Hospitals across the U.S. are adopting AI to predict patient needs and streamline care, but early use of the technology remains uneven, with adoption clustering in certain regions and lagging in communities with the greatest healthcare access needs.
A nationwide analysis published Jan. 15 in Nature Health found that hospitals’ use of AI-based predictive models in 2023 and 2024 formed clear “hotspots” and “coldspots” across the country. Hospitals serving areas with provider shortages or medical underservice were less likely to report using predictive AI. The study analyzed survey data from 3,560 hospitals that responded to the American Hospital Association’s IT Supplement in 2023 or 2024, alongside community-level indicators and CMS hospital quality metrics.
Here are five key findings from the analysis:
- Among hospitals that responded to the IT survey, 48.8% reported adopting AI-based predictive models, 16% reported using predictive models that were not AI-based, and 29.3% reported using no predictive models. Another 5.9% did not report an implementation status and were treated as nonusers in subsequent analyses, according to the study.
- Hospitals most commonly reported using predictive models to forecast patient health trajectories, and most of those models were developed by hospitals’ electronic health record vendors. However, many hospitals did not report evaluating model performance: 47.4% indicated no evaluation or did not respond regarding model accuracy, and 51.4% indicated no evaluation or did not respond regarding bias.
- Adoption varied by geography. The South Atlantic Census Division had the highest predictive AI implementation score, while the West South Central Division had the lowest. At the state level, South Carolina and South Dakota had the highest scores, while Idaho and Montana had the lowest.
- When researchers compared AI adoption with measures of community need, disparities were sharpest in areas facing healthcare workforce shortages and access constraints. In analyses stratified by indicators such as Health Professional Shortage Areas and medically underserved areas, hospitals in higher-need locations were consistently less likely to report using AI-based predictive models.
- The analysis also found that AI implementation was tightly clustered, driven in part by where hospitals are located. Researchers reported a consistent concentration of predictive AI implementation in metropolitan regions, and separate testing showed significant clustering of hospitals themselves.
The authors described the results as a baseline snapshot of early hospital AI deployment in 2023 and 2024 and noted several limitations. Only 58.4% of hospitals registered with the AHA responded to either the 2023 or 2024 IT Supplement, which could introduce selection bias. The study also relied on cross-sectional reporting of predictive AI implementation and lacked precise timing of adoption, limiting the ability to connect implementation to outcomes over time.
Still, the researchers said mapping hotspots and coldspots can help identify where adoption is concentrated — and where it lags — as AI technologies move further into clinical workflows. They called for standardized, detailed, model-specific metrics of AI implementation and approaches that account for local contexts rather than uniform deployment strategies.
At the Becker's 11th Annual IT + Revenue Cycle Conference: The Future of AI & Digital Health, taking place September 14–17 in Chicago, healthcare executives and digital leaders from across the country will come together to explore how AI, interoperability, cybersecurity, and revenue cycle innovation are transforming care delivery, strengthening financial performance, and driving the next era of digital health. Apply for complimentary registration now.
