Quantifying treatment differences in confirmatory trials under non-proportional hazards

Area of Science:

• Clinical Trials
• Biostatistics
• Oncology Research

Background:

• Proportional hazards is a standard assumption in oncology clinical trials.
• Novel therapies like immunotherapy often violate this assumption, complicating interpretation.
• Traditional methods like the log-rank test and hazard ratio face challenges under non-proportional hazards.

Purpose of the Study:

• To compare the performance of log-rank test, weighted log-rank test, and Restricted Mean Survival Time (RMST) test.
• To evaluate treatment effect estimation under various non-proportional hazard scenarios.
• To assess the clinical interpretability of different statistical measures.

Main Methods:

• A simulation study was conducted.
• Compared log-rank test, weighted log-rank test, and RMST test.
• Evaluated performance under different non-proportional hazard patterns.

Main Results:

• Hazard ratios lack straightforward clinical interpretation with non-proportional hazards.
• RMST ratio remains interpretable irrespective of the proportional hazards assumption.
• RMST test demonstrates power performance comparable to the log-rank test.

Conclusions:

• RMST provides a clinically meaningful measure for survival analysis when proportional hazards do not hold.
• RMST is a viable alternative to the log-rank test in modern oncology trials with novel therapies.
• Statistical methods must adapt to evolving treatment landscapes in clinical research.