Presentation (CTI)

OBJECTIVES: Network Meta Analysis (NMA) is an essential component of evidence generation in Health Economics and Outcomes Research (HEOR). NMA analyses supporting Joint Clinical Assessment (JCA) and rapid Health Technology Assessment (HTA) submissions, are increasingly conducted in short timelines and with increasingly complex evidence. This study aimed to assess how customized algorithms combined with artificial intelligence (AI), specifically large language models (LLMs) can be used to develop an efficient, intelligent & automated framework for NMA while remaining aligned with current best practices.
METHODS: The typical NMA workflow was decomposed into modular components. Those associated with high analyst time burden were identified. Algorithms/frameworks augmented with LLMs were developed (primarily in R software) to automate each component. Key algorithms/frameworks included: (1) a generalized framework for harmonizing data extraction tables derived from systematic literature reviews; (2) an automated study and intervention grouping algorithm with embedded feasibility checks; (3) an auto-suggestion module proposing analysis specifications consistent with best-practice guidance; and (4) automated reporting aligned with common HTA guidelines. The framework was evaluated using 6 completed NMA projects, assessing accuracy of outcomes and reductions in analyst effort.
RESULTS: The AI-enabled framework reproduced original NMA results with 100% accuracy across all evaluated projects. Substantial efficiency gains were observed, with estimated time reductions of approximately ~60-90% during early feasibility stages, particularly data harmonization, validation, and study and treatment grouping. Additionally, a potential time reduction of ~30%-50% in reported stages was estimated.
CONCLUSIONS: An AI-enabled NMA framework has the potential to improve efficiency, transparency, and consistency in NMA processes by applying customized algorithms and LLMs in key time-intensive steps. This approach can meaningfully enhance HEOR and HTA decision-making in time- and resource-constrained settings. Further development & validation is required to evaluate & exploit the capabilities of this framework.