Journal of Pharmacokinetics and Pharmacodynamics最新文献

Antimicrobial therapy is informed by quantitative models of drug disposition and action. These models utilize experimental and observational evidence, subject to uncertainties, to support drug selection and dosage regimen optimization, and interpret antimicrobial resistance data. The framework includes multiple components, which characterize mechanisms contributing to therapeutic outcome. The components must be combined in a logical sequence to generate predictions, so propagation of uncertainty is a critical consideration. Quantitative evaluation of this uncertainty has received apparently little attention. This essay argues for the importance of uncertainty quantification in antimicrobial pharmacology. The impact of parameter uncertainties and measurement errors on the validity of pharmacokinetic-pharmacodynamic modelling of antimicrobials is described. Major components of the modelling workflow are assessed, and uncertainties characterized. The influence of major design and statistical analysis decisions at each step is emphasized. Finally, using detailed simulations, the impact of these sources of uncertainty on outcomes including clinical breakpoints and dose individualization is illustrated. Measurement of antimicrobial potency as the minimum inhibitory concentration contributes approximately twofold error, which is important for individual dose determination. Interpretation of PK/PD parameters is generally conducted dichotomously as thresholds, which are empirically determined, and subject to error. Parameter uncertainties in the exposure-response relationship are potentially substantial, and contribute apparently major uncertainty to predictions at both population and individual levels. The importance of uncertainty in pharmacokinetics appears context-sensitive. Applications including dose optimization or susceptibility breakpoints appear overly confident, and point estimation from these models may be an unreliable basis for decision making. These observations highlight the importance of uncertainty quantification for rigorous antimicrobial pharmacology.

Regulatory evaluation of QT interval prolongation remains central to cardiac safety assessment in drug development. Since the 2015 revision of the ICH E14 Q&A, concentration-QT (C-QT) modelling has been formally recognized as an acceptable alternative to dedicated Thorough QT (TQT) studies, offering ethical and practical advantages. This study aimed to characterize how the European Medicines Agency (EMA) has assessed C-QT modelling approaches over the past five years, with particular focus on regulatory acceptance of TQT waiver requests and the recurring drivers of rejection. A retrospective review was performed of EMA Scientific Advice (SA) documents issued between January 2020 and January 2025. Using the internal text-mining platform Scientific Explorer, 524 SA cases (4,196 applicant questions) were screened for "QT" or "QTc." A custom Python tool extracted relevant discussions, which were subsequently categorized by expert review. Regulatory feedbacks were classified as supportive, conditionally supportive, or unsupportive. Among 110 QT-related requests, 81% sought TQT waivers, most justified by C-QT modelling. Of these, 70% were supported, 9% conditionally endorsed, and 21% rejected. Common rejection drivers included insufficient exposure margins (n = 8), study design limitations (n = 5), data gaps (n = 7), unclear methodological reporting (n = 5), QTc interpretation concerns (n = 5), and additional methodological weaknesses (n = 4). In nine supportive cases, safety margins were endorsed in principle but lacked detailed documentation. C-QT modelling is widely accepted by EMA when adequately supported. However, gaps in exposure justification and reporting continue to challenge regulatory confidence, emphasizing the need for standardized practices in QT risk assessment.

The pharmacokinetic literature is rich in aggregated concentration data that contain valuable information, yet tools to extract this information remain limited. This work introduces distributional physics-informed neural networks (D-PINNs), a novel algorithm designed to enable statistical modelling within the PINN framework, allowing recovery of pharmacokinetic parameter distributions at the population level from published concentration means and variances. Unlike traditional PINNs, which often focus on point estimates, D-PINNs incorporate distributional assumptions directly into the optimisation process. The framework utilises neural networks for predicting the mean and variance of the concentration over time. These predictions are then incorporated into a sampling-based procedure within the residual network, which uses the governing ordinary differential equation (ODE) system to compute the physics-informed loss term. The methodology accounts for both interindividual variability through the parameter distribution and measurement noise through a residual error model. The capability of D-PINNs to infer population-level parameter distributions from concentration summary statistics was demonstrated through a simple proof-of-concept using simulated data from a one-compartment pharmacokinetic model of intravenous drug administration. The model achieved high accuracy in estimating both the parameter distribution and the residual error. Hyperparameter tuning highlighted important aspects of model development. The modelling framework was then applied to real-world data to demonstrate its ability to recover information on the distribution of kinetic parameters in the studied population. Specifically, a minimal physiologically-based pharmacokinetic (mPBPK) model for monoclonal antibodies (mAbs) was fitted to aggregated plasma concentration data reported in the literature using D-PINNs. The same aggregated data were also analysed using a Markov chain Monte Carlo (MCMC) analogue to benchmark the proposed methodology.

PK-QTc analyses are routinely done as part of most drug development programs. Usually, the PK concentration of a single compound is related to the QTc effect. However, in many instances there are several active compounds, for example a parent drug and its metabolite, or combination drugs. Previous authors have shown that doing separate PK-QTc analyses for each of the potentially active compounds may lead to biased results, and recommended to do joint modeling of the impact of both compounds on the corrected QT interval. In this paper we go one step further and propose a formal hypothesis test to exclude a [Formula: see text]msec effect based on a joint modeling approach when there are potentially two active compounds. In analogy to the situation with just one active compound, where the upper limit of a [Formula: see text]% confidence interval for [Formula: see text] (with [Formula: see text] being the slope of a linear exposure-response relationship and [Formula: see text] being the expected maximum concentration of some supra-therapeutic dose) needs to be below [Formula: see text]msec, we use the upper confidence intervals for [Formula: see text], [Formula: see text], and [Formula: see text] and exclude a [Formula: see text]msec effect if all three upper confidence limits are below the [Formula: see text]msec threshold. We propose a bootstrap approach for decision making, and show via simulations that this approach controls the type I error of [Formula: see text]%. We focus on the situation where exposure-response is linear in both compounds, but also indicate how this can be extended to non-linear situations.

Phase I single and multiple ascending dose studies are more and more often used to evaluate QT liability of new drugs. However, these studies are not primarily tailored to concentration-QT analysis and to control or document influential factors such as meal intake. In addition, sampling times may vary over the day for operational reasons. This simulation analysis evaluates the reliability of the standard pre-specified linear model (PLM) proposed by a publication of Garnett et al. and an adjusted PLM accounting for food effect and clock time. The QTcF-time profile of a drug with a mild QT-liability (upper bound of the 90% confidence interval close to the 10 ms threshold) resulting from a well-controlled study was simulated 1000 times and evaluated with the unadjusted PLM (Scenario A, negative rate: 20.8%). Compared to suboptimal study designs with uncontrolled and unbalanced (i.e., differences between active treatment and placebo) differences in meal intake and dosing/sampling times, the unadjusted PLM led to an inflated negative rate (≤ 50%), while the adjusted PLM was able to correct for the imbalances resulting in similar negative rates as the reference scenario or lower, i.e., being more conservative. In conclusion, good documentation in Phase I trials and adjusting for known influential factors can help to analyze QT effects reliably and waive with relevance QT/QTc studies.