Bayesian Meta-analytic Pharmacokinetic Modeling and Pharmacokinetic/Pharmacodynamic-Integrated Clinical Trial Designs in Oncology

Shun Rao (Advisor: Yisheng Li, PhD)

Clinical trials play a critical role in oncology drug development by evaluating the safety and efficacy of investigational therapies in humans. While pharmacokinetic (PK) and pharmacodynamic (PD) information is routinely collected in early- and late-phase clinical studies, these data are often analyzed separately from primary clinical outcomes and are underutilized in statistical inference and trial design. Recent advances in model-informed drug development (MIDD) have highlighted the potential of mechanistic PK/PD models to improve the efficiency and scientific rigor of clinical drug development. However, existing statistical methods primarily focus on dose-finding trials, with relatively limited work integrating PK/PD information into treatment effect evaluation and decision-making across different phases of clinical oncology drug development.

This dissertation develops Bayesian methods that incorporate PK/PD information across three distinct stages of clinical oncology drug development: PK parameter estimation from published literature, phase II trial design, and post-marketing regimen optimization. In Chapter 2, we develop a three-level Bayesian hierarchical meta-analysis framework to estimate mechanistic model parameters using published summary statistics from early-phase studies of subcutaneously administered monoclonal antibodies. The proposed framework enables inference for key PK parameters despite the absence of individual-level data and provides informative prior distributions for downstream PK/PD modeling and trial designs. In Chapter 3, we extend a semi-mechanistic Bayesian modeling (SMB) framework for treatment effect evaluation in randomized phase II oncology trials. The proposed method jointly models PK exposure, PD biomarker response, and clinical outcomes, allowing mechanistic pharmacological information to directly inform inference on treatment effect. Simulation studies demonstrate improved statistical power compared with conventional approaches while maintaining Type I error control. In Chapter 4, we further extend the SMB framework to post-marketing treatment regimen optimization. Using ceritinib as a motivating example, the SMB-based post-marketing trial design integrates supplementary PK/PD information from pre-approval PK studies and food-effect evaluations to identify the optimal dose and regimen. Our simulation study demonstrates that the proposed design using a quasi-likelihood approach to modeling ordinal toxicity scores achieves improved statistical efficiency for regimen optimization, requiring fewer patients than a conventional comparative design.

Collectively, this dissertation addresses a common methodological gap: mechanistic PK and PD information that is routinely generated throughout drug development is rarely incorporated into the primary inferential framework of clinical trials. By demonstrating that this information can be recovered from aggregate published evidence, embedded within phase II efficacy evaluation, and leveraged for post-marketing regimen optimization within a coherent SMB paradigm, this dissertation provides a foundation for a more integrated approach to MIDD in which pharmacological knowledge systematically contributes to statistical inference, trial design, and decision-making across the full drug development continuum.

Advisory Committee:

• Yisheng Li, PhD, Chair
• Xuelin Huang, PhD
• Gabriel Lopez-Berestein, PhD
• Jing Ning, PhD
• Haitao Pan, PhD

Join via Zoom (Please contact Ms. Rao for her Zoom meeting info.)