Presentation (CTI)

OBJECTIVES: Immune checkpoint inhibitors (ICIs) have revolutionized the treatment of hepatocellular carcinoma (HCC). However, serious adverse events (SAEs) remain insufficiently characterized. This study aims to utilize an AI-driven pharmacovigilance approach to complementing RCTs to advance the understanding of SAEs associated with ICIs in treatment of HCC. This includes characterizing toxicity spectrum, estimating relative risk, and examining clinical outcomes.
METHODS: RCTs were used to compare SAE risks across ICI-based regimens. Pharmacovigilance analysis identified rare and severe toxicities, and evaluated heterogeneity in SAE-relative-risks between RCTs and real-world clinical settings. Multilayer Perceptron models, SHapley Additive exPlanations (SHAP), and Accumulated Local Effects (ALE) were employed to identify features contributing to fatal outcome of SAEs.
RESULTS: Thirty-six adverse events were identified in pharmacovigilance analyses but not captured in RCTs, including tumour hyperprogression and liver transplant rejection. RCT data showed anti-PD-L1 + anti-VEGF-A had the highest relative risk of acute kidney injury (SUCRA=84.60), and cardiac complications (65.9-70.7). In contrast, pharmacovigilance data suggested a distinct real-world profile: the highest relative risk of acute kidney injury (signal intensity=3.79) and myocardial infarction was associated with anti-PD-1 + TKI (7.05).
AI-based outcome classification modelling demonstrated that age (SHAP φ = 0.13), sex (0.04), and body weight (0.06) were major determinants of mortality. Mortality risk was elevated in females receiving anti-PD-1 + anti-CTLA-4 (ALE = 0.006) or anti-PD-L1 + anti-VEGF-A (0.006). Anti-PD-L1 therapy exhibited weight-dependent modulatory effects on mortality risk. Anti-PD-L1 + anti-VEGF-A + TKI was associated with increased vascular (ALE=0.030) and procedural complications (0.029), contributing to high risk of mortality.
CONCLUSIONS: Pharmacovigilance provides critical complementary evidence to RCTs. Discrepancies in relative-risk profiles between RCT and real-world settings underscore the importance of incorporating baseline comorbidities and medication strategies in safety assessment. Sex and weight were identified as drivers of fatal outcomes, which highlight the importance of risk stratification. Judicious planning of drug‑sequencing intervals was necessary.