The combination of atezolizumab and bevacizumab is an established first-line therapy for unresectable hepatocellular carcinoma (HCC). However bevacizumab is associated with an increased risk of gastrointestinal bleeding; in IMbrave150, upper GI bleeding occurred in ~7% of patients, leading to guideline recommendations for pre-treatment endoscopy and primary prophylaxis of oesophageal varices. Many centres therefore defer bevacizumab until endoscopic variceal eradication, delaying full systemic therapy. Two tertiary centres instead adopted an upfront bevacizumab strategy with endoscopic variceal ligation (EVL) in parallel. We aimed to assess the safety and feasibility of this approach in a real-world setting. 

This retrospective two-centre cohort included 354 consecutive Patients (290 males, 81,9%; 63 females, 17,8%, mean age 65.7 years) with HCC treated with atezolizumab-bevacizumab between 2020 and 2024. The primary endpoint was any variceal bleeding with EVL before bevacizumab initiation compared with a non-upfront ligation group. Cox regression was used to assess the association between EVL and variceal bleeding, adjusting for variceal grade, Child–Pugh score and macrovascular invasion.

Overall, 15 variceal bleeding events occurred after bevacizumab initiation (4.2%), of which 10 occurred during therapy (2.8 %) and 5 later than 6 months after treatment start (1.4%). In the varices subcohort (n = 145 with documented baseline varices), 57/145 (39.3%) patients underwent endoscopic variceal upfront ligation before therapy inititiation, whereas 88/145 (60.7%) did not. Patients who underwent upfront ligation therapy were substantially more likely to have higher-grade varices (grade II/III), whereas the no-ligation group predominantly had grade I varices (p < 0.001). Despite this imbalance toward more advanced varices in the ligation group, the overall proportion of variceal bleeding events did not differ significantly between groups (8.9% vs 6.8%; p = 0.751). In grade-stratified analyses, no significant differences in variceal bleeding were observed within individual variceal grades, consistent with very low event counts per subgroup. 

In time to event analyses among patients with complete follow up information for the Cox dataset (n = 124; 11 events), upfront ligation was not associated with time to variceal bleeding in an unadjusted Cox model (HR 1.24, 95% CI 0.38–4.09; p = 0.724). In the primary parsimonious model adjusting for baseline variceal grade (grade I reference; grade IV excluded), upfront ligation again showed no significant association with variceal bleeding (HR 0.65, 95% CI 0.13–3.28; p = 0.606), while baseline grade III varices were associated with a significantly increased hazard of bleeding compared with grade I (HR 8.25, 95% CI 1.23–55.31; p = 0.030); grade II was not significantly associated (HR 2.51, 95% CI 0.40–15.86; p = 0.329). 

Sensitivity analyses yielded consistent findings. After adjustment for macrovascular infiltration, upfront ligation remained non-significant (HR 0.81, 95% CI 0.21–3.07; p = 0.753), and macrovascular infiltration itself was not associated with bleeding risk (HR 0.81, 95% CI 0.19–3.45; p = 0.777). Similarly, models adjusting for Child–Pugh class or prior variceal bleeding did not identify significant predictors beyond variceal grade and did not materially change the null association between upfront ligation and time to variceal bleeding (HR 1.05, p = 0.944 and HR 1.27, p = 0.715, respectively). Using a Cox model stratified by baseline variceal grade (I–III), upfront ligation was also not significantly associated with time to variceal bleeding (HR 0.45, 95% CI 0.09–2.14; p = 0.312).

In this two-centre cohort uniformly applying bevacizumab during ongoing EVL, variceal bleeding incidence was low (4.2%) and not elevated in comparison to an non upfront ligation group. Upfront EVL itself was not independently associated with a higher bleeding incidence when adjusted to variceal grade, Child-Pugh Score or macrovascular invasion.