Bayesian Multistate Transition Modeling of Response, Progression, and Death to Elucidate Mechanisms of Survival Benefit in Oncology Trials

PURPOSE Innovate a novel Bayesian nonparametric approach for clinical trial analysis devised to infer intrinsic survival benefit mechanisms of action (SBMoA) in confirmatory randomized controlled trials (RCTs) with survival end point by comparing 12 intrinsic state transitions between stable disease, early response, progression/recurrence, and death. METHODS The method characterizes patient trajectories through states of stable disease, response, progression, and death using multiplicative transition intensities. Bayesian inference compares observed transition rate differences with benchmark distributions for five SBMoAs (Null, Survival Null, Mediation, Direct, Mixed), yielding posterior predictive probabilities. Performance was evaluated via simulations of phase III trials implemented to compare the accuracy of 12-month interim analyses with final survival analysis results, as well as to compare these results with those obtained using Bayesian mediation modeling, 2-in-1 seamless designs, and group sequential designs for survival analysis. RESULTS The accuracy of interim decision making in forecasting a trial's final outcome was highest for the SBMoA method (0.82), which outperformed conditional power (0.65), Pocock (0.60), Chen 2-in-1 (0.55), and O'Brien-Fleming (0.53). The method achieved positive predictive value (PPV) of 0.80 and negative predictive value (NPV) of 0.82, nearest to the ideal truth oracle (PPV 0.96, NPV 0.84) that assumes perfect knowledge of the true scenario. Comparators performed worse: conditional power (PPV 0.63, NPV 0.68), Pocock (PPV 0.72, NPV 0.58), Chen 2-in-1 (PPV 0.52, NPV 0.60), and O'Brien-Fleming (PPV 0.81, NPV 0.53). CONCLUSION Inferring SBMoA using the Bayesian approach provides reliable interim forecasts of survival benefit in the absence of mature survival data. The innovation provides avenues for accelerating interim decision making with enhanced accuracy in confirmatory trials.