PhD student: Valentin Vinnat
Title: Enrichment Bayesian designs for multiple classifier biomarkers
Supervisor: Sylvie Chevret
Doctoral school:ED 393 Epidemiology and Biomedical Information Sciences, Université Paris Cité
Promotion: 2020
Abstact: In the dynamic and complex setting of personalized medicine, clinical trials face substantial challenges, especially concerning statistical and ethical dimensions when incorporating biomarkers into research protocols. Accurately identifying the subpopulations that may benefit most from the considered treatment remains a central issue. To address these challenges, this thesis introduces methodological contributions that aimed at enhancing the precision and efficiency of such clinical trials.
The first section of this work delineates a Bayesian adaptive scheme for population enrichment, building upon a range of biomarkers and incorporating interaction measures into decision rules. This approach aims to progressively select a subpopulation of patients with a high probability of positively responding to the treatment of interest, focusing on the statistical interaction between the treatment effect and the biomarker.
The second part develops a hybrid Bayesian adaptive scheme, leveraging previously established decision rules and introducing adaptive randomization, thus providing additional optimization in the assignment of the treatment to patient subgroups.
The final component of this work introduces a sequential Bayesian scheme focusing on the identification and classification of biomarker-based subgroups. This strategy employs a truncated Poisson model with zero inflation and uses the SUCRA method to rank the subgroups most receptive to the experimental treatment based on their predictive effect, thereby offering a precise perspective on treatment efficacy based on a continuous judgment criterion.
Otherwise, each contribution of this thesis was based on clinical trials conducted in intensive care unit, adopting a Bayesian inference approach to develop and validate the proposed methods. The performance of these methods was confirmed through rigorous simulations, showcasing notable advantages over other existing techniques. This work not only offers advancements in precisely determining optimal subpopulations for therapeutic interventions but also paves the way for future research aiming at further refining these approaches, marking a crucial step towards more patient-centered clinical trials and effectively integrating the complex dynamics of biomarker-treatment interactions.
Keywords: Personalized medicine; Adaptive clinical trials; Biomarkers; Bayesian inference; Treatment-biomarker interaction
