pharmacokinetic character-
from the concentration 1 mg/L; thus typically 128, 64, 32, etc.
istics and the MIC value. Although pharmacokinetic variations
down to 0.016 and 0.008 mg/L. Therefore, the MIC is an endpoint
are addressed in population pharmacokinetic programs, few
measurement and represents the result of growth and bacterio-
clinicians are aware of the performance characteristics of MIC
static/bactericidal effects over time. In liquid media, visible growth
determination and its limitations. The truth is that with the
methods at hand and with few exceptions, laboratories are not occurs at bacterial densities of 107–108 cfu/mL. As the initial inoc-
capable of performing a sufficiently accurate and reproducible ulum in the test is 5%105 cfu/mL, growth may have occurred dur-
determination of an MIC value owing to the inherent assay varia- ing the period of incubation but not reached sufficient densities to
tion in the MIC test. In addition, semi-automated susceptibility become visible. The minimum in the MIC is therefore not a strict
testing machines yield truncated MICs and gradient tests can be minimum but rather an observable minimum. Should for some
quite often problematic. Even with perfectly performed tests the reason the inoculum be larger than that stipulated or the incuba-
variation may be pronounced, one reason being that the tests tion time longer, the MIC will probably be higher and vice versa for
produce discontinuous results on an interval scale (i.e. MIC values shorter incubation and smaller inoculum.
allocated to discrete values, usually a 2-fold dilution scale), Thus, the MIC does not represent a concentration that can be
which results in values falling across 2-fold intervals when varia- compared directly with any in vivo concentration found during
tion occurs. This may lead to significant dosing adjustment treatment. In particular, it does not predict the overall bacterial
errors, which could ultimately be harmful to patients. response whether the in vivo concentration is above or below the

C The Author 2017. Published by Oxford University Press on behalf of the British Society for Antimicrobial Chemotherapy. All rights reserved.
V
For Permissions, please email: [Link]@[Link].
564
For debate JAC
MIC.6,7 The MIC is the phenotypic endpoint in a defined and stand- requires a number of replicate measurements and not a single
ardized but highly relative system. Importantly, the MIC is meas- measurement.
ured in broth and not in the range of physiological environments
found in the human body. Bacterial growth conditions in the ‘test Assay variation
tube’ are significantly different from those in vivo and no direct
inference can or should be made about its relationship to in vivo Variation in MICs not explained by biological variation accounts for
concentrations.8 The MIC has value as a standardized measure of a significant part of the total variation observed in the WT distribu-
activity of the antibiotic against the microorganism. tion. Thus, the greater contribution to observed variation in MIC is
assay variation, which can mask any variation in MICs between
strains. If MICs are determined multiple times in several labs, it
What is the accuracy of an MIC? was estimated that at least half the variation observed was due to
The accuracy of an MIC measurement is, with a few exceptions for assay variation, and less than half the variation was due to strain-
to-strain differences.9 Although we often perceive a certain strain

Downloaded from [Link] by guest on 29 April 2023

some organism–antimicrobial combinations, not very high. If the
MIC for a certain isolate versus an antibiotic is determined repeat- as something absolute and not changing, in reality bacteria divide
edly in one laboratory using the same standardized conditions and continuously, leading to subtle differences over time of the culture.
by the same lab technician, there will be a log2 standard deviation The definition of a strain is therefore a certain parental lineage but
of 0.3–0.5 around the log2 (the 2-fold dilution series is a base not the same individual bacteria. As the number of generations
2-logarithmic series). If the same strain–antimicrobial combina- increases over time, small changes in environment have the
tion is tested in a range of laboratories, the log2 standard deviation impact of variations between individual cultures.
might increase to 0.5–1 2-fold dilution and more when media
obtained from several manufacturers are used.9 The variability of MIC and pharmacokinetic/pharmacodynamic
an MIC measurement thus consists of several components: inter-
strain differences, intra-laboratory variability and inter-laboratory
relationships
variability. In other words, there is biological variation (strain-to- Pharmacokinetic/pharmacodynamic relationships have been
strain differences) and assay variation as a result of differences in described for most antimicrobial classes, using free serum concen-
inoculum preparation, media, incubation temperature and incuba- trations as a surrogate for the concentrations at the site of infec-
tion time, and variation between laboratories because different tion.13 These relationships show a marked consistency, and the
laboratories have different facilities and technical skills and degree pharmacodynamic index values (such as fAUC/MIC or %[Link])
of training. This variation is also well recognized in the accepted that result in a certain effect have been determined for most
MIC ranges of quality control strains, which often span over two to classes of antimicrobials.14,15 If the effect is measured in terms of
three dilutions and even four dilutions in some cases.5 Moreover, bacterial counts, the ones most often used are the exposures
the acceptance criteria for reproducibility of an MIC device follow- (pharmacokinetic/pharmacodynamic index values), resulting in a
ing the ISO 20776-2 norm indicate an acceptable deviation of one static effect over time or a 1 log10 kill effect, often called pharma-
dilution from the mode for 95% of cases, or a range of at least two codynamic targets. These can be determined in various models,
2-fold dilutions.10 including animal models and hollow fibre in vitro models.
Important for the present discussion is the MIC in the equation.
Biological variation (strain-to-strain The MICs for the strains used in these studies are usually deter-
mined multiple times in dedicated laboratories and the assay vari-
differences) within a species ability of the MIC is thus restricted to limited intra-laboratory
Individual strains within a species, even when there are no variation. Important, however, is that these targets are derived
acquired resistance mechanisms, have individual characteristics from a number of strains with a range of MICs and thus the point
other than their susceptibility to the agent that will provide differ- estimate of the pharmacodynamic index is therefore reasonably
ences in MICs. These include the rate of division (growth rate), met- well described, provided that enough strains have been tested in
abolic status and a variety of antimicrobial receptors. The MIC the model. When translating this to the clinical setting, this means
value summarizes the impact of all characteristics in a single test that the pharmacodynamic target is reasonably accurate for a cer-
result. However, it has become apparent that in strains devoid of tain disease setting. However, the application of the pharmacody-
acquired resistance to the agent in question (the so-called WT) the namic target to adjustments of dosing regimens based on a single
difference between MIC values within a species is slight; for most MIC result from an individual strain is not accurate, as explained in
species and agents it is only three to five 2-fold concentrations and the next section.
the major and reproducible difference is between individuals with-
out and with acquired resistance. The upper end of the WT distribu- MICs, therapeutic drug monitoring and
tion (the highest MIC for isolates devoid of phenotypically
detectable resistance) is defined as the epidemiological cut-off
dose adjustments
value (ECOFF).11,12 Under strict standard conditions strain-to-strain Therapeutic drug monitoring is increasingly being used for
differences belonging to the WT can sometimes be determined. In antimicrobial agents, particularly in patients in the ICU with unpre-
one laboratory, under standard conditions, the reproducibility of dictable clearance.15,16 Several studies have shown that preset
the test is better, most likely because part of the variation in phe- pharmacokinetic/pharmacodynamic targets were not met in such
notype is better controlled, as explained above. However, this patient groups, one of the major reasons being augmented

565
For debate

clearance.17 MICs for strains that are cultured from patient speci- determine adjustment of the dosing regimen. This is particularly
mens and likely to be the cause of the infection are increasingly true for new drugs with a new mechanism of action, against which
used to determine individual pharmacokinetic/pharmacodynamic low- and high-level resistance mechanisms are infrequent.
targets.2 It should be clear from the arguments above that this is The second situation is an MIC immediately above the ECOFF.
not justified, because the measurement of an MIC is not suffi- Here, the actual distribution of MICs for the particular strain that
ciently accurate. Thus, there is a significant risk of underdosing if by would be observed with repeated testing is unknown. In most of
chance the measured MIC was at the low end of values that would the cases and provided the laboratory is proficient in the MIC deter-
have been observed if the test had been performed repeatedly. mination, it could be argued that the reported MIC could be correct
Thus, even if the MIC varies over only two dilutions, the pharmaco- within a range of three 2-fold dilutions. However, if the laboratory
kinetic concentrations to be reached vary 4-fold because the phar- is less proficient or if the initial value represented the outer boun-
macokinetic/pharmacodynamic target such as the AUC/MIC may daries of the distribution obtained with repeat testing, the reported
vary 4-fold. value could be as much as four dilutions off. In addition, there is a

Downloaded from [Link] by guest on 29 April 2023

paucity of published data in the literature on the reproducibility
Is there a solution? Practical guidance from outside the WT population. However, there is no clear reason to
assume a priori that the variability would be higher or lower.
an MIC point of view An overly conservative system could lead to a risk of overdosing
Given the inaccuracy of an individual MIC measurement, there is and assuming a 2-fold variation even outside the WT distribution
no universal solution for MIC-guided therapy as will be demon- could be a reasonable compromise. Thus, although there is no con-
strated by the following examples. Nevertheless, some practical sensus on this issue, one suggestion could be to use a margin of at
guidance can be given as discussed below to avoid errors and mis- least two 2-fold dilutions if used for individual target attainment
interpretations in dosing adjustment. calculations. This would significantly increase the margin of safety.
First, consider MICs from a distribution point of view. In general, The third situation is an individual MIC value that is clearly out-
three sections of an MIC distribution can readily be distinguished side the WT population and clearly (far) above the clinical resist-
(Figure 1). The first is the WT distribution (white area) representing ance breakpoint. There is little chance of increasing exposure to
MICs for isolates considered devoid of phenotypically detectable reach the desired pharmacodynamic target. MIC-guided therapy
resistance mechanisms. The second (striped area) represents MICs is not an option.
for strains just outside the WT distribution with low-level resistance The fourth situation sometimes arises as a complication of the
and in which isolates may or may not be different from each first two. Owing to emerging resistance and because of the inac-
other. An example here is the so-called plasmid-mediated quino- curacy of measurement of the MIC, a strain with a low-level resist-
lone resistance mechanism such as the qnr or aac(60 )-Ib-cr genes ance mechanism could be assigned to the WT, particularly if it is
in Escherichia coli18 or a low-level expression of certain based on a single measurement. As the probability of this is
b-lactamases.19 Sometimes resistance to the agent is so discrete dependent on the frequency of low-level resistance, this is unlikely
that another related agent is better suited to disclose the presence to occur in areas where these isolates are rare (as illustrated in
of the resistance mechanism.20 Finally, there are the MICs for Figure 1a) but may become an issue where they are common
strains with a phenotypical high-level resistance (black area). (Figure 1b). In that case, an MIC with a phenotype of WT borderline
The questions that arise are: Given the poor accuracy of an individ- MIC represents a non-WT strain in a significant fraction of occur-
ual MIC measurement for a particular isolate, to which of the three rences. It is not clear at present at which frequency level such a sit-
MIC distributions does the strain belong? Which MIC value could be uation should be taken into account; it will be different for each
used for pharmacodynamic attainment in the individual patient? drug–species combination and local or region specific. This clearly
As the MIC for a certain species and/or agent may well have a needs further study.
higher rate of accuracy than that obtained for another, does this The implications for MIC interpretation discussed above are
have a major impact on decision making? From a practical point of summarized in Table 1. It should be noted, however, that the num-
view there are four possibilities, and in each case the MIC used to ber of dilutions used as a safety margin, two in Table 1, are
determine the pharmacodynamic target in the individual patient dependent on laboratory proficiency and the extent of the WT dis-
warrants a different approach. tribution. For narrow drug–species distributions such as
The first scenario is that the MIC for the strain indicates that it is Staphylococcus aureus and vancomycin, the margin of error is
within the WT distribution. In this case, it should be clear that the likely much lower, as demonstrated by the studies showing a
MIC is not a fixed number but is within the range of values with the prominent pharmacodynamic relationship.22
ECOFF as the upper limit. The WT range of values follows a log-
normal distribution and is a composite of strain-to-strain variation
Conclusions
(biological variation) and both intra- and inter-laboratory variation
(assay variation). It follows that in this situation the ECOFF should We conclude that the use of an individual MIC value to modify a
be taken as the value to determine the pharmacokinetic/pharma- dosing regimen because a target is or is not likely to be attained is
codynamic target for the individual. ECOFF values are readily avail- not justified. The MIC should be interpreted in the context of assay
able at the web site of the EUCAST.21 In many situations, the variation, and species identification and WT distributions need to
ECOFF is similar to the clinical breakpoint; the isolate is reported as be taken into account. This is not always straightforward, in partic-
susceptible by the laboratory and it is therefore primarily the phar- ular if there is an abundance of local strains with low-level resist-
macokinetic and disease characteristics of the patient that will ance mechanisms.

566
For debate JAC
(a) 6000 (b) 1500
Number of strains

Number of strains
4000 1000

2000 500

0 0
0. 5
0. 03
0. 63
0. 5
25
5
1
2
4
8
16
32

124
258
516
2

0. 5
0. 03
0. 63
0. 5
25
5
1
2
4
8
16
32

124
258
516
2
01

12

0.

01

12

0.

Downloaded from [Link] by guest on 29 April 2023

0

0
0.

0.
MIC (mg/L) MIC (mg/L)

Figure 1. Schematic MIC distributions. White area, MICs with a WT phenotype as defined by EUCAST; striped area, MICs with a low-level resistance
phenotype; black area, MICs with a high-level resistance phenotype. (a) ECOFF " 0.25 mg/L, resistance rare. (b) ECOFF " 0.5 mg/L, resistance
common.

Table 1. Suggested interpretation of the MIC for target attainment isolates with low-level resistance become more frequent, gener-
under various conditions ally or locally, specific solutions need to be found to interpret MICs
of borderline susceptibility.
MIC found Interpretation for target attainment

Within WT, ECOFF ECOFF

.ECOFF MIC ! two 2-fold dilutionsa Transparency declarations
None to declare.
a
Number of dilutions could be higher or lower than two depending on
the proficiency of the lab and the drug–species distribution.
References
1 Economou CJP, Wong G, McWhinney B et al. Impact of b-lactam antibiotic
We must free ourselves from the misconception that there is therapeutic drug monitoring on dose adjustments in critically ill patients
such a thing as a ‘true’ MIC for a strain. Instead, each measure- undergoing continuous renal replacement therapy. Int J Antimicrob Agents
ment of an MIC generates a value that is a member of a probability 2017; 49: 589–94.
distribution. As discussed, all the evidence summarized in the 2 Charmillon A, Novy E, Agrinier N et al. The ANTIBIOPERF study: a nation-
wide cross-sectional survey about practices for b-lactam administration and
EUCAST WT distributions indicate that even though there are MIC
therapeutic drug monitoring among critically ill patients in France. Clin
differences between specific strains, these cannot be readily Microbiol Infect 2016; 22: 625–31.
detected or reproduced in the routine laboratory owing to the
3 Wong G, Brinkman A, Benefield RJ et al. An international, multicentre sur-
shortcomings of the system we employ. Moreover, for WT MIC dis- vey of b-lactam antibiotic therapeutic drug monitoring practice in intensive
tributions categorized as susceptible by the clinical breakpoints it is care units. J Antimicrob Chemother 2014; 69: 1416–23.
most likely that the difference between isolates, even when there, 4 Jager NG, van Hest RM, Lipman J et al. Therapeutic drug monitoring
is so small that it is doubtful as to the usefulness in a clinical of anti-infective agents in critically ill patients. Expert Rev Clin
situation. Furthermore, small changes to the test system (in pH, Pharmacol 2016; 9: 961–79.
incubation time, media, etc.) will drastically change the MIC of 5 International Organization for Standardization (ISO). ISO 20776-1. Clinical
many agents. We have otherwise deliberately not discussed clini- Laboratory Testing and In Vitro Diagnostic Test Systems—Susceptibility
cal breakpoints in relation to ECOFFs. The intention of the paper is Testing of Infectious Agents and Evaluation of Performance of Antimicrobial
to describe problems arising from accepting individual MIC deter- Susceptibility Testing Devices—Part 1. ISO, 2006.
minations to modify dosing regimens. 6 Mouton JW, Vinks AA. Pharmacokinetic/pharmacodynamic modelling of
Placing the MIC in the context of assay variation is particularly antibacterials in vitro and in vivo using bacterial growth and kill kinetics: the
minimum inhibitory concentration versus stationary concentration. Clin
important if it is used to optimize therapy in the individual patient
Pharmacokinet 2005; 44: 201–10.
using target attainment. It could very well be that the current
7 Bouvier d’Yvoire MJY, Maire PH. Dosage regimens of antibacterials.
paradigm to increase the pharmacodynamic target in the critically Implications of a pharmacokinetic/pharmacodynamic model. Clin Drug
ill to 100% [Link] of b-lactams because of decreased efficacy at Investig 1996; 11: 229–39.
lower targets23 is appropriate owing to variation in MIC measure- 8 Mouton JW, Vinks AA. Relationship between minimum inhibitory concen-
ments as discussed in this paper. Alternatively, it explains why tration and stationary concentration revisited: growth rates and minimum
multiples of the MIC trough concentration are sometimes found to bactericidal concentrations. Clin Pharmacokinet 2005; 44: 767–8.
be predictors of efficacy or emergence of resistance.24 This clearly 9 Voss A, Mouton JW, van Elzakker EPM et al. Linezolid susceptibility of
suggests the need for more specific research in this area. Finally, as glycopeptide-intermediately susceptible Staphylococcus aureus (GISA)—the

567
For debate

Dutch experience. In: Abstracts of the Thirteenth European Congress of Clinical 17 Udy AA, Roberts JA, Boots RJ et al. Augmented renal clearance: implica-
Microbiology and Infectious Diseases, Glasgow, UK, 2003. Abstract P1309. tions for antibacterial dosing in the critically ill. Clin Pharmacokinet 2010; 49:
ESCMID, Basel, Switzerland. 1–16.
10 International Organization for Standardization (ISO). ISO 20776-2. 18 Cavaco LM, Aarestrup FM. Evaluation of quinolones for use in detection of
Clinical Laboratory Testing and In Vitro Diagnostic Test Systems—Susceptibility determinants of acquired quinolone resistance, including the new transmissi-
Testing of Infectious Agents and Evaluation of Performance of Antimicrobial ble resistance mechanisms qnrA, qnrB, qnrS, and aac(60 )Ib-cr, in Escherichia
Susceptibility Testing Devices—Part 2. ISO, 2008. coli and Salmonella enterica and determinations of wild-type distributions.
11 Kahlmeter G, Brown DF, Goldstein FW et al. European harmonization J Clin Microbiol 2009; 47: 2751–8.
of MIC breakpoints for antimicrobial susceptibility testing of bacteria. 19 Kohner PC, Robberts FJ, Cockerill FR 3rd et al. Cephalosporin MIC distribu-
J Antimicrob Chemother 2003; 52: 145–8. tion of extended-spectrum-b-lactamase- and pAmpC-producing Escherichia
12 Kahlmeter G. The 2014 Garrod Lecture: EUCAST—are we heading coli and Klebsiella species. J Clin Microbiol 2009; 47: 2419–25.
towards international agreement? J Antimicrob Chemother 2015; 70: 20 Skov R, Matuschek E, Sjölund-Karlsson M et al. Development of a pefloxa-
2427–39. cin disk diffusion method for detection of fluoroquinolone-resistant

Downloaded from [Link] by guest on 29 April 2023

13 Craig WA. Pharmacokinetic/pharmacodynamic parameters: rationale for Salmonella enterica. J Clin Microbiol 2015; 53: 3411–7.
antibacterial dosing of mice and men. Clin Infect Dis 1998; 26: 1–10; quiz 21 EUCAST. [Link]
11–2. 22 Rybak MJ, Lomaestro BM, Rotschafer JC et al. Vancomycin therapeutic
14 Ambrose PG, Bhavnani SM, Rubino CM et al. Pharmacokinetics-pharma- guidelines: a summary of consensus recommendations from the Infectious
codynamics of antimicrobial therapy: it’s not just for mice anymore. Clin Diseases Society of America, the American Society of Health-System
Infect Dis 2007; 44: 79–86. Pharmacists, and the Society of Infectious Diseases Pharmacists. Clin Infect
15 Abdul-Aziz MH, Lipman J, Mouton JW et al. Applying pharmacokinetic/ Dis 2009; 49: 325–7.
pharmacodynamic principles in critically ill patients: optimizing efficacy and 23 Huttner A, Harbarth S, Hope WW et al. Therapeutic drug monitoring of
reducing resistance development. Semin Respir Crit Care Med 2015; 36: the b-lactam antibiotics: what is the evidence and which patients should we
136–53. be using it for? J Antimicrob Chemother 2015; 70: 3178–83.
16 Roberts JA, Abdul-Aziz MH, Lipman J et al. Individualised antibiotic dosing 24 Tam VH, McKinnon PS, Akins RL et al. Pharmacodynamics of cefepime in
for patients who are critically ill: challenges and potential solutions. Lancet patients with Gram-negative infections. J Antimicrob Chemother 2002; 50:
Infect Dis 2014; 14: 498–509. 425–8.

568