Highly appreciate the use of modeling and simulation methods
FDA published a white paper titledChallenge and Opportunity on the Critical Path to New Medical
Products in 2004, which emphasizes the importance of model-based drug
development in new drug
R&D
[22]. More than half (54.8%) of the 31 FDA-approved antimicrobial
agents are supported by population PK modeling analysis (Figure
2 ) for the safety and efficacy evaluation in the NDAs of 12
antibacterial agents, 4 antifungal agents, and 1 antituberculosis drug
(including 14 single-component and 3 combination products). Most
(73.1%) of the 26 EMA-approved antimicrobial agents are supported by
population PK modeling analysis (Figure 2 ) for the safety and
efficacy evaluation in the NDAs of 14 antibacterial agents, 3 antifungal
agents, and 2 antituberculosis drugs (including 17 single-component and
2 combination products).
Population PK modeling analysis is not available for some drugs, e.g.,
ozenoxacin (Xepi), efinaconazole (Jublia), and finafloxacin (Xtoro), due
to the low-level absorption or lack of systemic exposure after topical
administration [23-25]. As for the other drugs without population PK
modeling analysis, PK analysis in special populations (in terms of sex,
age, hepatic/renal impairment) was provided to support the submission
and evaluation of NDA.
Population PK model is built primarily upon the pooled data from phase 1
to 3 clinical trials in healthy subjects and the patients with target
indication, which are provided by the applicant. The PK data from
animals were also used to build models in some cases. As for combination
products (e.g., ceftolozane-tazobactam), population PK model should be
built for each drug component on a case-by-case basis [26]. If there
are major active metabolites, population PK analysis should be conducted
for such metabolites. If there are multiple routes of administration
(e.g., delafloxacin), multiple models should also be built on a
case-by-case basis. In the case of delafloxacin, population PK modeling
analysis was provided not only for intravenous administration but also
for oral administration [27].
Population PK models are used widely in the evaluation of various
antimicrobial agents. Furthermore, Monte Carlo simulation and other
tools are also used simultaneously in the PK-PD analysis of
antibacterial agents. The MICs of an antibacterial agent against various
target pathogens are combined with the PK data to calculate the PTA for
the proposed PK-PD target value. Based on the calculated PTA, the
probability of different dosing regimens to reach the expected maximalin vivo bactericidal or bacteriostatic effect is evaluated. In
this way, the optimal dosing regimen can be recommended with sound
rationale. Just take ceftolozane-tazobactam for example, its PTA of
PK-PD target was calculated via population PK modeling analysis of
ceftolozane alone and tazobactam alone, non-clinical PK-PD target
values, in vitro susceptibility testing results, and Monte Carlo
simulation [26]. Model selection and development may have a
substantial impact on the subsequent simulation. There are advantages
and disadvantages for different models (e.g. MIC-based model, killing
curve model, and other sophisticated methods), and we should select the
appropriate models with justification based on the characters of the
compound. In the model development, attention should also be paid to
whether it is a concentration-dependent or time-dependent pattern of
bactericidal activity observed in time-kill studies, and then select the
relevant parameters to predict of efficacy in PK-PD model. Adequacy of
the reasons and justification of the extent are essentially important in
model building. NONMEM and R are the software commonly used for
modelling. The applicant is required to set up the parameters according
to specific data characteristics. For example, a three-compartment model
featuring zero-order absorption and first-order elimination was used to
describe the plasma concentration-time data pooled from phase 1, 2 and 3
clinical trials of plazomicin. Model selection can be informed by
goodness-of-fit plots, including the accuracy of parameter estimates,
scatter plot, correlation, and convergence between parameter estimates,
modification and conditions of the object function. Non-parametric
bootstrapping methods are also appropriate for evaluating the final
model and estimating the index with standard deviation and 95%
confidence interval (CI) of population PK parameters [28]. The
regulators would repeat the population PK analysis submitted by the
applicant to assess its rationality and adequacy in describing the PK
data of the drug and the effects of relevant covariates on PK
parameters. If any fault is identified, the regulator would correct it,
e.g., point out that the parameters and range of sample selection in the
model of meropenem-vaborbactam were not set up appropriately by the
applicant, and propose to run an independent population PK analysis of
updated dataset to assist evaluation.
FIGURE 2 PLACEHOLDER