Primary Submission Category: Propensity Scores
Propensity-score-based predictive probability of success using surrogate endpoints
Authors: Jun Lu, Sanjib Basu,
Presenting Author: Jun Lu*
Over 50% of Phase III clinical trials failed due to a lack of significant efficacy, despite proven efficacy in the completed Phase II. This discordant efficacy can be attributed to different populations and endpoints investigated in two phases. Phase III trial failures have significant consequences for both patients and investigators, making it crucial to try to prevent them from occurring. One way to mitigate the risk of failure is to use the predictive probability of success (PPoS), which is a quantitative tool based on prior knowledge and available evidence. Properly applied, PPoS can identify failing trials early and allow for their termination. However, when calculating PPoS using accumulated data, it is important to consider heterogeneity in populations and endpoints. To address this issue, we propose a method for predicting the success of future Phase III trials based on the results of past trials. Our approach adjusts for heterogeneity in populations using the propensity scores method. PPoS can be calculated using either surrogate or both surrogate and final endpoints. Additionally, in a Bayesian framework, the propensity-score-based informative prior can be used to increase sample size with reduced bias. We have applied our method to the development of a drug for multiple sclerosis.