Modest Clostridiodes difficile infection prediction using machine learning models in a tertiary care hospital

Area of Science:

• Medical Informatics
• Infectious Diseases
• Machine Learning

Background:

• Machine learning (ML) models show potential for predicting patient health outcomes.
• Clostridioides difficile infection (CDI) is a significant healthcare-associated infection.
• Accurate prediction of CDI can aid in timely intervention and patient management.

Purpose of the Study:

• To develop and evaluate the performance of various ML models in predicting CDI among hospitalized patients.
• To assess the feasibility of using ML for CDI risk stratification in a real-world clinical setting.

Main Methods:

• Retrospective cohort study of 3514 inpatients tested for C. difficile between 2015-2017.
• CDI defined by positive glutamate dehydrogenase and toxin assays.
• Ten ML models were tested, including logistic regression, random forest, and naïve Bayes, with and without resampling.

Main Results:

• The overall incidence of CDI in the cohort was 4% (136/3514).
• Age and recent antibiotic use (within 90 days) were significantly associated with CDI (P < 0.01).
• The best performing ML models (logistic regression, random forest, naïve Bayes) achieved a modest Area Under the Receiver Operating Characteristic Curve (AUC ROC) of 0.6.

Conclusions:

• Predicting CDI in hospitalized patients tested for the infection proved challenging.
• ML models demonstrated only moderate predictive performance in this real-world cohort.
• Further research is needed to improve ML model accuracy for CDI prediction in clinical practice.