CPT: Pharmacometrics & Systems Pharmacology最新文献

Exposure–response analyses are central to dose selection in drug development. The estimand framework, formalized in ICH E9(R1) regulatory guidance, provides a structured approach to define scientific objectives with precision. We apply the estimand framework to dose–exposure–response analyses. For simulated example studies inspired by real-world scenarios, we define dose–response estimands of clinical interest. The estimands are formalized using the potential outcome notation. Assumptions on the setup of the studies and the relation between treatment, exposure and response are expressed as a directed acyclic graph (DAG). The estimand is transformed using the assumption into expressions to identify the estimand based on the observed data. Three types of expressions are obtained. First, a pooled dose–exposure–response (DER) analysis that corresponds to a standard DER analysis as executed for many projects. Second, a pooled, covariate adjusted dose–response (DR) analysis, and third summaries of the outcomes in each randomized cohort. In our example, DER provides more precise estimates than DR as judged by the mean square error (MSE) of repeated simulation estimation. This work advances methodological rigor in DER analyses by integrating with causal inference methodologies and the estimand framework, enabling clearer interpretation of modeling assumptions and results. This has important concrete advantages. We obtain different estimation methods for the same estimand that may be compared to validate them. The potential for bias in the different estimation methods can be formally assessed. The proposed approach provides a generalizable strategy to improve exposure–response analyses for dose selection, particularly when the relevant evidence includes data from multiple studies.

Pediatric physiologically-based pharmacokinetic (PBPK) modelling plays an increasing role in selecting doses in children and addressing clinical pharmacology questions. Ethical concerns often limit clinical pharmacology studies that have no direct therapeutic benefit in children, highlighting the value of PBPK model predictions. However, regulatory acceptance of pediatric PBPK models remains limited because of uncertainties in system-specific information and inadequate model qualification. Ambiguous ontogeny data of drug metabolizing enzymes (DME) and transporters are recognized as significant obstacles to the accurate pharmacokinetics (PK) prediction in children and the leading cause of insufficient pediatric PBPK model qualification. To address this challenge, a population PBPK modeling approach is proposed. This method is analogous to whole-body PBPK modeling and allows the estimation of DME/transporter ontogenies using sparse PK data collected from children and adults by nonlinear mixed-effect modeling. Well-characterized ontogeny functions of key DME/transporters enhance the extrapolation ability of PBPK models and facilitate model-informed drug development (MIDD) in children. This article proposes a strategy for pediatric PK extrapolation using population PBPK modeling, illustrated through the case example of risdiplam, approved for the treatment of spinal muscular atrophy. The ontogeny modeling, extrapolations of PK to unstudied pediatric populations, and drug–drug interaction (DDI) risk assessment are also discussed. The population PBPK modeling approach is intended to address the inconsistencies in ontogeny data and augment PBPK modeling for quantitative clinical pharmacology assessments in children. It will accelerate optimal dose finding and provide guidance for adequate use of drugs in pediatric patients, which is especially important for developing treatments for progressive pediatric rare diseases.

We introduce the Regularized Horseshoe (RHS) in the context of covariate selection for population PK/PD models. Unlike stepwise approaches which are commonly used in this context, the RHS can simultaneously assess all possible parameter-covariate relationships in a single model fit by leveraging the fact that such relationships are usually sparse in practice. Furthermore, the RHS avoids the over-estimation of effect sizes that commonly occurs with stepwise approaches and avoids overfitting by averaging over the posterior uncertainty of possible parameter-covariate relationships. This leads to improved predictive performance on held-out data. We first give an overview of common covariate selection methods for population PK/PD modeling, then we define the RHS and provide intuition for how the method works. We then provide Stan code and a set of hyperparameters applicable to general population PK/PD models that can readily be applied by practitioners. Using an extensive simulation study, the beneficial properties of the RHS are illustrated and compared to popular covariate selection methods that are commonly used on population PK/PD models. Lastly, we compare the RHS to other commonly used methods on four real-world PK/PD datasets and illustrate its superior predictive performance on held-out data.

Understanding immunogenicity is crucial to improving therapeutic protein development. By comparing Phase I to III antidrug antibody (ADA) incidence data of Roche-internal and approved monoclonal antibodies, we demonstrate a bias toward lower ADA incidence in published data, partly because ADA data from early trials—often discontinued for reasons related to high immunogenicity—are rarely published. Through an empirical model, we show how this bias affects ADA incidence time-course predictions, underscoring the need for cross-industry transparent data reporting.

The investigational antigen-specific immunotherapy mRNA-4359 is a lipid nanoparticle–encapsulated mRNA-based immunotherapy that encodes for the immunogenic indoleamine 2,3-dioxygenase (IDO) and programmed death-ligand 1 (PD-L1) antigens. An ongoing first-in-human (FIH) phase 1/2 clinical trial (NCT05533697) will evaluate the safety and antitumor activity of mRNA-4359 when administered alone and in combination with the anti–programmed death-1 agent pembrolizumab in participants with advanced solid tumors. The current analysis applied a novel immunostimulatory/immunodynamic (IS/ID) modeling approach to determine a plausible starting dose of mRNA-4359 for the FIH trial. The model used for the FIH dose prediction was calibrated to previously published clinical trial data obtained for an immunomodulatory peptide-based vaccine activating IDO- and PD-L1–specific T cells in patients with metastatic melanoma. The analysis found that a 180 μg dose of mRNA-4359 would possibly elicit a T-cell response similar to a 200 μg dose of the peptide-based vaccine with a range of 45–360 μg, assuming a potential 4-fold higher to 2-fold lower efficiency (the ability to elicit IFN-γ secreting T cells, indicative of cytotoxic potential). Model simulations further predicted that a 15-cycle every 3 weeks regimen of mRNA-4359 could be expected to provide longer responses than other feasible simulated regimens. Finally, the IS/ID modeling analysis determined that a 100 μg dose of mRNA-4359 would be the most appropriate starting dose for FIH trials. The described approach represents a unique application of IS/ID modeling to determine a therapeutically relevant FIH starting dose in the absence of supporting preclinical animal data.

This work investigates how correlations between covariates influence the estimation of their effects in pharmacometric models. The focus is on quantifying the impact on conditional and unconditional covariate effect estimates and assessing the consequences for model interpretation, communication, and dosing recommendations. A theoretical framework was used to describe the mathematical relationship between conditional and unconditional coefficients. This was verified by simulations across a wide range of covariate correlation strengths and relative covariate effect sizes. The practical consequences of misinterpreting conditional effects were evaluated in the context of dose selection and a priori dose individualization. As predicted by theory, covariate correlation had a substantial effect on the conditional covariate coefficient estimates, while unconditional estimates remained stable. Interpreting conditional covariate effects in isolation led to incorrect conclusions about dosing needs and introduced bias and imprecision in individual dose predictions. In contrast, both the complete conditional model and the unconditional model gave accurate predictions when applied appropriately. Unconditional covariate effects offer greater interpretability, making them more suitable for communicating individual covariate impacts in drug labels, publications, and forest plots. We demonstrate that conditional effects are highly sensitive to model context and covariate correlation, making them poor proxies for the unconditional effect, which is often the quantity of interest for dosing and communication. To minimize misinterpretation, unconditional effects should be reported when describing the influence of individual covariates, while the complete conditional model should be used for simulations and exposure predictions. This dual approach can improve clarity and reduce the risk of misunderstanding in model-informed decision-making.

Vildagliptin (VDG) is a dipeptidyl-peptidase-4 (DPP-4) inhibitor used for type 2 diabetes (T2DM) treatment. Viewing to improve VDG treatment, a population pharmacokinetic (popPK) model was built to describe drug plasma, free liver and muscle concentrations determined by microdialysis in healthy and diabetic animals following 50 mg/kg i.v. bolus administration. A four-compartment popPK model with linear elimination and bidirectional transport between tissues and the central compartment described the data with diabetes as a covariate in Q₁ and Q_out,liver. The pharmacokinetic parameters of VDG were scaled to humans using allometry, and used to simulate VDG tissue concentrations in patients with T2DM and relate them with the DPP-4 inhibition by an I_max model. The efficacy of VDG was evaluated considering 80% and 92% DP-IV inhibition during the entire dosing interval. VDG 100 mg q24 h achieved 80% DPP-4 inhibition in plasma, but not in tissues. Although q12 h dosing interval reached 80% enzyme inhibition in plasma for > 25 mg doses, only the 100 mg reached this goal in muscle. The 92% enzyme inhibition was achieved in plasma for 50 and 100 mg q12 h but none of the dose regimens investigated reached this inhibition in tissues.

To characterize the relationship between cendakimab exposure and the longitudinal efficacy endpoint dysphagia days (DD), E–R analyses were performed using data from the EE-001 study (N = 427) with eosinophilic esophagitis. DD—a bounded, discrete endpoint assessed over 14-day period via modified daily symptom diary (mDSD)—was modeled using a latent variable indirect response (IDR) model coupled with a combined uniform-binomial (CUB) distribution. The latent variable, representing the underlying disease status, was dynamically modulated by placebo and drug effects (a function of individual-predicted exposure) to govern the binomial probability of DD, while the uniform component captured the residual variability in patient-reported outcomes. Inter-individual variability was estimated for baseline DD, maximum placebo effect, and maximum drug effect. Covariates, including steroid inadequate response or intolerance (Steroid IR/I) status and baseline DD, were incorporated in the final model based on the clinical relevance. The estimated placebo half-life was ~28 weeks, estimated EC₅₀ was 76.5 μg/mL, corresponding to an EC₉₀ of ~688 μg/mL, indicating steepness of the E_max curve. Model-based simulations showed that both 360 mg QW and QW-to-Q2W regimens reduced DD compared to placebo at Week 48, with mean reductions of ~1.65 and ~1.36 days, respectively. Covariate-stratified simulations suggested consistent responses across sex, age, and race. Steroid IR/I and baseline DD influenced treatment response magnitude but did not warrant dose modification. These findings support QW-to-Q2W as an effective maintenance posology and the utility of latent variable IDR models with appropriate likelihoods for modeling bounded, discrete longitudinal endpoints in E–R analyses.

Aspirin is one of the most commonly used medications in pregnancy, particularly for the prevention of hypertensive disorders. Despite aspirin's widespread use in pregnancy for preeclampsia prevention, its pharmacokinetics (PK) across all trimesters remain poorly characterized, complicating optimal dosing recommendations. To develop a pregnancy-specific physiologically based pharmacokinetic (PBPK) model for aspirin that could be individualized to patient-specific parameters, illustrating differences in aspirin PK across the different trimesters of pregnancy. A PBPK model was developed using GastroPlus (a mechanistically driven simulation software) for nonpregnant and pregnant people at each trimester of pregnancy. The nonpregnant PBPK model was first established and validated against existing data from healthy adult volunteers. Once validated, the model was adapted for pregnant people and verified using observed pharmacokinetic profiles. The simulated PK parameters of aspirin in pregnant and nonpregnant women closely matched the clinical observations reported in the literature, with fold errors ≤ 1.04 (less than 1.5 is considered an acceptable simulation model). The predicted systemic exposure (AUC_0-24h) of salicylic acid (SA), the active metabolite of aspirin decreased throughout gestation, showing a reduction of approximately 20% at 10 weeks and 30% at 40 weeks. An increase in clearance was observed as gestation progressed. The model predicted a modest decrease of 10% in systemic exposure in pregnant women and a 20% increase in fetal exposure to SA as pregnancy progresses. A PBPK model using GastroPlus was developed to describe the PK and pharmacodynamics of aspirin in both pregnant and nonpregnant healthy adults.

Quantitative systems pharmacology (QSP) models offer a useful platform to integrate drug pharmacology with knowledge about biological mechanisms across multiple scales and data sources into a unified quantitative framework. This makes them invaluable to address many relevant questions in drug research and development. Despite their potential, however, QSP models are seldom employed in the population analysis context due to their complexity and dimensionality. Model order reduction (MOR) techniques can be used to tackle this challenge. However, a single MOR technique might not be sufficient to achieve an applicable reduced model. Furthermore, to date there is no tool to judge whether the reduced model retains important mechanistic features of the original model. In this tutorial, we present a workflow employing index analysis that guides the selection and combination of MOR techniques and includes a check of the preservation of important mechanistic features by the reduced model. To demonstrate the value of the proposed approach, we first explain the concepts in the context of a small-scale example model and then expand to a well-known large-scale QSP model—the blood coagulation model.