Early and late survival dynamics of immunotherapy combinations in advanced HCC: Anchored indirect comparison of atezolizumab plus bevacizumab versus durvalumab + tremelimumab.

BACKGROUND AND AIMS: Atezolizumab plus bevacizumab (A+B) and STRIDE (tremelimumab plus durvalumab) represent two of the approved first-line immunotherapy strategies for unresectable hepatocellular carcinoma (HCC). As no head-to-head trial exists, we assessed temporal differences in survival benefit using complementary analytic frameworks.

METHODS

Three independent analyses were conducted: (1) anchored indirect comparison using reconstructed survival from IMbrave150 and HIMALAYA; (2) a real-world A+B cohort (n = 883) weighted via MAIC to match STRIDE baseline characteristics; and (3) a propensity-matched population-level dataset from TriNetX (n = 309 per arm). Time-interval hazard ratios (HRs), RMST, and conditional landmark survival were calculated. Proportional hazards were assessed by Schoenfeld testing; pooled interval HRs were synthesized through meta-analysis.

RESULTS

Across all models, A+B showed a consistent early advantage. In the first 6 months, the pooled HR favored A+B (0.75; 95% CI 0.64-0.88), with absolute survival differences of + 5-7%. From 6-12 months, HRs remained numerically favorable but non-significant. Between 12 and 24 months, pooled HRs approached neutrality (0.93-1.06), with confidence intervals crossing unity and no meaningful absolute survival differences. Beyond 24-36 months, the effect reversed in favor of STRIDE (anchored HR 1.41; RWD HR 2.52; TriNetX HR 1.47). Meta-analysis confirmed a progressive late benefit for STRIDE from 36 to 60 months (pooled HR 1.54-1.75). Schoenfeld tests indicated time-dependent effects in all frameworks (p < 0.05), and ΔRMST trajectories were highly concordant (Pearson r > 0.97).

CONCLUSIONS

A+B is associated with stronger early tumor control, while STRIDE provides more durable long-term survival beyond 2-3 years. These findings support selecting therapy based on temporal treatment objectives and advocate for biomarkers predicting early versus sustained treatment benefit.