Safety and efficacy of atezolizumab with bevacizumab (A+B) for hepatocellular carcinoma (HCC) in patients with Child-Pugh (CP) class B cirrhosis.

e16219 Background: A+B is a first-line treatment with demonstrated safety and efficacy for patients with advanced HCC and CP class A (CPA) cirrhosis. Less is known about the tolerance and outcomes of this treatment for patients with CP class B (CPB) cirrhosis, a population with a more limited prognosis. Methods: We performed a retrospective cohort study using EHR records at the University of California, San Francisco from 2015 to 2025. Using drug administration data and ICD-10 codes, we identified 39 patients with a diagnosis of advanced HCC who received first-line treatment with A+B. CP class was defined at the start of treatment using averaged data in the 30 days prior to treatment start. A validated large-language-model (LLM) pipeline was used to grade and extract treatment-emergent adverse events (TEAEs) from unstructured clinical notes, and additional lab abnormalities were extracted and graded from lab values. Adverse events related to worsening liver function tests (LFTs) were counted separately. Median overall survival (OS) was estimated with Kaplan-Meier methods, and hazard ratios were estimated with Cox-proportional hazard models with CP class as a single covariate. Results: The distribution of CP classes was 46.2% CPA (18 patients), 51.3% CPB (20 patients), and 2.6% CPC (1 patient), who was excluded from this analysis. The majority of patients experienced at least one grade ≥3 (≥G3) TEAE regardless of CP class (CPA 72.2% vs. CPB 75.0%), with similar average numbers of unique TEAEs per patient (CPA 2.3 [95% CI, 1.6 – 3.0] vs. CPB 2.2 [95% CI, 1.6 – 2.9]). However, patients with CPB cirrhosis tended to experience more episodes of ≥G3 TEAEs per 100 days on treatment (4.7 [95% CI, 3.8 – 5.7] vs. 3.6 [95% CI, 2.7 – 4.4]). The most common ≥G3 TEAEs in the CPB group were lymphopenia (30%), hypoalbuminemia (20%), anemia (20%), diarrhea (15%), and hepatic failure (15%), and in the CPA group abdominal pain (17%) and lymphopenia (17%). TEAEs ≥G3 related to hemorrhage/thromboembolism were only observed in the CPB group, for 3 of 20 patients. Worsening ≥G3 LFTs after treatment start occurred more frequently in the CPB group (55.0% vs. 38.9%), and there was a trend towards higher rates of ED visits and hospitalizations on treatment for CPB patients (1.5 [95% CI, 0.9 – 2.0] vs. 0.8 [95% CI, 0.4 – 1.3]). Median OS from time of treatment initiation was higher for CPA patients (22.1 months [95% CI, 5.3 months to not estimable] vs. 7.9 months [95% CI, 4.8 months to 15.7 months]; HR 2.5 [95% CI, 0.1 – 1.8], p = 0.03). Conclusions: Compared to CPA patients receiving A+B, CPB patients experienced a comparable number of unique TEAEs, but experienced episodes more frequently. This study augments knowledge of A+B TEAEs and outcomes in CPB patients, an understudied but important population, and demonstrates the utility of an LLM pipeline in facilitating retrospective real-world outcomes analysis in this population.