All Times EDT
Keywords: Probability of Study Success, Bayesian Network Meta Analysis, Selection Bias
Drug development is a complex and costly process, and one of the most critical decision is if a compound should move to late-stage clinical trial for registration purpose once the proof of concept study result is available. One big challenge is the effectiveness of the drug is unknown and the treatment effect is not easy to quantify. Bayesian Network Meta-Analysis (BNMA) is commonly used to synthesize relevant data and derive the distribution of multiple treatment effects, which is essential for Probability of Study Success (PrSS) calculation. Then PrSS could be used to facilitate the investment decision. However, selection bias exists since only compounds with promising results from the small study (Ph2) would move to the next development stage (Ph3).
The BNMA with selection-bias adjustment has been used to address this issue. A hierarchical model for distribution of the efficacy for a portfolio of compounds was considered, which can serve as the prior distribution. We further extended the original BNMA to bias-adjusted BNMA (BaBNMA) for composite endpoints. Theoretical and numerical results will be discussed. The BaBNMA approach with composite endpoints was illustrated via a hypothetical late stage compound where the triggered Go or NO-GO decision making need to be made.