Hospital Artificial Intelligence Tools and Inpatient Utilization and Costs in Older Adults with Alzheimer's Disease and Related Dementias

BACKGROUND: Hospitals are increasingly adopting artificial intelligence and machine learning (AI/ML) tools to support clinical decision making and care management. However, evidence on how hospital AI/ML adoption relates to inpatient utilization and spending among clinically complex populations remains limited. Older adults with Alzheimer's disease and related dementias (ADRD) experience higher rates of readmissions and potentially avoidable hospitalizations. METHODS: Cross-sectional study was conducted using 2023 inpatient claims linked to the Medicare Beneficiary Summary File and the American Hospital Association Annual Survey Information Technology Supplement to examine associations between hospital adoption of patient-related AI/ML tools and inpatient utilization and spending among Medicare fee-for-service (FFS) beneficiaries with ADRD. The study included 340,509 FFS beneficiaries aged 65 years or older with ADRD who experienced at least one inpatient hospitalization in 2023. Hospital adoption of patient-related AI/ML tools was measured using four indicators reflecting AI/ML use to predict inpatient health risks, identify high-risk outpatients, monitor patient health, and recommend treatments. Outcomes included frequent hospitalization, 30-day readmission, preventable acute and chronic hospitalizations, total Medicare payments, and beneficiary out-of-pocket (OOP) spending. Multivariable regression models adjusted for beneficiary and hospital characteristics. RESULTS: Greater hospital adoption of patient-related AI/ML tools was associated with lower odds of frequent hospitalizations, 30-day readmissions, and preventable acute hospitalizations. Inpatient risk prediction and high-risk outpatient identification tools were consistently associated with lower inpatient utilization. Inpatient risk prediction was associated with lower total Medicare spending, while high-risk outpatient identification was associated with higher Medicare spending. Treatment recommendation tools were associated with higher beneficiary OOP spending. CONCLUSIONS: Among Medicare FFS beneficiaries with ADRD, hospital adoption of patient-related AI/ML tools, particularly those focused on risk prediction and high-risk patient identification, was associated with lower inpatient utilization without increasing overall spending. Heterogeneity across AI/ML tool types suggests the importance of evaluating AI/ML tools based on their specific clinical functions.