82 – Contributed Poster Presentations: Government Statistics Section
Predicting Future OS Behavior Using Bayesian Posterior Predictive Distributions
Pourab Roy
US Food and Drug Administration
Time-to-event endpoints such as overall survival (OS) and progression-free survival (PFS) are commonly used to determine the efficacy of drugs in oncology. Usually efficacy is determined by performing hypotheses tests at pre-specified interim and final analyses. The results based on interim analyses often exhibit large variability and uncertainty due to immature data. A common clinical question of relevance is to forecast how these results will look with continued follow-up. We adopted the use of Bayesian predictive probability to implement a prediction model, assuming an underlying piece-wise exponential distribution of the survival times, to simulate future behavior based on the current data. The goal of the approach is to check whether this method can predict the robustness of the interim OS analysis results from different approved clinical trials and determine whether they are representative of the final results. To evaluate the approach, we present a comparison of the predicted and actual performance of six different clinical trial datasets. We have also developed a Rshiny app based on the methodology.