Operational AI Adoption in US Inpatient Rehabilitation Facilities: Organizational Selection and Domain Heterogeneity, 2021-2024.

OBJECTIVE

To examine inpatient rehabilitation facilities (IRFs), market, and regional characteristics associated with operational artificial intelligence (AI) adoption across 3 functional domains and breadth of use.

DESIGN

Retrospective cohort study using multivariable linear probability models.

SETTING

IRFs across the United States.

PARTICIPANTS

A total of 276 IRFs (N=766 hospital-years of data) that responded to the American Hospital Association Annual Survey (with a mean response rate of 58%).

INTERVENTIONS: Not applicable (observational study).

MAIN OUTCOME MEASURES: Any operational AI use; domain-specific use (predicting patient demand, staff scheduling, and optimizing operational efficiency); breadth of adoption.

RESULTS

Operational AI use more than doubled from 11.1% in 2021 to 24.9% in 2024, led by system-affiliated, nonprofit, and teaching facilities. Organizational structure explained 50%-67% of adoption variance, exceeding the combined role of market conditions, volume, and region. Optimizing operational efficiency was the dominant domain, reaching 20.0% by 2024, compared with 8.1% for predicting patient demand and 7.0% for staff scheduling. Staff scheduling declined after 2022, suggesting domain-specific implementation barriers beyond resource constraints.

CONCLUSIONS

Operational AI in IRFs is spreading along the same organizational lines as earlier health information technology, concentrating in system-affiliated, nonprofit, and teaching facilities. Reaching smaller, independent, and for-profit IRFs will require targeted financing and shared infrastructure. Staff scheduling tools lag the most, so widespread use of these tools will depend on clear rules and clinician input on how algorithms guide staff schedules.