CPT: Pharmacometrics & Systems Pharmacology最新文献

This study demonstrates the application of a model based meta analysis (MBMA) framework to characterize the safety and efficacy profiles of therapies in relapsed and refractory multiple myeloma (RRMM). Published clinical trial data were analyzed to evaluate the incidence of Grade ≥ 3 neutropenia and overall response rate (ORR), providing a quantitative foundation for model-informed drug development. The final model incorporated trial- and treatment-level covariates and was evaluated using visual predictive checks and predictive simulations. Results revealed increased neutropenia risk associated with alkylating agents and higher ORR in regimens with background corticosteroids and in patients with only one prior line of therapy. MBMA-derived estimates facilitated systematic comparisons across regimens, accounting for heterogeneity in trial design and populations. The MBMA estimates can also support benchmarking of internal regimens against current standards. A quantitative systems pharmacology (QSP) model, developed in parallel, was also used to simulate patient responses across a broad array of RRMM treatments, including novel combinations involving T-cell engagers (TCEs) and CELMoD agents. Trial-calibrated virtual patients and a classifier for prior therapy exposure enabled the prediction of regimen-specific ORR across different treatment histories. Together, the MBMA-informed and QSP-supported modeling strategy enabled a comprehensive benefit–risk assessment by combining statistical estimation with mechanistic simulation. This coordinated approach enhances clinical decision-making by enabling comparison of novel or investigational therapies to the evolving treatment landscape, particularly in the absence of head-to-head trials.

Multiple sclerosis (MS) is a chronic disorder that typically shows accumulation of disability, affecting the ability to work. The disease severity is usually graded by physicians with the expanded disability status scale (EDSS) in the clinic, but patient-reported outcome questionnaires are also available, like the Multiple Sclerosis Impact Scale (MSIS-29) or Fatigue Scale for Motor and Cognitive Functions (FSMC). The aim of this work was to investigate the quantitative link between disease severity and the number of days with MS-related sickness benefits from registry data (the Swedish MS registry and the Swedish Social Insurance Agency's Micro Data for Analyzes of Social Insurance registry). An item response theory model for the disability was built, linking the EDSS, MSIS-29, and FSMC to the same underlying disease construct through five correlated latent variables. A Markov state model for the level of sickness benefits was also developed, in which the disease severities from the disability model were tested as covariates, on top of age. The latent variable for EDSS was the most important predictor of work ability. Patients with low disability (EDSS < 3) hardly had any sickness benefit days, while patients with severe disability (EDSS ≥ 6) were found to spend over 50% of their time with sickness benefits. Physical aspects of the disease were found to be more important than psychological aspects in predicting work ability. This underlines the patient-specific nature of MS, and the need for predictive models such as these to evaluate treatment effects, make risk assessments, and calculate societal and individual costs.

Generative Artificial Intelligence (AI) frameworks, such as Variational Autoencoders (VAEs), have proven powerful in learning structured representations from complex, high-dimensional data. In pharmacometrics (PMX), nonlinear mixed effects (NLME) modeling is widely used to capture inter-individual variability and link covariates to characterize parameters with the goal of informing key decisions in drug research and development. This research combines the strengths of both approaches by introducing a VAE framework specifically designed for NLME modeling. The proposed method integrates the flexibility of generative AI with the interpretability and robustness of mechanism-based PMX modeling. To advance covariate selection in PMX, we replace the Evidence Lower Bound objective in VAEs with an objective function based on the corrected Bayesian information criterion. This enables the simultaneous evaluation of all potential covariate-parameter combinations, thereby allowing for automated and joint estimation of population parameters and covariate selection within a single run. Manual selection and repeated model fitting across covariate combinations are no longer required. We demonstrate the effectiveness of this combined AI-PMX approach with two representative cases. As the first generative AI-based optimization method for NLME modeling, the VAE achieves high-quality results in a single run, outperforming traditional stepwise procedures in terms of efficiency. As such, the presented approach facilitates automated model development, advancing PMX and its applications in model-informed drug development.

Artificial intelligence (AI) is increasingly being explored as a tool to support pharmacometric modeling, particularly in addressing the coding challenges associated with NONMEM. In this study, we evaluated the ability of seven Large Language Models (LLMs) to generate NONMEM codes across 13 pharmacometrics tasks, including a range of population pharmacokinetic (PK) and pharmacodynamic (PD) models. We further developed a standardized scoring rubric to assess code accuracy and created an optimized prompt to improve LLM performance. Our results showed that the OpenAI o1 and gpt-4.1 models achieved the best performance, both generating codes with great accuracy for all tasks when using our optimized prompt. Overall, LLMs performed well in writing basic NONMEM model structures, providing a useful foundation for pharmacometrics model coding. However, user review and refinement remain essential, especially for complex models with special dataset alignment or advanced coding techniques. We also discussed the applications of AI in pharmacometrics education, particularly strategies to prevent overreliance on AI for coding. This work provides a benchmark for current LLMs' performance in NONMEM coding and introduces a practical prompt that can facilitate more accurate and efficient use of AI in pharmacometrics research and education.

Pirtobrutinib is a reversible Bruton tyrosine kinase (BTK) inhibitor. In vitro, pirtobrutinib is metabolized by cytochrome P450 (CYP) 3A4 and uridine 5′-diphosphoglucuronosyl transferases (UGTs) and causes reversible and time-dependent inhibition and induction of CYP3A4. Coadministration of itraconazole, a strong CYP3A4 inhibitor, with pirtobrutinib in healthy human subjects, resulted in a pirtobrutinib area under the plasma concentration-time curve (AUC) ratio of 1.49, while rifampin, a strong CYP3A4 inducer, decreased pirtobrutinib AUC by 71%. Oral administration of pirtobrutinib 200 mg once daily (QD) increased the AUC of oral and intravenous midazolam by 1.70- and 1.12-fold, respectively. A physiologically based pharmacokinetic (PBPK) model was developed for pirtobrutinib using physicochemical properties, in vitro data, and clinical pharmacology study results. The PBPK model captured the clinically observed interactions for itraconazole, rifampin, and midazolam, with predicted pirtobrutinib and midazolam AUC ratios within 0.91- to 1.16-fold of observed. The model predicted 1.20- to 1.73-fold increases in the pirtobrutinib AUC with strong and moderate CYP3A4 inhibitors. Furthermore, the predicted pirtobrutinib AUC ratios were within 0.51–0.86 with moderate and weak CYP3A4 inducers. The predicted effects of CYP3A4 modulators on pirtobrutinib pharmacokinetics, together with the known exposure-response relationships for safety and efficacy in patients with hematological malignancies, were used for recommending appropriate dosing regimens during coadministration.

Quantitative Systems Pharmacology (QSP) is a powerful approach to provide decision-making support throughout the drug development process. QSP comes with many challenges in model development, validation, and applications. Traditional QSP workflows are limited by slow knowledge integration, labor-intensive model construction, inconsistent validation practices, and restricted scalability. In this work, we introduce QSP-Copilot, the first end-to-end AI-augmented solution designed to improve QSP modeling workflows by integrating a multi-agent system utilizing large language models (LLMs). QSP-Copilot provides modular support from project scoping and model structuring to model evaluation and reporting. Through the automation of routine tasks, QSP-Copilot reduces model development time by approximately 40% and improves methodological transparency through systematic documentation of literature sources and modeling assumptions. We demonstrate QSP-Copilot's application for two rare diseases of blood coagulation and Gaucher disease. In the blood coagulation case, automated extraction from ten peer-reviewed articles yielded 179 biological entity interaction pairs; out of these, only 105 unique mechanisms were retained after standardization. For Gaucher disease, screening nine articles produced 151 pairs, which were consolidated into 68 distinct biological interactions following the same post-processing workflow. The extraction precision for blood coagulation and Gaucher disease is 99.1% and 100.0%, respectively. QSP-Copilot extractions can be incorporated into effect diagrams with minimal expert filtering, significantly reducing the manual curation burden. The integration of AI-augmented workflows like QSP-Copilot represents a pivotal shift toward enhanced scalability and impact for QSP across the drug development pipelines, especially in disease areas where biological knowledge is sparse, such as rare diseases.

Accurate prediction of human pharmacokinetics (PK) for lead compounds is one of the critical determinants of successful drug development. Traditional methods for PK parameter prediction, such as in vitro to in vivo extrapolation and physiologically based pharmacokinetic modeling, often require extensive experimental data and time-consuming calibration of parameters. Machine learning (ML) has been widely applied to predict ADME and physicochemical properties (ADMEP descriptors), but studies focusing on concentration-time (C-t) profile prediction remain limited. In this study, we developed a Long Short-Term Memory (LSTM) based ML framework to predict C-t profiles following intravenous (IV) bolus drug administration in humans. The model used ADMEP descriptors generated by ADMETlab 3.0 and dose information as input. A total of 40 drugs were used for training and 18 for testing, with concentration data simulated from published PK models. Our approach achieved R² of 0.75 across all C-t profiles, and 77.8% of C_max, 55.6% of clearance, and 61.1% of volume of distribution predictions within a 2-fold error range, demonstrating predictive performance comparable to previously published ML methods. Furthermore, model performance was found to be associated with the input dose level and ADMEP descriptors, suggesting the accuracy and confidence of the prediction may be expected in advance via these descriptors. This LSTM-based framework using a small number of compounds enables efficient prediction of human PK profiles with IV dosing, offering a practical alternative to traditional PK prediction models. It holds promise for improving early-phase prioritizing lead compounds and reducing reliance on animals in drug development.

In vitro–in vivo correlation/relationship (IVIVC/R) models such as physiologically based biopharmaceutics modeling (PBBM) are crucial tools that link biopharmaceutical properties to clinical performance. They accelerate development, reduce costly experimental studies and clinical trials, and justify regulatory decisions for drug formulation related questions of interest (QOI). This paper consolidates insights from academia, industry, and service providers, exploring future opportunities, organizational challenges, regulatory perspectives, and competency gaps for further enhanced application in pharmaceutical development and regulatory decision-making.

The objectives of this study were to quantify the relationship between vaccine-induced immunogenicity responses and the protection against respiratory syncytial virus (RSV) infection-related clinical outcomes, and to evaluate immunogenicity as a surrogate marker for vaccine efficacy (VE) to accelerate RSV vaccine development. Serum neutralizing activity (SNA) and cell-mediated immunity (CMI) may serve as surrogate markers for the protection against RSV infection and are evaluated as immunogenicity endpoints in clinical trials of RSV vaccine candidates. Two meta-analytical approaches were applied to data from seven randomized placebo-controlled clinical trials that investigated RSV vaccines in older adults. The primary analysis examined the relationship between SNA and VE across three different clinical severity levels: (1) acute respiratory infection, (2) RSV lower respiratory tract disease (LRTD) with ≥ 2 clinical symptoms, and (3) RSV LRTD with ≥ 3 clinical symptoms (LRTD 3+). Furthermore, the additional contribution of CMI to VE, after accounting for the effect of SNA, was explored in a secondary analysis. The results demonstrated a positive correlation between SNA and VE across three clinical severity levels. Higher CMI was associated with higher VE specifically for RSV LRTD 3+, the most severe clinical level, suggesting that CMI may be correlated with additional clinical benefits in mitigating the severity of RSV infection. These findings provided preliminary evidence for immune correlates of protection against RSV infection and may aid in accelerating the development of new RSV vaccines.

Classic and hybrid pharmacometric-machine learning models (hPMxML) are gaining momentum for applications in clinical drug development and precision medicine, especially within the oncology therapeutic area. However, standardized workflows are needed to ensure transparency, rigor, and effective communication for broader adoption. In this tutorial, we review pharmacometric (PMx) and machine learning (ML) reporting standards and evaluate them against hPMxML works in oncology contexts as a motivational example to identify current deficiencies and propose mitigation strategies for future efforts. The revealed gaps include insufficient benchmarking, absence of error propagation, feature stability assessments, and ablation studies, limited focus on external validation and final parametrization, and discrepancies between the performance metrics chosen and the original clinical questions. To address these, we propose a checklist for hPMxML model development and reporting, consisting of steps for estimand definition, data curation, covariate selection, hyperparameter tuning, convergence assessment, model explainability, diagnostics, uncertainty quantification, validation and verification with sensitivity analyses. This standardized approach is expected to enhance the reliability and reproducibility of hPMxML outputs, enabling their confident application in oncology clinical drug development, while fostering trust among all stakeholders.