Use of predictive AI in hospitals is growing

Wouldn’t it be great if we had a crystal ball?

Well, we don’t. (Bummer!) But hospitals are using what they hope is the next best thing: predictive AI. Instead of magic, it uses statistical analysis and machine learning to analyze patterns in order to forecast the future.

In 2024, 71% of surveyed hospitals reported using predictive AI integrated into their electronic health records, up from 66% in 2023, per a September data brief by the Department of Health and Human Services’s Assistant Secretary for Technology Policy (ASTP), analyzing data from the 2023 and 2024 American Hospital Association (AHA) information technology supplement survey.

And the investments are still coming. Healthcare spend on this and other “digital-first” strategies may shift $1 trillion away from other healthcare spending by 2035, according to consulting firm PwC. 

“Predictive AI is definitely here to stay,” Julia Croxen, VP of strategy consulting at digital health strategy group Rock Health Advisory, told Healthcare Brew.

AI’s use cases grow. Hospitals that used predictive AI in 2024 reported most frequently using it to forecast health trajectories for inpatients (92%) and identify high-risk outpatients in need of intervention (87%), per the ASTP brief.

Croxen pointed to the tool’s effectiveness in predicting sepsis, which occurs when a patient’s immune system has an extreme, potentially deadly reaction to an infection.

An October systematic review in the journal BMC Infectious Diseases found machine learning and deep learning could predict sepsis earlier than traditional diagnostic methods in patients, though performance depended on a model’s data quality.

That’s old news, though. Hospitals have long used mathematical models to predict patient outcomes.

“Predictive [AI] isn’t necessarily new,” Croxen said. “But [it] has definitely gained more focus as AI awareness has intersected with social trends and dynamics in this space.”

The big news is predictive AI’s uses are expanding. In 2023, 36% of hospitals reported using it for simplifying or automating billing procedures. In 2024, that jumped to 61%, according to the ASTP brief. This entails tasks like predicting claim denials, per an AHA fact sheet.

Gaps remain. Not all AI is created or distributed equally, however.

The ASTP data echoes other reports in suggesting that small, rural, government-owned as well as critical access hospitals aren’t adopting AI as quickly as others, likely due to a lack of resources: 96% of large hospitals compared to 59% of small hospitals used it in 2024.

Another limitation is that some forms of predictive AI are unproven—or even worse, seem to overlook some at-risk patients, according to a March study in the journal Communications Medicine.

Croxen describes the healthcare industry’s previous attitude as “a race” to apply AI in as many ways as possible, but says leaders are becoming more discerning. Sometimes the most expensive or technologically complex solution isn’t the right one.

“Not every problem requires AI as a solution, and right now, I think we’re starting to see that acknowledgement,” Croxen said.