A comparative study to alternatives to the log-rank test

Area of Science:

• Biostatistics
• Medical Research

Background:

• Time-to-event data analysis is crucial in medical research for comparing survival across groups.
• The log-rank test is standard but assumes proportional hazards, which is often not met.
• Recent developments include omnibus tests and restricted mean survival time methods, showing promise in biometrics.

Purpose of the Study:

• To conduct a comprehensive simulation study comparing established statistical tests with newer omnibus and restricted mean survival time methods.
• To evaluate test performance under various conditions, including non-proportional and crossing hazards, unequal censoring, and small or unbalanced sample sizes.
• To provide updated recommendations for survival data analysis in medical research.

Main Methods:

• A large-scale simulation study was designed to assess statistical test performance.
• Various scenarios were simulated, manipulating survival distributions, censoring patterns, and group sizes.
• The power of different tests, including traditional and recent methods, was compared across these simulated settings.

Main Results:

• Omnibus tests demonstrated superior robustness and power when the proportional hazards assumption was violated.
• The performance of various tests varied significantly depending on the specific simulation setting.
• Newer methods, particularly omnibus tests, showed advantages in complex survival scenarios.

Conclusions:

• Omnibus tests are recommended for their robustness against deviations from proportional hazards assumptions.
• When uncertainty exists regarding survival time distributions, omnibus approaches provide more reliable group comparisons.
• These findings support the adoption of more flexible statistical methods in survival analysis.