A dual-level Bayesian network meta-analysis of first-line immunotherapy in advanced hepatocellular carcinoma: strategy-level and regimen-level evidence.

Phase III trials of first-line immunotherapy-based combinations for advanced hepatocellular carcinoma (HCC) have yielded heterogeneous results, and head-to-head comparisons remain scarce. We evaluated mechanism-based strategy classes and individual regimens using a dual-level Bayesian network meta-analysis to provide a structured interpretation of the current evidence. We systematically searched the literature up to December 28, 2025, and identified 15 phase III randomised controlled trials. A dual-level Bayesian network meta-analysis was performed using separate models at the strategy level (PD-(L)1 plus VEGF monoclonal antibody (VEGF(Ab)), PD-(L)1 plus tyrosine kinase inhibitor (TKI), PD-(L)1 plus CTLA-4, and PD-(L)1 monotherapy) and the regimen level. Outcomes included overall survival (OS), progression-free survival (PFS), objective response rate (ORR), and grade ≥3 treatment-related adverse events (TRAEs). Fifteen trials comprising 9,792 patients were included. At the strategy level, PD-(L)1 plus VEGF(Ab) showed the most favourable OS effect versus TKI monotherapy and the highest efficacy-oriented ranking, followed by PD-(L)1 plus CTLA-4 and PD-(L)1 plus TKI. PD-(L)1 plus TKI also improved PFS but was associated with a higher toxicity and discontinuation burden, whereas PD-(L)1 monotherapy showed the most favourable tolerability profile. At the regimen level, atezolizumab plus bevacizumab, sintilimab plus IBI305, camrelizumab plus rivoceranib, and STRIDE showed OS benefit versus sorafenib. In this evidence network, PD-(L)1 plus VEGF(Ab) showed the most efficacy-oriented profile and was supported by relatively consistent regimen-level estimates. However, the clinical interpretation of strategy-level findings differed across classes: PD-(L)1 plus TKI required closer regimen-level assessment because of greater within-class variability and a less favourable toxicity profile, whereas PD-(L)1 plus CTLA-4 and PD-(L)1 monotherapy showed distinct benefit-risk patterns. This dual-level framework complements regimen-centric evidence by distinguishing strategy-level patterns from regimen-level findings and by supporting a structured interpretation of efficacy-toxicity trade-offs in first-line ICI-based therapy for advanced HCC. https://www.crd.york.ac.uk/PROSPERO/view/CRD420251273309, identifier CRD420251273309.