Gastrointestinal and Variceal Bleeding Under Atezolizumab–Bevacizumab in Hepatocellular Carcinoma: Evidence from Trials to Real-World Practice

Atezolizumab–bevacizumab is established as first-line therapy for unresectable hepatocellular carcinoma (HCC) based on phase III randomized evidence. Although overall safety outcomes were acceptable in the registration trial, the risk of gastrointestinal (GI) and variceal bleeding remains a clinically relevant concern, particularly in patients with cirrhosis and portal hypertension. Differences between trial-based safety estimates and observational data necessitate a focused evaluation of hemorrhagic risk in this setting. Randomized trial data indicate low rates of high-grade bleeding under protocol-driven endoscopic screening and predefined eligibility criteria. In contrast, real-world cohorts report higher incidences of GI and variceal hemorrhage, especially among patients with prior bleeding, untreated or high-risk varices, reduced hepatic reserve, and extensive portal vein tumor thrombosis. Pooled analyses confirm an increased prevalence of bleeding with atezolizumab–bevacizumab compared with non-antiangiogenic systemic therapies, although event rates vary across studies due to differences in patient selection and bleeding definitions. Severe and fatal hemorrhage occurs in a minority of cases and is concentrated in clinically high-risk subgroups. Bleeding during atezolizumab–bevacizumab therapy is influenced by baseline portal hypertension severity, hepatic functional status, and tumor-related vascular involvement. Trial-derived safety data reflect outcomes under controlled conditions and may underestimate risk in broader populations. Structured baseline assessment, endoscopic evaluation, and risk-adapted portal hypertension management are integral to clinical implementation. Prospective studies incorporating standardized hemorrhage definitions and predefined risk stratification frameworks are required to refine patient selection and optimize safety in routine practice.