Abstract
Artificial intelligence (AI) technologies are increasingly integrated into healthcare, yet their economic value remains uncertain. Traditional economic evaluation methods may not adequately capture the unique features of AI, including dynamic model evolution, scalability, and broader societal impacts. This systematic review synthesized existing evidence on the cost-effectiveness of AI-based healthcare interventions and assessed the methodological rigor of published studies. A comprehensive search identified health economic evaluations of AI applications published between September 2019 and March 2025, following PRISMA and SWiM guidelines and registered in PROSPERO (CRD42025641230). Eligible studies were full economic evaluations comparing AI-based interventions with non-AI alternatives, and data were extracted on study characteristics, analytical methods, decision-analytic models, perspectives, outcomes, and AI-specific costs. Methodological quality was evaluated using the CHEERS checklist. A total of 52 studies from 15 countries were included, most published after 2020, focusing on diabetic retinopathy screening, cancer detection, and cardiovascular disease applications. Cost-utility analysis was the predominant method (79%), followed by cost-effectiveness analysis (15%). Nearly all studies (98%) concluded that AI-based strategies were cost-effective, cost-beneficial, or cost-saving. However, reporting of AI-specific costs was inconsistent, while over 90% of studies detailed expenses such as software licensing, per-test charges, or maintenance fees, some omitted cost information entirely, limiting comparability. Overall, AI-based healthcare interventions are generally reported as cost-effective, but methodological heterogeneity, incomplete cost reporting, and potential publication bias constrain the reliability and comparability of current evidence. Standardized economic evaluation frameworks that incorporate comprehensive cost structures and account for the evolving nature of AI are urgently needed.