A Modified Power Prior for Time-to-Event Analysis with Piecewise Constant Hazards and Outcome Similarity Driven Bounded Weighting

Music City Center 

The power prior is a widely used Bayesian approach for incorporating historical data in clinical trials, but its application to time-to-event data presents challenges, particularly in defining prior-data compatibility within the contexts of interpolation and extrapolation and ensuring appropriate borrowing. We propose a variation of the power prior for time-to-event analysis under a piecewise constant hazard model, using outcome similarity-driven bounded weighting approach to dynamically adjust historical data contribution. We evaluate its performance through simulations assessing type I error, power, bias, and effective sample size (ESS) and compare it with Normalized Power Prior (NPP) and Commensurate Power Prior (CPP) methods. A case study further illustrates its practical implementation.