Panel - Bridging Causal Inference and Pharmacometric Modeling

Causal graphs are employed to introduce the nature of the relationships between baseline patient characteristics, early markers of drug effect, and clinical outcomes. They also help in reasoning about the estimand and the potential impact of intercurrent events (such as drug discontinuation), which may be considered during the assessment of the TGI-OS relationship. Once the estimand is defined, a discussion can happen on the choice of the estimator. Sequential and joint models have been utilized to analyze the relationship between drug-induced tumor growth inhibition (TGI) and overall survival (OS) in oncology clinical development. Sequential models operate under the assumption that TGI precedes and influences OS (which introduces a question of uncertainty propagation), while joint models allow for bidirectional relationships and correct censoring and weighting of the early marker data. In this talk, we elucidate the nuances between sequential and joint models and illustrate their application using data from oncology clinical trials.
Title of Speaker's Presentation #2 (if this speaker is in more than one proposal)
Panelist (Session Title: Bridging Causal Inference and Pharmacometric Modeling)
Title of Speaker's Summary #2 (if this speaker is in more than one proposal)
Join us for an engaging and insightful discussion on the integration of causal inference methodologies with pharmacometric modeling to enhance drug development and clinical practice. Our esteemed speakers will share their expertise on the primary challenges and innovative solutions for incorporating these methodologies into pharmacometric models, ultimately enhancing their capability to inform critical decisions in drug development and clinical settings. We will delve into the principles of causal mediation analyses, focusing on how to evaluate longitudinal biomarkers effectively. By comparing landmarking strategies with other common approaches, such as joint modeling, we will explore the implications for interpreting surrogate endpoints and gain a deeper understanding of their role in clinical outcomes. The discussion will also critically examine the application of instrumental variable techniques in assessing causality within exposure-response relationships. We will explore how these techniques can provide quantitative measures of the impact of different origins of exposure variability on response. By understanding the underlying model assumptions, we aim to comment on the reliability and applicability of these techniques in pharmacometric modeling. Furthermore, we will address the impact of model misspecification and discuss the utility of non-linear mixed effects modeling in managing unmeasured confounders in longitudinal settings. By employing these models, we can adjust for unmeasured confounders and provide unbiased estimates of hypothetical estimands, thereby enriching our understanding of causal dose-exposure-response relationships. These discussions will be supported whenever possible with concrete application examples. Your active participation and insights are warmly welcomed as we explore these complex and fascinating topics together. This discussion aims to foster a deeper understanding of how causal inference methodologies can be seamlessly integrated into pharmacometric modeling to drive innovation and improve outcomes in drug development and clinical practice.