Performance of Restricted Mean Survival Time Based Methods and Traditional Survival Methods: An Application in an...

Area of Science:

• Biostatistics
• Survival Analysis
• Epidemiology

Background:

• Traditional survival analysis methods, such as the log-rank test and Cox regression, rely on assumptions like proportional hazards (PH).
• Violations of the PH assumption can complicate the interpretation of results, particularly hazard ratios (HRs), making them difficult to report as single values.
• Restricted mean survival time (RMST)-based methods offer an alternative approach to survival analysis that is less sensitive to the PH assumption.

Purpose of the Study:

• To compare the performance of RMST-based methods against traditional survival analysis techniques.
• To evaluate RMST-based methods in the context of multiple covariates.
• To assess the interpretability and robustness of RMST-based measures compared to traditional effect measures like HR.

Main Methods:

• Utilized data from 4405 osteosarcomas from the Surveillance, Epidemiology, and End Results (SEER) Program Database.
• RMST-based methods included Kaplan-Meier (KM) group comparison, pseudovalue (PV) regression, and inverse probability of censoring probability (IPCW) regressions.
• Traditional methods included log-rank test, Wilcoxon test, and Cox regression (including time-dependent variables), with assessment of PH and censoring assumptions.

Main Results:

• RMST-based methods, particularly PV regression and IPCW with group-specific weights, provided consistent estimations, outperforming IPCW with individual weights.
• PV regression demonstrated more robust statistical power compared to IPCW regressions with group-specific weights.
• When the PH assumption was violated, RMST difference offered a single, interpretable value, unlike time-varying HRs; logarithmic relationships were observed between HR and RMST difference.

Conclusions:

• The difference in RMST is more interpretable than time-varying hazard ratios, especially when proportional hazards assumptions are not met.
• Kaplan-Meier and pseudovalue regression are recommended as preferred RMST-based methods, with IPCW regression suitable for sensitivity analysis.
• Comprehensive covariate effect assessment is encouraged by adopting both traditional and RMST-based survival analysis methods.