Researchers at Boston-based Massachusetts General Hospital, Ann Arbor-based University of Michigan and Cambridge-based Massachusetts Institute of Technology are developing hospital-specific machine learning models that predict patients' risk of Clostridium difficile infections much sooner than current diagnostic methods allow, according to a study published in Infection Control & Epidemiology.
"Despite substantial efforts to prevent C. diff infection and to institute early treatment upon diagnosis, rates of infection continue to increase," co-senior study author Erica Shenoy, MD, PhD, said in a press release. "We need better tools to identify the highest risk patients so that we can target both prevention and treatment interventions to reduce further transmission and improve patient outcomes."
The study authors noted most previous approaches to C. diff infection risk were limited in usefulness since they were not hospital-specific and were developed as "one-size-fits-all" models that only included a few risk factors.
Therefore, to predict a patient's C. diff risk throughout the course of their hospital stay, the researchers took a "big data" approach that analyzed the entire EHR. This method allows for institution-specific models that could be tailored to different patient populations, different EHR systems and factors specific to each facility.
"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," said co-senior study author Jenna Wiens, PhD. "To mitigate these issues, we take a hospital-specific approach, training a model tailored to each institution."
With this machine learning-based model, the researchers looked at de-identified data, which included individual patient demographics and medical history, details on admissions and daily hospitalization, and the likelihood of C. diff exposure. The data was gathered from the EHRs of roughly 257,000 patients admitted to either MGH or to Michigan Medicine over two-year and six-year periods, respectively.
The models proved to be highly successful at predicting patients who would ultimately be diagnosed with C. diff. In half of these infected patients, accurate predictions could have been made at least five days before collecting diagnostic samples, which would allow hospitals to focus on antimicrobial interventions on the highest-risk patients.
The study's risk prediction score could guide early screening for C. diff if validated in subsequent studies. For patients who receive an earlier diagnosis, treatment initiation could curb illness severity, and patients with confirmed C. diff could be isolated to prevent transmission to other patients.
The algorithm code is freely available here for hospital leaders to review and adapt for their institutions. However, Dr. Shenoy notes facilities looking to apply similar algorithms to their own institutions must assemble the appropriate local subject-matter experts and validate the performance of the models in their institutions.