Assessment of traditional and novel effect measures for time-to-event endpoints: a meta-epidemiological study of...

Time-to-event endpoints are essential for evaluating treatment efficacy in oncology trials. The hazard ratio, although commonly used, captures only the relative effect and may not suffice in diverse clinical contexts. Traditional measures based on incidence rates and restricted mean survival time, along with novel measures based on average hazard (AH), offer both relative (ratio-based) and absolute (difference-based) perspectives. However, these measures have not been systematically evaluated in real-world oncology trials, limiting their practical application.

This meta-epidemiological study analyzed individual patient data reconstructed from Kaplan-Meier curves of 46 randomized controlled oncology trials published in five high-impact journals, involving 35,994 patients and 52 curves. The reconstruction used a validated algorithm to build patient-level data from published curves. Seven effect measures were evaluated: hazard ratio, incidence rate ratio and difference, restricted mean survival time ratio and difference, and AH ratio and difference. Pairwise concordance among these measures was assessed using visualizations, median differences, Spearman rank correlation, and intraclass correlation coefficients.

There was a high agreement in clinical direction and statistical significance among the 7 measures. Among the four ratio-based (relative) measures, hazard ratio, incidence rate ratio, and ratio of AH demonstrated high agreement, with median differences ≤0.004, correlations >0.96, and intraclass correlation coefficients >0.87. The restricted mean survival time ratio showed substantial inconsistency, with other relative measures being approximately 1.25 times higher. The nonproportionality of hazards further increased this ratio to 1.38. Among the three difference-based (absolute) measures, incidence rate difference and AH difference were closely aligned, whereas the restricted mean survival time difference exhibited substantial variability, with median ratios of treatment effect magnitude exceeding 300.

AH-based measures provide promising alternatives to the hazard ratio by incorporating both relative and absolute perspectives. The restricted mean survival time based measures offer clinically relevant insights but exhibited substantial differences in magnitude; therefore, they should be interpreted independently and not directly compared with other measures. Incidence rate-based measures may serve as practical approximations when other metrics are unavailable. Routine reporting of multiple effect measures in oncology trials can enhance clinical interpretation and support more nuanced, evidence-based decision-making.