Chair: Miaomiao Ge, PhD (Boehringer Ingelheim Pharmaceuticals, Inc.)
Co-Chair: Jingjing Schneider, PhD (BeiGene)
Abstract: In an era where data-driven decisions are paramount, the application of Bayesian Statistics has emerged as a powerful tool. This session will explore innovative Bayesian statistical approaches that dynamically leverage prior knowledge and observed data to make informed decisions during drug development. Specifically, we will delve into the application of Bayesian designs, including Bayesian Dynamic Borrowing, in HIV drug development, an innovative Bayesian approach leading to a dynamic ‘Go’ probability, and the Bayesian joint modeling techniques addressing complex challenges in longitudinal data analysis and survival modeling.
Speaker: Chengxue Zhong, PhD (Boehringer Ingelheim Pharmaceuticals, Inc.)
Title: Dynamic Probability of ‘Go’ in Early Oncology
Abstract: In early oncology, expansion cohorts are typically non-randomized, and data is available with minimal delay. This raises the question about the projected probability of reaching the quantitative ‘Go’ / PoCP criterion. Effective use of individuals at-risk and a Bayesian time-to-event model can improve the quantification of the probability of ‘go’ and can be used to initiate frontloading of activities. Therefore, a Bayesian multi-state model was developed, it allows imputing future trial data based on the posterior-predictive distribution, and the generated data can be used to evaluate quantitative ‘go’ decision criteria to determine whether the sampled data would lead to a ‘go’ decision. The average sampled ‘go’ decision rate over repetitive data sets is the MCMC approximation of the probability of ‘go’. This approach can be taken at any interim time point from the posterior predictive distribution conditioning on any observed data, which leads to a dynamic probability of ‘go’. We will illustrate statistical details and use an oncology trial as a retrospective example to demonstrate the performance of the proposed model.
Speaker: Gaby Ghita, PhD (GSK)
Title: Innovation Bayesian Design/Implementation in HIV Clinical Trials
Abstract: The pharmaceutical industry is increasingly adopting Bayesian methods in clinical trials to leverage existing data for improved decision-making, accelerated development timelines, and cost reduction. A significant factor driving this shift is the ability of Bayesian methods to effectively integrate relevant external information, whether from historical data or expert opinion. Bayesian Dynamic Borrowing (BDB) provides a flexible approach by adjusting the extent of information borrowed based on the alignment between the source data and the target study population, which enhances the probability of success of drug development and facilitates informed regulatory decision-making. This presentation will discuss the implementation of Bayesian designs, including BDB, and their associated go/no-go criteria across various stages of HIV drug development, with a focus on applications in Early, Late and pediatric development.
Speaker: Yan Gao, PhD (Medical College of Wisconsin)
Title: Accelerated Fitting of Joint Models of Survival and Longitudinal Outcomes While Accounting for Cumulative Variations of CD4 Biomarker in an HIV/AIDS Clinical Trial
Abstract: Stochastic volatility often implies increasing or hidden risks that are difficult to capture in dynamic real-world applications. We propose a novel statistical measurement aligned with arc length, a mathematical concept, to quantify cumulative variations of longitudinal biomarkers in survival analysis. We define cumulative variation as the total variability over time to characterize stochastic volatility fully. The hazard rate defined by the Cox proportional hazards model is assumed to be impacted by the instantaneous value of a longitudinal variable. However, when cumulative variations significantly impact the hazard, this assumption is questionable. Our proposed Bayesian arc model infuses arc length into a united statistical framework by synthesizing three parallel components: Bayesian joint models, distributed lag models, and arc length. We illustrate the usage of the proposed model in simulation studies to assess its performance. Furthermore, we apply it to an HIV/AIDS clinical trial to assess the impact of cumulative variations of CD4 count, a critical longitudinal biomarker of immune function in HIV patients, on mortality while accounting for measurement errors and relevant key variables, including treatment.
