Hospital adoption of EHR-integrated predictive AI spikes

New federal data shows that hospital adoption of predictive AI rose over a one-year period, with 71% of hospitals using predictive AI integrated into their EHR in 2024, up from 66% in 2023.

Published by the Assistant Secretary for Technology Policy/Office of the National Coordinator for Health IT (ASTP/ONC), the report examines trends in the utilization, evaluation and governance of predictive AI. The report defines predictive AI as "using statistical analysis and machine learning to classify or produce a risk score for individuals (such as readmission risk prediction, early disease detection, appointment no-show, and treatment recommendations)."

The data for the report comes from the Information Technology Supplement to the American Hospital Association Annual Survey from 2023 and 2024. The 2024 survey was fielded from April to September 2024 among 2,253 non-federal acute care hospitals with a response rate of 51%. The 2023 survey, conducted from March to August 2023, had a 58% response rate of 2,547 non-federal acute care hospitals.

The report shows that medium, large and non-critical access hospitals used predictive AI at higher rates compared to small and critical access hospitals. Similarly, system-affiliated and urban hospitals used predictive AI at higher rates than rural and independent hospitals.

Further, hospitals' use of predictive AI varied by EHR vendor. In 2024, a majority of hospitals (90%) using the market-leading EHR vendor used predictive AI versus 50% of hospitals using other vendors.

Not only that, but hospitals were more likely to use predictive AI provided by their EHR vendor. Most hospitals (80%) used predictive AI sourced from their EHR developer in 2024, while 52% used AI developed by a third party and 50% used AI that they developed.

From 2023 to 2024, hospitals increased their use of predictive AI primarily to simplify or automate billing procedures, facilitate scheduling and identify high-risk outpatients to guide follow-up care. However, the source of the predictive AI also played a role in which use case hospitals focused on.

For instance, the use of predictive AI for simplifying or automating billing was higher among hospitals using third-party or self-developed AI (73%) compared to AI sourced from their EHR developer (58%). Additionally, the use of predictive AI to identify high-risk outpatients increased at higher rates among hospitals using third-party or self-developed AI compared to those using EHR vendor-developed AI.

Alongside growing rates of predictive AI use, a majority of hospitals reported evaluating these models. In 2024, 82% of hospitals evaluated predictive AI for accuracy, 74% assessed the models for bias and 79% conducted post-implementation evaluation or monitoring.

Additionally, in 2024, most hospitals had multiple governance entities for predictive AI. The most-reported entities responsible for evaluating predictive AI were a specific committee or task force (66%) and division/department leaders (60%). The least-reported AI evaluation entities were IT staff.

AI governance is proving vital as the technology becomes increasingly integrated into healthcare delivery. Not only are health systems establishing internal AI governance committees and frameworks, but they are also entering into collaboratives to create industry-wide guardrails and guidance. 

"As more resources are becoming available to improve AI governance, it will be critical to understand how different frameworks contribute to effective evaluation and monitoring practices," the report concluded.

Anuja Vaidya has covered the healthcare industry since 2012. She currently covers the virtual healthcare landscape, including telehealth, remote patient monitoring and digital therapeutics.