AI Guidance Tool Prevents Spread of C. Difficile Infection in Hospital Settings

Clostridioides difficile (C. difficile) is a type of bacteria that can be particularly harmful to patients with compromised health. Outside the human body, C. difficile transforms into spores that can persist on surfaces for extended periods, sometimes for months. The resilient pathogen, which leads to severe diarrhea and inflammation in the gut, is not easily eliminated by many disinfectants, including alcohol-based hand sanitizers. Its persistence and the presence of susceptible patients make C. difficile a serious threat in healthcare environments.

This danger is heightened by antibiotic prescriptions, which increase the likelihood of infection in at-risk patients—making them ten times more susceptible—because antibiotics eliminate the gut’s natural microbial defenders, weakening the body's resistance to pathogens like C. difficile. Now, artificial intelligence (AI)-based tools deployed in a hospital for the first time have helped clinicians reduce the spread of this infection.

In a study led by University of Michigan (Ann Arbor, MI, USA), the new protocol deployed in a hospital setting led to a significant reduction in antibiotic prescriptions—cutting antimicrobial use by 10% to 15%. Notably, this decrease in antibiotic use did not lead to longer hospital stays, increased readmission rates, or higher mortality. Although the incidence of C. difficile showed a downward trend during the study period, it did not reach the threshold for statistical significance. This clinical application marked the culmination of a decade-long development process.

Initially, the researchers created a predictive model that analyzed past hospital data to identify patients at high risk for C. difficile infection. The machine learning algorithm was trained using data such as patient medications, laboratory results, prior hospital admissions, coexisting health conditions, demographic details, and proximity to other infected patients. When tested on a new group of patients who hadn’t been included in the original dataset, the model’s predictions matched actual infection risks, confirming its accuracy. The approach remained effective even when adapted specifically for use at Michigan Medicine.

Advancing toward practical implementation, the team conducted a real-time validation study in 2022 across two academic hospitals. The model generated immediate risk estimates for patients, and these predictions were later compared to actual infection outcomes. Following this success, the researchers developed an infection prevention strategy that would deliver live risk scores and targeted recommendations to clinicians through the hospital’s electronic health record system.

This multi-pronged strategy was developed by a team of engineers, clinicians, and hospital personnel. The guidance offered to healthcare providers included measures such as mandatory handwashing with soap and water before entering a patient’s room, limiting the use of high-risk antibiotics, and reevaluating penicillin allergies. Since many patients who were once allergic to penicillin may no longer be, reclassification could broaden the antibiotic options available—options that pose a lower risk of triggering C. difficile infections.

An intensive care nursing team also devised a novel use for the patient risk score. When assigning rooms, the charge nurse ensured that a nurse caring for an actively infected patient would not also be responsible for a high-risk patient, reducing the possibility of transmission. To evaluate the impact of the AI tool, the researchers compared data from the one-year intervention period to a baseline period before the AI system was introduced. According to results published in JAMA Network Open, C. difficile infection rates slightly decreased from 5.76 to 5.65 per 10,000 patient-days, though the change was not statistically significant. However, there was a statistically significant reduction in antibiotic usage, with patients spending 10% to 15% fewer days on antimicrobial medications.

“It’s rewarding to see an algorithm grow into something with a measurable impact at the bedside,” said Jenna Wiens, an associate professor of computer science and engineering at U-M and senior author of the study, who is in the process of handing the C. difficile monitoring project over to Michigan Medicine, as she looks forward to future AI modeling projects working towards healthcare solutions.

Related Links:
University of Michigan