OBJECTIVES: Artificial intelligence (AI) enables precision medicine (PM) via intensively learning big health data to diagnose and/or tailor treatment. We reviewed the value for money and economic evaluation (EE) methods of AI-PM tools, and identified sources of value heterogeneity.

METHODS: We searched Embase, Medline, Web of Science and Tufts Registry for studies between 2013 and 2023. All types of EEs were included, and data were extracted from cost-effectiveness profiles and evaluation methods. The ECOBIAS checklist was used for quality assessment. For cost-utility analyses (CUAs), incremental net monetary benefits (ΔNMBs) were calculated using the willingness-to-pay threshold of one-time GDP per capita of each study country, with costs inflated and converted to 2023 US dollars. Next, we performed subgroup and regression analysis with generalized linear mixed models on Δcosts, ΔQALYs, and ΔNMBs to identify key value drivers.

RESULTS: Among 48 included EEs of AI-PM, the most common tool was digital diagnostics (66%), with 80% being conventional EEs for market access and 20% early EEs for guiding product development. The healthcare perspective (76%) and lifetime horizon (46%) were frequently applied. Most studies found AI-PM to be cost-saving (64%) or cost-effective (20%). Median ΔQALYs, Δcosts, and ΔNMBs per person were 0.006, $-26, and $212, respectively. AI-PM had the highest ΔNMB when used for disease control and clinical risk prediction (median, $3968, $686). Regression analysis indicated that AI-PM developed or applied in high-income countries brought higher ΔNMBs, and EEs funded by private-for-profit funders showed lower ∆costs of AI-PM.

CONCLUSIONS: Despite rapid development of AI-PM, EEs are limited in number. Generally, AI-PM was reported as cost-effective or cost-saving. Major sources of value heterogeneity included AI-PM types, country settings, and funder types. Future studies should investigate the challenges and provide recommendations for unbiased and standardized EEs of AI-PM.