Machine Learning-Enabled Real-World Evidence in Synthetic Arms Trials

• Activity Number: 332 - Synthetic Clinical Trials Design to Accelerate FDA Approvals
• Type: Invited
• Date/Time: Thursday, August 12, 2021 : 10:00 AM to 11:50 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Prater Edmund and Kay-Yut Chen and Sridhar Nerur*
• Companies: The University of Texas at Arlington and The University of Texas at Arlington and The University of Texas at Arlington
• Keywords: Real World Data; Clinical Trials; Machine Learning; Randomization; Real World Evidence; Sparsity

Abstract

In this talk, we will discuss strategies for clinical trial designs which use observational data. which combine cohorts who volunteered for an experimental therapy (treatment group) with a synthetically generated control group of patients found in real world data (RWD). Random assignment of patients to treatment and control groups yield comparable results. In contrast, a non-random selection process results in misleading comparisons when patients exposed to the treatment group differ systematically from observed data. So, adjustment for lack of randomization is paramount. Under lack of randomization, we show equivalence between cohorts. Due to sparsity of data (very few volunteers in the treatment group), and very few matching patients in the control group (observed data), establishing equivalence between the two groups is a monumental challenge. We will address these two challenges in the context of a clinical trial conducted by Gilead Sciences for a cancer drug.