Clinically deployed AI guidance for preventing C. difficile spread

This story was originally published by Computer and Science Engineering at the University of Michigan.

AI guidance for clinicians aimed at reducing the spread of Clostridioides difficile—a bacteria that can be deadly for sick patients—was deployed for the first time in a hospital setting, according to a University of Michigan-led study published in JAMA Network Open.

The protocol significantly reduced antibiotic prescriptions, a factor that increases infection risk for vulnerable patients, with 10% to 15% fewer days on antimicrobials. Importantly, reducing days on antimicrobials did not increase in length of stay, readmission rate or mortality.

The incidence of C. difficile trended downwards during the study, but did not reach statistical significance.

“The technical side of computer science has always been something I love, but it’s the potential for impact that keeps me going. 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.

When outside of the body, C. difficile forms spores that can remain on surfaces for months. 

The stubborn microbe that causes severe diarrhea and gut inflammation is resistant to many cleaning products, including alcohol-based hand sanitizer. 

This, along with the availability of sick hosts, make C. difficile particularly dangerous in hospital settings. 

In particular, hospital patients taking antibiotics are ten times more likely to get a C. difficile infection, as the medications wipe out the indigenous microbes in the gut, which normally form a barrier against invaders such as C. difficile. 

Getting to this point of measuring clinical impact has been ten years in the making. In the early stages, the research team built a predictive model to identify which patients are at the greatest risk of C. difficile infection using past hospital records. 

The machine learning model trained on factors including medications, lab results, previous hospitalizations, comorbidities, demographics and even proximity to other infected patients in the hospital.

Applied to a new set of never-before-seen patients, the predictions aligned with true patient risk, proving the model worked.

The approach held up again when a model was trained specifically for Michigan Medicine, the academic medical center affiliated with the University of Michigan.

Taking another step towards application, a prospective validation study in 2022 used the model to estimate risk in real time at two academic medical centers, Michigan Medicine and Massachusetts General Hospital, and then evaluated which patients became infected. 

The model’s success prompted the research team to work towards an infection prevention bundle that could give clinicians real-time risk predictions and  systematic recommendations, communicated through electronic health records.

A team including engineers, clinicians and hospital staff developed the approach with multiple angles of attack.

“While I had anticipated challenges in navigating the various hospital committees needed to get approval for the project, I was pleasantly surprised by the engagement and enthusiasm I observed in the medical teams, which should not be taken for granted when asking people to change how they work,” said Krishna Rao, an associate professor of internal medicine at the U-M Medical School and co-author of the study.

Guidance clinicians received included implementing hand washing protocols—requiring hand washing with soap and water before entering the room, reducing use of high-risk antibiotics and testing for penicillin allergies.

Many patients labeled as allergic to penicillin actually lose the allergy over time, opening up a new class of antibiotics that lowers the patient’s risk for getting C.difficile infection.

“Pharmacists were acutely aware of vulnerable patients. This prompted handwashing when entering the patient’s room to minimize the chance of spreading infection along with recommending medication changes, including changes to antibiotic regimens and antacid therapy which reduce the risk of getting infection,” said Jerod Nagel, a clinical pharmacist and assistant professor of pharmacy at U-M and co-author of the study.

“All the providers I worked with during this project had first-hand experience with treating patients with C. difficile infection and the horrible impact it has on lives, and wanted to help in whatever way they could if it would be of benefit,” said Rao.

A team of intensive care nurses even implemented an unexpected use for the patient risk score in their own work flow. 

When assigning rooms, the charge nurse ensured a nurse caring for a patient with an active infection was not also assigned to a high risk patient.

The research team compared the year-long intervention period with a pre-AI period to assess changes.

The incidence of C. difficile trended downwards from 5.76 per 10,000 patient-days to 5.65, but did not reduce enough to garner statistical significance.

Antibiotic prescriptions did reduce significantly with 10% to 15% fewer days on antimicrobials. 

“It goes without saying—but I’ll say it anyway—that it truly takes a team. I have been privileged to work with such a dedicated interdisciplinary team on a problem with real-world impact, and I’m proud of what we’ve accomplished,” said Shengpu Tang, a former U-M doctoral student, now an assistant professor of computer science at Emory University and the study’s first author. 

“At Emory, I look forward to continuing this line of work and exploring innovative ways AI can help improve patient care,” added Tang.

While Wiens is in the process of handing the C. difficile monitoring project over to Michigan Medicine, she looks forward to future AI modeling projects working towards healthcare solutions.

University of Michigan AI & Digital Health Innovation helped make this collaborative project possible. 

This work was funded by the Alfred P. Sloan Foundation, the Agency for Healthcare Research and Quality (R01HS02743), the National Institute of Allergy and Infectious Diseases of the National Institutes of Health (U01AI124255), and the Department of Veterans Affairs (RCS 11-222).

Paper cited: "Guiding Clostridioides difficile Infection Prevention Efforts in a Hospital Setting with AI," JAMA Network Open. DOI: 10.1001/jamanetworkopen.2025.15213

Additional authors: Shengpu Tang, Stephanie Shepard, Rebekah Clark, Erkin Ötleş, Chidimma Udegbunam, Josh Tran, Melinda Seiler, Justin Ortwine, Akbar K Waljee, Jerod Nagel, Sarah L Krein, Jacob E Kurlander, Paul J Grant, Jihoon Baang, Anastasia Wasylyshyn, Krishna Rao and Jenna Wiens