Biostatistics Seminar Series - Devan V. Mehrotra, PhD

Title: Survival Analysis using a 5-STAR Approach in Randomized Clinical Trials Abstract: The logrank test, or, equivalently, the score test from an unstratified Cox proportional hazards (PH) model, is commonly used for the primary analysis of time-to-event data in randomized clinical trials.  Two challenges emerge when the PH assumption is violated: the estimated hazard ratio (HR) is hard to interpret and the power of the logrank test can be substantially diminished.  Statistical strategies to guard against the potential adverse effects of non-PH include (i) use of a stratified logrank test (stratified Cox PH model) based on pre-specified stratification factor(s), (ii) a versatile use of weighted logrank test(s) and (iii) a between-group comparison of restricted mean survival time (RMST).  We propose a novel approach motivated by the observation that overall non-PH is often due to the overall population being a mixture of underlying risk-based homogeneous subpopulations ("strata"), with PH or approximate PH evident within each risk stratum.  Our approach entails using a pre-specified tree-based algorithm to separate patients into risk-based strata based on conditional associations between baseline covariates and observed survival times.  A treatment effect is estimated within each risk stratum, and the stratum-level effects are combined for overall estimation and inference.  We describe the application of our 5-step stratified testing and amalgamation routine (5-STAR) using three real datasets.  Extensive simulations show that 5-STAR is a promising alternative to the aforementioned methods for tackling non-PH in terms of bias, power and interpretability. Devan V. Mehrotra* and Rachel Marceau West