Dive Brief:
- Hospitals are increasingly using predictive artificial intelligence in healthcare delivery, but small, rural, independent and critical-access hospitals are lagging behind, according to data released Wednesday by the Assistant Secretary for Technology Policy/Office of the National Coordinator for Health IT.
- Use of AI rose from 2023 to 2024 across hospital types, according to the analysis. Still, some facilities are adopting the technology at much lower rates. For example, 86% of hospitals affiliated with health systems reported using predictive AI last year, compared with just 37% of independent facilities.
- The data suggests a “persistent digital divide” in hospitals’ use of predictive AI, the researchers wrote.
Dive Insight:
Predictive AI, which uses machine learning to predict future outcomes like risk of readmissions, has been used in the healthcare sector for years.
However, adoption of the technology has grown significantly over the past decade, according to the analysis, which used survey data from the American Hospital Association.
Last year, 71% of non-federal acute care hospitals reported using predictive AI integrated into their electronic health records, a notable bump up from the 66% who said they’d adopted the tools in 2023.
Hospitals’ use of AI predictive technology increased significantly year over year in three use cases: simplifying or automating billing procedures, helping with appointment scheduling and identifying high-risk outpatients for follow-up care.
However, using predictive AI to monitor health and recommend treatments is still less common, possibly due to the high risk of errors, the researchers wrote. Hospitals might adopt these tools more frequently as they become more comfortable with using the technology for non-administrative purposes, they added.
Still, predictive AI adoption is uneven — and some hospitals appear to be falling through the cracks. Only half of critical access hospitals, small facilities that are located at least 35 miles from another hospital, used predictive AI last year, compared with 80% of non-critical access hospitals.
Additionally, only 56% of rural hospitals reported using the tools, compared with 81% of urban facilities.
Adopting AI tools can be challenging for providers, given the high stakes of inaccuracies and the amount of labor needed to manage the tools. For example, health systems need to set up governance structures on AI use, and continually monitor the tools in case their performance begins to decline.
Most hospitals using the technology are evaluating their predictive AI tools, according to the ASTP report. Last year, 82% evaluated their AI for accuracy, 74% checked the tools for bias and 79% conducted post-implementation evaluation or monitoring.
Most hospitals report evaluating predictive AI for accuracy, bias
A number of people are often involved in this work, according to the analysis. Nearly three-quarters of hospitals said multiple entities were accountable for evaluating predictive AI, with a quarter reporting four or more entities were responsible. Specific task forces or committees and division and department leaders were mostly commonly in charge of evaluating predictive AI, according to the data.
The study stands in contrast to other analyses researching hospitals’ readiness to implement other forms of AI, like Generative AI, which can create new original content like texts or images. There, researchers have found that few pilots have been fully implemented so far.
