Predicting C. Diff Risk with Big Data and Machine Learning
A new model analyzes a wealth of information to better predict which patients are more prone to the dangerous infection.
Nearly 30,000 Americans die each year from an aggressive, gut-infecting bacteria called Clostridium difficile. Resistant to many common antibiotics, C. diff can flourish when antibiotic treatment kills off beneficial bacteria that normally keep the deadly infection at bay.
But doctors often struggle to determine when to take preventive action.
New machine learning models tailored to individual hospitals could offer a much earlier prediction of which patients are most likely to develop C. diff, potentially helping stave off infection before it starts.
The models are detailed in a paper published recently in Infection Control and Hospital Epidemiology.
Developed by researchers at the University of Michigan, Massachusetts General Hospital and MIT, the models can predict a patient’s risk of developing C. diff much earlier than current methods.
“C. difficile is one of the most problematic health care-associated bugs,” says Jenna Wiens, Ph.D., an assistant professor of computer science and engineering at the University of Michigan and a senior author on the paper. “Despite best efforts, hospitals have had little success in reducing incidence of infections.”
That’s where the new model can play a vital role.
“This represents a potentially significant advance in our ability to identify and ultimately act to prevent infection with C. difficile,” says study co-author Vincent Young, M.D., Ph.D., the William Henry Fitzbutler Professor in the Department of Internal Medicine at U-M and the study’s co-author.
“The ability to identify patients at greatest risk could allow us to focus expensive and potentially limited prevention methods on those who would gain the greatest potential benefit.”
A complex approach
Most previous models of predicting C. diff infection risk were designed as “one-size-fits-all” approaches, the authors note. Those models included only a few risk factors in their calculations, which limits their utility.
Given the diversity of health care systems and their demographics, taking such variations into account is crucial.
“When data are simply pooled into a one-size-fits-all model, institutional differences in patient populations, hospital layouts, testing and treatment protocols, or even in the way staff interact with the EHR can lead to differences in the underlying data distributions and ultimately to poor performance of such a model,” Wiens says. “To mitigate these issues, we take a hospital-specific approach, training a model tailored to each institution.”
Along with their colleagues, co-lead authors Jeeheh Oh, a U-M graduate student in computer science and engineering, and Maggie Makar, M.S., of MIT’s Computer Science and Artificial Intelligence Laboratory, took a “big data” approach that analyzed the entire electronic health record (EHR) to predict a patient’s C. diff risk throughout the course of hospitalization.
The analysis reviewed the EHRs of almost 257,000 patients admitted to either Massachusetts General Hospital or to Michigan Medicine’s University Hospital over periods of two and six years, respectively.
Using their machine-learning model, the investigators analyzed anonymous patient data — including demographics and medical history, details of admission and daily hospitalization, and the likelihood of exposure to C. diff.
Applied in other settings, their method allows for the development of institution-specific models that could accommodate different patient populations, different EHR systems and factors specific to each institution.
Timely, targeted results
The prediction model generated daily risk scores for each individual patient that, when a set threshold is exceeded, classify patients as high-risk.
Overall, the models were highly successful at predicting which patients would ultimately be diagnosed with C. diff.
And they were determined in a shorter period of time: In half of those who were infected, accurate predictions could have been made at least five days before diagnostic samples were collected, thus allowing highest-risk patients to be the focus of targeted antimicrobial interventions.
If validated in prospective studies, the risk prediction score could guide early screening for C. diff. For patients diagnosed earlier in the course of disease, initiation of treatment could limit the severity of the illness, and patients with confirmed cases could be isolated — with contact precautions instituted to prevent transmission to other patients.
Meanwhile, the model could serve as an inspiration for other efforts.
Says Young: “I think that this project is a great example of a ‘team science’ approach to addressing complex biomedical questions to improve health care, which I expect to see more of as we enter the era of precision health.”
Additional co-authors of the Infection Control and Hospital Epidemiology paper are Erin E. Ryan, MPH, CCRP, Lauren West, MPH, and David Hooper, M.D., MGH Division of Infectious Diseases; Krishna Rao, M.D., M.S., Laraine Washer, M.D., and Vincent Young, M.D., Ph.D., University of Michigan Medical School; John Guttag, Ph.D., MIT Department of Electrical Engineering and Computer Science; and Christopher Fusco and Robert McCaffrey, Partners HealthCare Information Systems. The study was supported by the MGH-MIT Grand Challenge, National Science Foundation award IIS-1553146, National Institute of Allergy and Infectious Diseases grants U01 AI124255 and K01 AI110524, and a Morton N. Swartz Transformative Scholar Award.
C. diff image courtesy of the Centers for Disease Control and Prevention.