Researchers Use AI to Identify Healthcare-Based Outbreaks

By Eric Wicklund

Healthcare organizations are training an AI tool to rapidly identify outbreaks within a health system, giving clinicians more time to contain the infection and treat patients.

A four-year study in 82 hospitals across the US, recently posted in The New England Journal of Medicine, found that the automated tool reduced potential outbreaks by 64% compared to traditional methods of identifying an outbreak. The tool identified potential outbreaks, on average, three times per year per hospital.

“Outbreaks in hospitals are often missed or detected late, after preventable infections have occurred,” Meghan A. Baker, MD, ScD, an assistant professor of population medicine at Harvard Medical School’s Harvard Pilgrim Health Care Institute and lead investigator of the study, said in a press release. “This study provides a practical and standardized approach to identify early transmission and halt events that could become an outbreak in hospitals.”

Funded by the U.S. Centers for Disease Control and Prevention (CDC), the CLUSTER study was conducted in 2019-22 at hospitals within the HCA Healthcare system by a team of investigators from HCA, the Harvard Pilgrim Health Care Institute, and the University of California, Irvine (UCI) Health.

“Despite significant progress in reducing healthcare-associated infection outbreaks, including of antimicrobial-resistant pathogens, they remain an industry challenge and can present as clusters that signal potential for transmission to patients,” Joseph Perz, DrPH, MA, senior advisor for public health programs in the CDC’s Division of Healthcare Quality Promotion and a committee member for the CDC’s Council for Outbreak Response: Healthcare-Associated Infections, said in the release. “The CLUSTER trial provides evidence that early detection powered by automation tools and quick action can prevent outbreaks from growing.”

In this trial, researchers created an “algorithm-driven statistical detection tool” that combed through laboratory data for signs of more than 100 bacterial and fungal infections, then posted real-time alerts to infection control programs. The process included both an automated review of patients’ clinical cultures and a statistical assessment of whether patients with these specific infections were increasing in number.

The results of the study were affected by the COVID-19 pandemic. According to researchers, automated alerts weren’t as effective during the pandemic because hospital staff were so busy that they weren’t able to respond to the alerts in time. Researchers decided instead to focus on the results gained prior to the pandemic.

Eric Wicklund is the associate content manager and senior editor for Innovation, Technology, and Pharma for HealthLeaders.