Presentation (CTI)

OBJECTIVES: Indirect treatment comparisons (ITCs) are essential when head-to-head trials are unavailable but require substantial effort for feasibility assessment, selection of appropriate analyses, and clear interpretation of results. This study evaluated the use of Generative Artificial Intelligence (GenAI) to support the ITC workflow from feasibility assessment through analysis planning and result interpretation in first-line hepatocellular carcinoma (HCC). Statistical analyses were conducted using an R Shiny-based platform supporting Bayesian and frequentist network meta-analysis (NMA).
METHODS: Published randomized controlled trials in first line HCC were included and data was extracted in structured Excel format capturing trial design, treatments, patient characteristics, and outcomes. GenAI was first applied to assess ITC feasibility by evaluating network connectivity, availability of common comparators, and similarity of trial populations and endpoints. Based on the feasibility assessment, GenAI then recommended appropriate analytical approaches. NMA were conducted using an in-house R Shiny tool with predefined model settings. Finally, GenAI was used to interpret the statistical outputs and support structured reporting of comparative effectiveness results. All GenAI-generated feasibility assessments and reports were reviewed by experienced ITC experts to evaluate usability and accuracy.
RESULTS: GenAI identified a connected evidence network of 12 trials and confirmed the feasibility for the Bayesian NMA. Bayesian NMA results showed significant overall survival benefits for sintilimab plus bevacizumab (HR 0.38; 95% CrI 0.21-0.79), camrelizumab plus rivoceranib (HR 0.42; 0.21-0.82), and atezolizumab plus bevacizumab (HR 0.45; 0.23-0.87) compared with placebo. Treatment rankings favored combination immunotherapy-antiangiogenic regimens. Expert review indicated that GenAI-generated feasibility assessments and reports were 80-90% usable, with approximately 10% requiring human refinement, mainly for clinical nuance and contextual interpretation.
CONCLUSIONS: GenAI can streamline ITC workflows by supporting feasibility assessment, analysis planning, and result interpretation, and when combined with standard NMA tools and expert review enhances efficiency while preserving rigor and enabling faster evidence generation for HTA and JCA use cases.