Evaluating Statistical Methods for Sequential Parallel Comparison Designs Under Non-Proportional Hazards

Under proportional hazards, commonly used statistical test statistics maintain appropriate type I error control in sequential parallel comparison designs with time-to-event endpoints, but their performance diverges substantially when the assumption of proportional hazards is violated. In a systematic evaluation examining trial designs with time-to-event outcomes, weighted log-rank and Cox Wald test statistics achieved the highest statistical power when proportional hazards held true.

The sequential parallel comparison design was developed to address high placebo response rates by enriching the study population with placebo nonresponders. Although earlier work established statistical inference for time-to-event endpoints under the assumption of proportional hazards, treatment effects in clinical investigations frequently vary over time. Consequently, it remained unclear whether standard time-to-event methods retain their statistical properties under the two-stage structure of a sequential parallel comparison design when proportional hazards do not hold.

To resolve this question, researchers Xinlin Lu and Guogen Shan conducted a systematic evaluation of test statistic robustness under non-proportional hazards in sequential parallel comparison designs featuring a time-to-event endpoint. Their analysis evaluated commonly used test statistics derived from the Cox proportional hazards model, alongside nonparametric rank-based tests and the restricted mean survival time approach.

When proportional hazards were violated, the performance of the evaluated methods diverged according to the specific pattern of the treatment effect. The nonparametric Gehan-Wilcoxon statistic demonstrated greater sensitivity when treatment effects were early and diminishing, whereas the restricted mean survival time approach performed more reliably under crossing hazards accompanied by early benefit. These findings demonstrate that selecting appropriate test statistics must align directly with anticipated treatment dynamics in time-to-event trials that use the sequential parallel comparison design.

Subject of Research: Statistical methods for sequential parallel comparison designs with time-to-event endpoints

Article Title: Comparison of statistical methods for sequential parallel comparison designs with time-to-event endpoints under non-proportional hazards

Article References: Lu, X., & Shan, G. (2026). Comparison of statistical methods for sequential parallel comparison designs with time-to-event endpoints under non-proportional hazards. BMC Medical Research Methodology. https://doi.org/10.1186/s12874-026-03053-2

Image Credits: AI Generated

DOI: 10.1186/s12874-026-03053-2

Keywords: sequential parallel comparison design, time-to-event endpoints, non-proportional hazards, Cox proportional hazards model, restricted mean survival time, Gehan-Wilcoxon statistic, clinical trial methodology, type I error, Comparison, statistical, methods, sequential