OBJECTIVES: While there is ample research on the cost-effectiveness of AI/robotic treatment methods, there is a limited understanding of the use of machine learning (ML) methods in health economics and outcomes research (HEOR). We systematically reviewed peer-reviewed literature on machine learning and HEOR to understand the intersection of the two fields, identify trends, and suggest areas for future investigation

METHODS: A systematic search was conducted in Medline and Embase databases for the period between January 2004 and September 2021. Title and abstract screening were performed by two independent reviewers per PRISMA guidelines. The text of included studies mentioned both ML and cost-effectiveness/health economics.

RESULTS: Of the 861 articles screened, 43 articles were included in the review. The two most common AI/ML application areas in HEOR were prediction/classification (51%) and automated image analysis (35%). The most common study objectives included: improving clinical pathway and operational efficiency (33%), investigating the cost implications of automated image analysis for screening, detection, or diagnosis of a disease (33%), and predicting disease and treatment outcomes (19%). Supervised learning methods such as Neural Networks, Decision trees, and Support Vector Machines were the most commonly utilized ML methods. Image analysis and electronic health records were the most commonly used data sources. The studies were conducted in a range of disease areas, with cardiovascular (16%), infectious (12%), and ophthalmic diseases (12%) being the dominant areas. The most commonly used health economic analysis was cost-minimization (44%).

CONCLUSIONS: Machine learning can potentially influence HEOR, but the current applications remain limited in scope and depth. Most of the economic analyses were simple cost-minimization analyses. Future studies can explore emerging areas, including digital interventions, wearable technologies, and social media data.