Journal of Pharmacokinetics and Pharmacodynamics最新文献

Toxicokinetic and pharmacokinetic (PK) summary parameters, such as C_max (peak concentration), AUC (time-integrated area under the plasma concentration curve), and t_1/2 (elimination half-life from the body), are important information for understanding chemical safety in both pharmaceuticals and commercial industry. Although standardized tools exist for PK analysis of individual chemicals, new workflows can enhance chemoinformatic trend analysis. The Concentration versus Time Database (CvTdb) is a public repository of PK data at the U.S. Environmental Protection Agency (EPA). The CvTdb contains manually curated, standardized toxicokinetic data from hundreds of publications. Experimental time-course data of chemical concentrations in body fluids and tissues are extracted along with descriptive metadata. The advantage of standardized data is that it can be analyzed systematically. For example, we observe that 88.6% of replicate measurements of blood or plasma concentrations of chemicals after intravenous or oral dosing are within two-fold of the mean concentration. Although most experimental data have final timepoints within three days, some experiments extend up to a year, usually for long-lived chemicals. Here we have estimated PK parameters of CvTdb data using a custom R package, invivoPKfit. Standardized 1- and 2- compartmental PK model parameters were estimated using all data associated with a particular compound, including data that spans multiple references. We used invivoPKfit to estimate PK parameters such as volume of distribution (V_d) and t_1/2. The parameter values estimated with invivoPKfit are distributed similar to estimates made in the literature by a variety of methods. Overall, CvTdb serves as a standardized set of open data and for calibrating and evaluating PK models, while invivoPKfit allows for batch processing of this data type in a transparent and scalable manner. In addition to scientific insights, chemical risk assessment may be better informed by transparent, reproducible, and open-source workflows for PK informatics.

Two recent papers offer contrasting perspectives on integrating Quantitative Systems Pharmacology (QSP) and Artificial Intelligence/Machine Learning (AI/ML): one views QSP as the primary driver using AI/ML to enhance computational tasks, while the other argues that AI/ML should provide an alternative mechanistic framework. Rather than perpetuate this tension, we used Large Language Models (LLMs) to examine both papers in two tests-one comparing their core arguments and another probing which methodology LLM should take precedence. Repeating each test multiple times with an identical and neutral prompt, the LLM revealed that each perspective suits specific stages of the drug development pipeline. QSP offers mechanistic rigor and regulatory clarity, and AI/ML excels in high-dimensional data analysis and exploratory modeling. A hybrid approach might best serve researchers and decision-makers, especially when harmonizing data-driven insights with mechanistic integrity. This exercise also highlights the potential of LLMs as promising tools for synthesizing complex information, offering an arguably less biased viewpoint that can trigger deeper discussion from the broader community seeking to align QSP and AI/ML in model-informed drug development (MIDD). By combining our human expertise with AI-driven analyses, we hope to further discuss with the scientific community how QSP and AI/ML-and the synergy between them-can drive innovation in therapeutic discovery and optimization.

Standard pharmacodynamic models are ordinary differential equations without the features of stochasticity and heterogeneity. We develop and analyse a stochastic model of a heterogeneous tumour-cell population treated with a drug, where each cell has a different value of an attribute linked to survival. Once the drug reduces a cell's value below a threshold, the cell is susceptible to death. The elimination of the last cell in the population is a natural endpoint that is not available in deterministic models. We find formulae for the probability density of this extinction time in a collection of tumour cells, each with a different regulator value, under the influence of a drug. There is a logarithmic relationship between tumour population size and mean time to extinction. We also analyse the population under repeated drug doses and subsequent recoveries. Stochastic cell death and division events (and the relevant mechanistic parameters) determine the ultimate fate of the cell population. We identify the critical division rate separating long-term tumour population growth from successful multiple-dose treatment.

The objective of this work was to develop a translational physiologically-based pharmacokinetic (PBPK) model for antibody-drug conjugates (ADCs), using monomethyl auristatin E (MMAE)-based ADCs. A previously established dual-structured whole-body PBPK model for MMAE-based ADCs in mice was scaled to higher species (i.e., rats and monkeys) and humans. Species-specific physiological and drug-related parameters for the payload and antibody backbone of ADCs were obtained from literature. Parameters associated with payload release, including the deconjugation rate, were optimized using an allometric scaling approach, and antibody degradation rate was adjusted to account for the enhanced clearance of ADCs due to conjugation across different species. The translational PBPK model predicted the PK profiles for various ADC analytes in rats, monkeys, and humans reasonably well. The optimized PBPK model suggested decreased rate of deconjugation for ADCs in higher species, whereas the effects of payload conjugation on ADC clearance were more pronounced in higher species and humans. The translational PBPK model presented here may enable prediction of different ADC analyte PK at the site-of-action, offering valuable insights for the development of exposure-response relationships for ADCs. The modeling framework presented here can also serve as a platform for the development of PBPK model for other ADCs.

To evaluate the role of diffusion process dimensionality in drug absorption following subcutaneous or intramuscular administration. The diffusion dimensionality model is based on analytical solutions of the 1-, 2- or 3-dimensional diffusion equations for a bolus input linked to a central compartment with first-order elimination. The model equations were reparameterized to contain three parameters for the time needed for the drug diffusion from the administration site, drug absorption into the central compartment, and the elimination rate constant. The diffusion dimensionality models were challenged with previously published subcutaneous absorption data for 13 antibody drugs and insulin lispro, and the long-acting injectable antipsychotic drugs: subcutaneous Perseris™, intramuscular Invega Sustenna®, Risperdal Consta®, and olanzapine. The Bayesian information criterion was used for model selection. The solution to the diffusion equation for a bolus dose administration is strongly dependent on the number of dimensions. The maximal concentration is lowest for the 3-dimensional diffusion equation. The pharmacokinetic profiles of all 13 antibodies were satisfactorily approximated by a diffusion dimensionality model. Three antibodies (CNTO5825, ACE910 and ustekinumab) were best described by the 2-dimensional diffusion equation. The 2- and 3-dimensional diffusion equations were equivalent for ABT981, guselkumab, adalimumab, nemolizumab, omalizumab, and secukinumab. Golimumab, DX2930, AMG139, and mepolizumab were best described by the 3-dimensional diffusion equation. All the long-acting antipsychotic dosage forms except Risperdal Consta were modeled satisfactorily. Diffusion dimensionality models are a parsimonious and effective approach for modeling drug absorption profiles of subcutaneously and intramuscularly administered small molecule and protein drugs and their dosage forms.

The human milk/plasma (M/P) drug concentration ratio is crucial in pharmacology, especially for breastfeeding mothers undergoing treatment. It determines the extent to which drugs ingested by the mother pass into breast milk, potentially affecting the infant. This study conducted a comprehensive evaluation of multiple machine learning algorithms to assess their effectiveness in predicting the M/P ratio. The dataset consists of 162 drugs and 11 predictor variables. M/P ratios were categorized into two groups of (0, 1) and (≥ 1), and a refined three-category system: (0, < 0.5), (0.5, < 1), and (≥ 1). The ML techniques utilized include K-Nearest Neighbors (KNN), Random Forest, Support Vector Machine (SVM), and Neural Networks. We implied the five-fold cross-validation to ensure the model's robustness and Principal Component Analysis (PCA) was applied for data visualization. Bayesian Information Criterion (BIC) was used in the KNN model selection to balance complexity and explanatory power. In our study, KNN achieved average accuracies of 79% for the two-category system and 60% for the three-category. Random Forest models show 77 and 64% average accuracy, respectively. SVM achieved similar results with 78 and 67%, while Neural Networks have the overall best result among the other models with average accuracies of 82 and 76% accuracy. The study highlights the potential of machine learning (ML) techniques in predicting M/P ratios, offering valuable insights for risk assessment during drug development. These predictive models can serve as a valuable tool for estimating drug transfer into breast milk, helping to bridge knowledge gaps in drug safety for lactating individuals. Further validation and refinement by incorporating larger datasets can enhance their reliability and applicability. Advancing these techniques can support safer medication use and informed clinical decision-making for lactating individuals.

This study presents a translational modeling framework designed to predict tumor size dynamics in cancer patients undergoing anticancer treatment, using data from patient-derived xenograft (PDX) mice. In the first step, a population tumor growth inhibition (TGI) model to estimate the distribution of exponential tumor growth rates and anticancer drug potency in PDX mice was built. Then, model parameters were allometrically scaled from mice to humans to inform a TGI model predicting tumor size dynamics in cancer patients. Longitudinal tumor dynamics predicted by the PDX-informed TGI model were expressed in terms of tumor progression events to allow validation against literature time-to-progression (TTP) data. The proposed approach was tested on two case studies: gemcitabine treatment for pancreatic cancer and sorafenib treatment for hepatocellular cancer. The framework successfully predicted median tumor size dynamics, closely aligned with clinical TTP curves for gemcitabine-pancreatic cancer case study. While predictions for extreme tumor size percentiles highlighted potential avenues for refinement, such as incorporating resistance mechanisms, the overall accuracy underscored the goodness of the approach. For the sorafenib-hepatocellular cancer case study, the framework provided plausible tumor size predictions, with TTP curves closely aligned with clinical observations, despite the limited availability of clinical data prevented a full validation. Overall, the translational modeling framework showed potential for predicting tumor dynamics in cancer patients, with results suggesting its applicability as a valid tool to support early decision-making in oncology.

Modeling interoccasion variability (IOV) of pharmacokinetic parameters is challenging in sparse study designs. We conducted a simulation study with stochastic simulation and estimation (SSE) to evaluate the influence of IOV (25, 75%CV) from numerous perspectives (power, type I error, accuracy and precision of parameter estimates, consequences of neglecting an IOV, capability to detect the 'correct' IOV). To expand the scope from modeling-related aspects to clinical trial practice, we investigated the minimal sample size for IOV detection and calculated areas under the concentration-time curve (AUC) derived from models containing IOV and mis-specified models. The power to correctly detect an IOV increased from one to three occasions (OCC) and the type I error rate to falsely include an IOV was not elevated. Two sampling schemes were compared (with/without trough sample) and including a trough sample resulted in better performance throughout the different evaluations in this simulation study. Parameters were estimated more precisely when more OCCs were included and IOV was of high effect size. Neglecting an IOV that was truly present had a high impact on bias and imprecision of the parameter estimates, mostly on interindividual variabilities and residual error. To reach a power of ≥ 95% in all scenarios when sampling in three OCCs between 10 and 50 patients were required in the investigated setting. AUC calculations with mis-specified models revealed a distorted AUC distribution as IOV was not considered.

About 15-20% of women experience postnatal depression and may seek advice about medication use whilst breastfeeding. Venlafaxine is a potent and selective neuronal serotonin-norepinephrine reuptake inhibitor indicated for treating major depressive disorders. The drug is mainly metabolised by cytochrome P450 2D6 (CYP2D6) to its active metabolite O-desmethylvenlafaxine (ODV), with small contributions from CYP2C9 and CYP2C19. Subsequently, the formed ODV undergoes CYP3A4- and UGT-mediated metabolism and renal excretion. A physiologically based pharmacokinetic (PBPK) model describing the disposition of both venlafaxine and ODV was developed. Consistent with observed data, simulations showed that exposure of the combined active moieties (venlafaxine plus ODV) was similar for both CYP2D6 extensive (EM) and poor metaboliser (PM) subjects. Clinical lactation data for venlafaxine were available from several studies but CYP genotypes were not recorded. Interestingly, based on simulated exposures in breast milk, the estimated average relative infant daily dose (RIDD) ranged from 3.8% for all EMs to 7.6% for all PMs of CYP2D6, CYP2C9 and CYP2C19. Furthermore, simulations in breastfed infants indicated that both CYP polymorphisms and enzyme ontogenies contribute to the significant variability that is observed clinically but the combined exposures of venlafaxine and ODV remain below the thresholds that have been reported for adverse events in adults and children. The data generated here add to the existing knowledge base and can help clinicians and their patients make a more informed decision on the use of venlafaxine during breastfeeding.

Statistical modelling of covariate distributions allows to generate virtual populations or to impute missing values in a covariate dataset. Covariate distributions typically have non-Gaussian margins and show nonlinear correlation structures, which simple descriptions like multivariate Gaussian distributions fail to represent. Prominent non-Gaussian frameworks for covariate distribution modelling are copula-based models and models based on multiple imputation by chained equations (MICE). While both frameworks have already found applications in the life sciences, a systematic investigation of their goodness-of-fit to the theoretical underlying distribution, indicating strengths and weaknesses under different conditions, is still lacking. To bridge this gap, we thoroughly evaluated covariate distribution models in terms of Kullback-Leibler (KL) divergence, a scale-invariant information-theoretic goodness-of-fit criterion for distributions. Methodologically, we proposed a new approach to construct confidence intervals for KL divergence by combining nearest neighbour-based KL divergence estimators with subsampling-based uncertainty quantification. In relevant data sets of different sizes and dimensionalities with both continuous and discrete covariates, non-Gaussian models showed consistent improvements in KL divergence, compared to simpler Gaussian or scale transform approximations. KL divergence estimates were also robust to the inclusion of latent variables and large fractions of missing values. While good generalization behaviour to new data could be seen in copula-based models, MICE shows a trend for overfitting and its performance should always be evaluated on separate test data. Parametric copula models and MICE were found to scale much better with the dimension of the dataset than nonparametric copula models. These findings corroborate the potential of non-Gaussian models for modelling realistic life science covariate distributions.