Objectives/Goals: Misinterpreting hazard ratio (HR) as absolute risk reduction is common in randomized trials (RCTs) with time-to-event endpoints, inflating benefit. We quantify this by describing risk ratio (RR) estimators for survival data and compare HRs with milestone and time-averaged RRs across a large collection of cancer RCTs. Methods/Study Population: We conducted a meta-epidemiological study in a convenience sample of more than 650 cancer RCTs (2002–2024), excluding trials without reported benefit, those stopped early, or non-inferiority designs. Pseudo-individual patient data were reconstructed from survival curves for overall survival and other event-free endpoints (such as progression-free survival). From these, we calculated Cox HRs and milestone-based RRs on common time supports. We also calculated time-constant RRs and weighted-averages of time-varying RRs. Primary contrasts were inverse-variance pooled HR/RR ratios for each RR definition. HR/RR < 1 indicates larger effects with HR as compared to RR. Results/Anticipated Results: In a preliminary subset (n=120 trials), the pooled HR/RR ratio was 0.80 (95% confidence interval, 0.76 to 0.82): interpreting the HR as a RR would overstate benefit by 20%. Although all included trials reported statistically significant HR < 1, only 73% had RR < 1 and statistically significant: over one quarter failed to show benefit on the absolute risk scale. Results were consistent when using time-constant and weighted-average RRs. We anticipate comparable findings in the full sample. Discussion/Significance of Impact: This study quantifies the inflation that arises when HRs are read as RRs. Reporting RR estimators for time-to-endpoints alongside HRs would improve communication of trial effects. Our findings also have design implications: to detect absolute risk reductions, RCTs would often need larger N or longer follow-up.
Nolan et al. (Wed,) studied this question.