Abstract Body

Under- or over-estimation of the hypothesized failure rates in the definition of non-inferiority bounds for a hazard ratio (HR) estimand can significantly impact on the probability of a trial demonstrating non-inferiority for a hazard ratio and complicate the interpretation of the study findings. The restricted mean survival time (RMST) measure have not been used as primary measure of efficacy in HIV/AIDS clinical trials and may offer a powerful alternative to the hazard ratio. We compared analysis based on the difference in RMST (Δ-RMST) measure with 2 treatment-effect measures in a recent HIV equivalence trial, and investigated the performance and characteristics of Δ-RMST-based analysis.

Primary and secondary virologic failure (VF) outcome measures from ACTG A5257 were reanalyzed using hazard ratio (HR) and Δ-RMST estimands and compared the to the original study results based on risk difference estimated by Kaplan-Meier (RDKM). A5257 equivalence bounds were transformed for each estimand assuming exponential VF distributions and A5257 design characteristics. The performance and operating characteristics of Δ-RMST-based analysis in the setting of non-proportional hazards ratio were investigated in a simulation study.

Table summarizes results of the analyses in the ACTG 5257 study and alternative analyses. Analyses based on Δ-RMST globally led to similar conclusions as the published finding based on RDKM. In contrast, analyses based on HR provided some discordant equivalence conclusions compared both with the initial analyses based on RDKM and the Δ-RMST despite that appeared driven by very low failure rates in one group. Results of our simulation study indicated that the violation of the proportional hazards assumption may negatively an impact the probability of declaring equivalence for a Δ-RMST based analysis.

The RMST based analysis could be a promising alternative measure of efficacy in HIV/AIDS clinical trials although further discussion to define non-inferiority bounds is needed.