Chairs:
Kentaro Takeda, PhD (Astellas Pharma Inc.)
Yusuke Yamaguchi, PhD (Astellas Pharma Inc.)
Abstract: Making quantitative Go/No-Go decisions for advancing investigational therapies to Phase 3 is a central challenge in clinical development, where efficiency, uncertainty, and strategic risk must be balanced under increasing scientific and operational constraints. This invited session brings together recent advances in Bayesian quantitative decision-making and evidence integration to highlight how statistical innovation can directly support program-level strategy. The three presentations showcase complementary perspectives across development scenarios, including accelerated programs, rare diseases, and evidence synthesis using only aggregated data. The first talk discusses program-level risk quantification in accelerated development, where sparse information and compressed timelines require transparent modeling of uncertainty and clear communication of risk to multidisciplinary decision makers. The second presentation extends Bayesian quantitative decision-making to settings with co-primary or dual endpoints, demonstrating how modeling correlation between endpoints can meaningfully reduce inconclusive outcomes in small-sample rare disease trials. The third talk introduces a Bayesian hierarchical framework for extracting insights about disease trajectories and mechanisms of action using only published aggregated outcomes, enabling cross-study integration when subject-level data are unavailable. Together, these presentations illustrate how Bayesian modeling, principled integration of external information, and transparent communication of uncertainty can strengthen Go/No-Go decision-making and improve the rigor and efficiency of clinical development. A discussant will synthesize themes across the talks and reflect on the evolving role of quantitative methods in Phase 3 transition decisions.
Speaker: Andrew Bean, PhD (Novartis)
Title: Quantifying Program-Level Risk for Accelerated Drug Development
Abstract: Modern pharmaceutical drug development increasingly demands efficiency and speed. In the face of rising costs and competition, acceleration is demanded in not only study execution, but also in strategic clinical development planning and decision making. However, acceleration of clinical development (for example, skipping traditional steps of early development in order to quickly accelerate to larger confirmatory trials) entails risk. Statistical perspective is crucial to transparently quantify these risks and support decision making. Bayesian approaches are well suited for such situations, quantifying uncertainty in key quantities like the treatment effect on pivotal endpoints, and predicting trial outcomes in light of this uncertainty. The challenges in bringing Bayesian thinking to this problem are twofold. First, the development of a suitable prior distribution for evaluation brings unique challenges in accelerated plans where relevant data may be sparse or even absent. Second, clear and transparent communication is indispensable, to ensure the risks involved in clinical development plan options are apparent to diverse stakeholders and decision makers, from biostatisticians to decision boards. We experiences and some lessons learned in addressing these challenges.
Speaker: Yusuke Yamaguchi, PhD (Astellas Pharma Inc.)
Title: Bivariate Bayesian Quantitative Decision Making in Rare Disease Drug Development
Abstract: Quantitative decision-making (QDM) frameworks are essential for efficient clinical development in rare diseases, where limited patient populations necessitate making definitive Go/No-Go decisions based on small proof-of-concept studies. While established Bayesian approaches exist for single-endpoint evaluation, many clinical development programs require the simultaneous assessment of multiple endpoints, such as co-primary endpoints and a combination of surrogate and clinical endpoints. We extend the Bayesian QDM framework to accommodate the joint evaluation of two endpoints, explicitly modeling their correlation structure while maintaining the capability to integrate external data through power priors. The present talk provides a general framework of the two-endpoint Bayesian QDM and proceeds to an application to specific settings. Through a case study, we demonstrate that properly incorporating correlation substantially reduces inconclusive decisions. This framework enhances rare disease drug development by providing principled bivariate endpoint evaluation that maximizes information extraction from resource-constrained settings, facilitating more definitive and efficient Go/No-Go decision-making.
Speaker: Yan Chu, PhD (Amgen)
Title: Generating Insights into Disease Trajectories and Treatment Mechanisms of Action (MOAs) from Aggregated Outcomes Published Across Multiple Sources Using a Bayesian Hierarchical Model
Abstract: Clinical trial publications in Oncology and Hematology routinely report short-term endpoints such as overall response rate and duration of response, along with long-term outcomes including progression-free survival (PFS) and overall survival (OS). These aggregated data contain important information on disease behavior and treatment MOAs. Without subject-level data, establishing connections between short-term and long-term endpoints across studies is difficult, and evidence integration is hindered by incomplete reporting, heterogeneous eligibility criteria, variable sample sizes, distinct therapeutic mechanisms, and differences in study duration. To address these limitations, we propose a Bayesian Hierarchical Hypoexponential Model that decomposes published survival summary statistics into a parameterized continuous-time Markov chain (CMC). Building on the previously developed PubPredict framework, disease progression is represented through predefined states, with progressive disease (PD) and/or death, which are digitized from PFS/OS curves, serving as absorbing states. Transition intensity parameters define pathways beginning with stable disease (SD), followed by response and PD, and capture the relationship between short- and long-term outcomes. A Gamma prior structure is used to borrow strength across publications with varying levels of uncertainty, incorporating both sample size and prior knowledge. A weighted resampling approach further incorporates clinical plausibility and similarity between selected publications and the target study. Model outputs include posterior parameter estimates with credible intervals and visual representations, such as Sankey diagrams, to illustrate transition pathways and inferred MOAs. Case studies in non-small cell lung cancer and multiple myeloma demonstrate the applicability of simplified and full model variants. This framework provides a robust approach for leveraging heterogeneous published evidence to characterize disease trajectories and treatment MOAs in the absence of subject-level data and enables comparative assessment of investigational and competitor therapies using aggregated outcomes.
Discussant: Cong Han, PhD (Astellas Pharma Inc. )
