Strategies to address non-proportional hazards between survival curves - Lessons from phase III trials in hepatocellular carcinoma
Ezequiel Mauro, Tiago de Castro, Marcus Zeitlhoefler, Allan Hackshaw, MinJae Lee, Tim Meyer, Amit G. Singal, Josep M. Llovet
Abstract
BACKGROUND & AIMS: Non-proportional hazards (NPH) can lead to discrepancies between interim (IA) and final analyses (FA) in randomized controlled trials (RCTs) of hepatocellular carcinoma (HCC). We assessed the impact of NPH in pivotal HCC trials and proposed strategies for more robust analyses. METHODS: Phase III pivotal HCC RCTs (2008-2024) were reviewed. Proportional hazards were tested using the Grambsch-Therneau method. For trials with NPH, we proposed an optimal IA timing (twice the estimated median of the primary endpoint in the control arm) or a minimum event threshold (≥60%). In NPH scenarios, the MaxCombo test, restricted mean survival time (rRMST), and piecewise hazard ratios were applied. RESULTS: NPH was present in 4/20 (20%) phase III trials in HCC, all involving immunotherapies, and displayed three patterns: 1) diminishing effects, 2) delayed effects, and 3) crossing hazards. Two RCTs (IMbrave050, LEAP-012) reported positive IA results with diminishing effects. In IMbrave050, discrepancies were observed when comparing IA and FA using MaxCombo analysis (p = 0.02 and p = 0.33, respectively), rRMST at 12 and 36 months (1.11 [p <0.001] and 1.08 [p = 0.08], respectively) and piecewise hazard ratios before vs. after 12 months (0.59 [95% CI 0.43-0.73] vs. 1.12 [95% CI 0.88-1.37]). In LEAP-012, results were consistent across 12 and 24 months (MaxCombo test p <0.001) and rRMST (1.20 [p <0.001) and 1.27 [p <0.001]). HIMALAYA and Checkmate 9DW reported positive results, which were confirmed by MaxCombo test. HIMALAYA showed delayed effects (rRMST at 12 and 36 months: 1.04 [p = 0.13] and 1.15 [p = 0.004]), while CheckMate 9DW displayed crossing hazards (rRMST at 12 and 36 months: 0.95 [p = 0.07] and 1.12 [p = 0.03]). CONCLUSION: NPH contributed to discrepancies between IA and FA in IMbrave050. Robust IA require a minimum follow-up duration or event count before prematurely stopping RCTs in the presence of NPH. IMPACT AND IMPLICATIONS: Non-proportional hazards (NPH) impact phase III randomized controlled trials in hepatocellular carcinoma, particularly in immunotherapy trials, potentially causing discrepancies between the interim and final analyses. In fact, halting trials at the interim analysis can be premature when NPH is present. Thus, we propose a framework to ensure study maturity based on follow-up duration and event accruals to optimize interim analyses in the presence of NPH. Whenever NPH is identified, distinct statistical tools should be used to assess reliable differences between arms (MaxCombo) and to assess the effect size (restricted mean survival time and piecewise hazard ratios) for regulatory decisions and clinical guidance. Implementing these strategies can improve trial design, and better support clinical decision-making.