Competing risks of treatment discontinuation of atezolizumab plus bevacizumab in unresectable hepatocellular carcinoma.

BACKGROUND & AIMS: Atezolizumab plus bevacizumab (Atezo/Bev) is considered standard first-line therapy for unresectable hepatocellular carcinoma (HCC), yet reasons for treatment discontinuation vary, and conventional survival endpoints do not capture the time-varying probability of each cause. This multicentre retrospective study aimed to quantify the cause-specific discontinuation and its predictors. METHODS: We analysed 685 consecutive patients treated with Atezo/Bev (10 excluded). Time to discontinuation was defined from treatment initiation to cessation and right-censored at last follow-up if ongoing. Discontinuation reasons were categorized as progressive disease (PD), adverse event (AE)-driven discontinuation (treatment-related AEs without hepatic decompensation), hepatic dysfunction-driven discontinuation (decompensation irrespective of mechanism), response/strategy change, and other causes. Cumulative incidence functions (CIFs) were estimated, and cause-specific Cox models were fitted with multiple imputation. RESULTS: Discontinuation causes included PD (n=334), AE (n=125), hepatic dysfunction (n=48), response/strategy change (n=44), and other causes (n=35); 99 were censored. One-year CIFs were 0.394, 0.153, 0.056, 0.042, and 0.041, respectively. Baseline AFP predicted PD-driven discontinuation (HR 1.18 per log10[AFP+1]; p=3.95×10-5). AE-driven discontinuation was independently associated with age (HR 1.03/year), ECOG performance status ≥1 (HR 1.82), and ALBI (HR 1.07 per 0.1 increase; p=7.43×10-4). Hepatic dysfunction-driven discontinuation was strongly associated with age (HR 1.07/year; p=4.28×10-4) and ALBI (HR 1.29 per 0.1 increase; p=8.65×10-13). AFP (<400 vs ≥400 ng/mL) × mALBI (1/2a vs 2b/3) defined four strata. Low AFP/poor mALBI showed the highest 1-year toxicity-related discontinuation (CIF 0.339), whereas high AFP/good mALBI showed predominantly PD-driven discontinuation (CIF 0.520). CONCLUSIONS: Four clinically defined risk strata based on AFP and mALBI differentiated distinct patterns of treatment discontinuation, capturing both PD and toxicity-pathways. This framework may support risk-adapted monitoring and early treatment optimisation, but external validation is warranted. IMPACT AND IMPLICATIONS: Atezolizumab plus bevacizumab is a standard first-line regimen for unresectable HCC, yet discontinuation occurs for multiple reasons that are not captured by conventional survival summaries. Analysis of a large multicentre cohort revealed distinct patterns of treatment discontinuation along with clinical predictors. Baseline AFP primarily forecasted PD-driven discontinuation, whereas baseline liver function (ALBI) and older age dominated the risk of AE and hepatic decompensation. An AFP×mALBI matrix separated patients into distinct discontinuation pathways, identifying a low-AFP/poor-mALBI subgroup with high early toxicity-related discontinuation and a high-AFP/preserved-mALBI subgroup with predominantly PD-driven discontinuation. Since this framework has not yet been externally validated, it should be regarded as hypothesis-generating. Nevertheless, it may help clinicians consider risk-adapted monitoring, toxicity preparedness, and supportive care according to the dominant discontinuation risk.