New research indicates that an artificial intelligence (AI)-driven infection prevention tool did not significantly reduce cases of Clostridioides difficile (C. difficile) infection in a hospital setting. Nevertheless, its application did result in the reduced use of antimicrobials that contribute to the risk of infection. These insights were part of a study conducted by Jenna Wiens, PhD, an Associate Professor of Computer Science and Engineering at the University of Michigan.
The study highlighted how risk models developed using machine learning technology can be seamlessly integrated into clinical workflows to guide antimicrobial stewardship. However, Wiens cautioned that the manner of integration is critical, as significant alterations to existing workflows may not yield the desired results. Clostridioides difficile was named one of the top five urgent threats in 2019 by the Centers for Disease Control and Prevention (CDC) and continues to be prevalently stubborn in hospital environments.
The AI-based model developed by Wiens and her team, harnessing data from electronic health records, can predict patients at risk for C. difficile about five days before infection occurs. However, they sought to confirm if the model could realistically impact patient outcomes in practical clinical environments, leading to their research on this AI tool implementation. In an attempt to drive infection prevention efforts at Michigan Medicine, they integrated the risk model into their system.
The study was a prospective, quality improvement investigation assessing adult patient hospitalizations before and after AI tool implementation. It primarily focused on preventing infection in high-risk hospitalizations by promoting handwashing and antimicrobial oversight to limit pathogen exposure and host susceptibility. These preventive measures were integrated through best practice advisory alerts, medication reviews by inpatient pharmacists, and health record assessments carried out by study physicians.
The study found a non-significant reduction in C. difficile infection cases following the implementation of the AI tool, but a significant drop was observed in the use of certain antimicrobials. Moreover, staff interviews and field observations noted a ‘limited and potentially inadequate’ use of some preventive strategies, pointing towards challenges in workflow adaptation.
Wiens acknowledged limitations in their study as a clinical trial could not be conducted in a single-hospital setting due to the high transfers of patients and staff. She suggests more extended study periods for observing a statistically significant difference in the incidence of C. difficile.