Machine Learning Prediction of Clostridioides difficile Infection in Hospitalized COVID-19 Patients Across Pandemic...

Area of Science:

• Infectious Disease Epidemiology
• Computational Medicine
• Health Informatics

Background:

• Clostridioides difficile infection (CDI) is a significant healthcare-associated complication, particularly in COVID-19 patients with prolonged hospitalization and intensive treatments.
• Antibiotic exposure and other clinical factors increase CDI risk in hospitalized populations.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting CDI occurrence in hospitalized COVID-19 patients.
• To interpret these models to identify key predictors of CDI risk.
• To create a decision-support tool for clinical risk stratification.

Main Methods:

• Analysis of anonymized clinical data from 3848 COVID-19 patients using CRISP-DM methodology.
• Comparison of logistic regression, Random Forest, XGBoost, and multilayer perceptron models.
• Handling of missing data and class imbalance, with performance assessed by PR-AUC and AUROC.
• Utilizing SHAP, LIME, and odds ratio analysis for model interpretability.

Main Results:

• The best machine learning models demonstrated a significant improvement in predicting CDI risk (PR-AUC up to 0.160).
• XGBoost achieved the highest AUROC (0.823), with Random Forest also performing well (0.798).
• Inflammatory markers were identified as key predictors for CDI risk; a web-based decision-support tool was developed.

Conclusions:

• Interpretable machine learning models offer valuable CDI risk stratification for imbalanced COVID-19 datasets.
• The developed decision-support tool shows promise for clinical workflow integration.
• External and prospective validation are necessary for widespread clinical adoption.