Economic Evaluations of AI-Assisted Technology in Healthcare: How Are We Assessing the Cost-Effectiveness of These New Therapies?

OBJECTIVES: The integration of Artificial Intelligence (AI) in healthcare is rapidly increasing. This umbrella review aimed to synthesize the findings from systematic reviews (SRs) of economic evaluations (EEs) of AI-assisted interventions to provide an understanding of the value proposition of AI use in healthcare and to draw implications for future EEs.

METHODS: PubMed, Cochrane Reviews and NICE HTA guidelines were searched to identify relevant SRs of EEs on AI-assisted interventions published in the last 10 years. Articles were included if they reported SRs of EEs of AI-assisted interventions. Data on SRs guidelines followed, databases searched, search period, inclusion/exclusion criteria, number of EE studies identified alongside details on the use of AI, disease area, EE methods, findings, and quality assessment of identified EEs were extracted for narrative analysis.

RESULTS: Total 9 SRs were included covering 115 EEs. The evaluated AI interventions involved general medicine/healthcare (4), diagnosis/screening (3) and surgery (2). The main EE methods were cost-effectiveness, cost-utility or cost-minimisation analyses and perspectives were healthcare system or payer. Traditional decision trees, Markov models or combinations were commonly used with time horizons ranging from 28 days to lifetime. Quality of the included EEs were reported as moderate. Due to study heterogeneity, insufficient clinical evidence, limitations in costing and a lack of transparent reporting on EEs especially on AI interventions, no conclusive value proposition could be identified by the SRs to inform decision making.

CONCLUSIONS: There is a growing body of SRs on the EEs of AI in healthcare. Findings suggest that while AI-assisted interventions may demonstrate economic benefits, EEs often lack methodological robustness and supportive clinical evidence which then underplays the value of these technologies. This underscores the necessity for long-term, real-world data on AI performance along with transparent reporting guidelines for EE to inform decision-making in investments in AI-assisted interventions.