OBJECTIVES: This study examines the relationship between the level of artificial intelligence (AI) maturity and completeness of health economic evaluations (HEEs). Recent reviews reveal inadequate quantity and quality of HEEs on AI and limited evidence to assess the feasibility of AI solutions. Concerns arise regarding the maturity of evaluated AI technologies due to early-stage development and limited data availability, potentially leading to suboptimal HEEs.

METHODS: Following PRISMA 2020 guidelines, a comprehensive search was conducted in six databases (EMBASE, MEDLINE, Web of Science, Cochrane, NHS-EED, and Google Scholar). Maturity of AI applications was assessed using the Technology Readiness Level (TRL), a scale (range: 1-9) that gauges the maturity and readiness for implementation of a technology. HEE quality was evaluated using CHEERS checklist compliance rate (range: 0-100%).

RESULTS: Of 6,503 articles, 73 met the inclusion criteria, with most published between 2020 and 2022 (65%). The primary application of AI was for prevention and screening (39%). TRL scores ranged from 4 to 9, with 77% falling within levels 4-5. The most common HEE approach was CMA (41%), followed by CUA (26%) and CEA (25%). Average CHEERS score was 61%, and no correlation with TRL level was found. Significant proportion of HEEs omitted implementation (83%) and running costs (62%); and HEEs with lower TRL AI were more likely to exclude these costs (OR=8.85, [95% CI: 1.91, 45.52] and OR=4.35, [95% CI: 1.15, 18.69]; respectively).

CONCLUSIONS: Evaluated technologies are primarily in early development-stages, potentially limiting data availability and comprehensive assessments. TRL should be reported in AI HEE assessments, as it provides a snapshot of technology maturity at a given point in time. For HEEs of technologies with lower TRL, estimation of implementation and running costs is crucial, combined with sensitivity and scenario analyses for realistic cost-effectiveness assessment. The exclusion of these costs in the analysis must always be disclosed.