Diverging conclusions from risk difference and difference in restricted mean survival time in quantifying absolute...

Area of Science:

• Clinical Trials
• Biostatistics
• Survival Analysis

Background:

• Risk-based analyses are increasingly used for heterogeneous treatment effects (HTEs) in clinical trials.
• Traditional assumptions for time-to-event analyses may not hold for absolute treatment effects across risk strata.
• Absolute treatment effects can be measured by risk difference (RD) or difference in restricted mean survival time (ΔRMST).

Purpose of the Study:

• To examine risk-based HTE analyses in time-to-event data.
• To identify patterns of absolute HTE across risk strata.
• To compare the utility of ΔRMST versus RD in guiding treatment decisions.

Main Methods:

• Utilized artificial and empirical time-to-event data.
• Compared RD (difference in Kaplan-Meier estimates at a time point) and ΔRMST (area between Kaplan-Meier curves) across risk strata.
• Explored scenarios with constant hazard ratios (HRs), varying event rates, and risk model discrimination.

Main Results:

• When event rates and discrimination were low, both RD and ΔRMST showed monotonic increases, favoring high-risk patients.
• With increased event rates and/or discrimination, a 'sweet spot' pattern emerged, benefiting intermediate-risk patients.
• RD identified the 'sweet spot' pattern even when ΔRMST favored higher-risk patients, potentially understating benefits and leading to mistargeting.

Conclusions:

• The pattern of HTE assessed by RD can differ significantly from ΔRMST.
• Using RD may lead to suboptimal treatment decisions and mistargeting of therapies.
• ΔRMST is recommended for evaluating absolute HTE in time-to-event data.