Using Machine Learning and the Electronic Health Record to Predict Complicated Clostridium difficile Infection

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Infectious Disease Epidemiology

Background:

• Clostridioides difficile infection (CDI) is a significant healthcare-associated infection with severe potential complications.
• Accurate identification of patients at high risk for complicated CDI remains a clinical challenge.

Purpose of the Study:

• To explore the utility of machine learning (ML) for patient risk stratification of CDI complications.
• To leverage electronic health record (EHR) data for predicting complicated CDI.

Main Methods:

• Adult patients diagnosed with CDI between October 2010 and January 2013 were analyzed.
• Complicated CDI was defined by ICU admission, colectomy, or 30-day mortality.
• An ML model was trained on EHR data to predict complications, with performance evaluated using AUROC.

Main Results:

• The ML model achieved an AUROC of 0.69 on the day of CDI diagnosis.
• Model performance significantly improved to an AUROC of 0.90 two days after diagnosis.
• The EHR-based ML model outperformed a model using manually curated features (AUROC 0.84).

Conclusions:

• EHR data can be effectively utilized to accurately stratify CDI patients by complication risk.
• This ML approach holds potential for guiding clinical studies on interventions to prevent or mitigate complicated CDI.