Budget Impact Analysis of Machine Learning Models for Diagnosing Alzheimer's Dementia

OBJECTIVES: To investigate the potential of machine learning (ML) as a cost-effective approach for diagnosing Alzheimer's dementia (AD), and to estimate the budget impact of utilizing ML versus standard diagnostic approaches from the perspective of Medicare.

METHODS: A budget impact model (BIM) was developed to assess the cost difference of using ML to assist AD diagnosis compared to standard AD diagnosis from the perspective of Centers for Medicare and Medicaid Services. A bespoke neural network model (NNM), developed in previous research to inform model variables specific to the application of ML for AD diagnosis, was used to determine the most cost-effective ML diagnosis approach, resulting in no diagnostic imaging being used for the ML approach. All other model probabilities and epidemiology inputs were sourced from systematic literature review and remained the same in both. Billing codes and costs were sourced from the April 2023 Medicare cost addendum and the 2019 American Psychological Association Service billing and coding guide (cost adjusted to 2023 using CPI data).

RESULTS: Implementing ML for diagnosing AD in an estimated 3.8 million Medicare patients per year could lead to a 36% reduction in Medicare's diagnosis-related AD costs, saving an estimated $6.08 billion annually. This translates to a potential savings of $7.71 per-member-per-month.

CONCLUSIONS: Using ML to help diagnose AD reduces the emphasis on imaging, while providing similar diagnostic accuracy, resulting in a cost reduction to the payer. Imaging has benefits in certain circumstances, but it may not always be cost-effective to perform. With advancements in ML, and barriers existing to equitable access to health care for the Medicare population, this study proves to be an example of how ML can be leveraged to address health inequities by providing timely diagnostic insights that could improve patient outcomes.