Machine Learning Model Predicts Recurrent Clostridioides difficile Infection in Patients With Inflammatory Bowel...

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Gastroenterology

Background:

• Clostridioides difficile infection (CDI) poses significant challenges for inflammatory bowel disease (IBD) patients, often leading to high recurrence rates and complications.
• Accurate prediction of recurrent CDI (rCDI) is essential for timely interventions and improved patient management in this vulnerable population.

Purpose of the Study:

• To develop and validate a supervised machine learning model, termed RecurCDI-IBD, for predicting rCDI in adult IBD patients.
• To identify key clinical and demographic factors associated with rCDI risk in IBD patients.

Main Methods:

• A Gradient Boosting Machine (XGBoost) model was trained using data from adult IBD patients diagnosed with CDI between 2013 and 2021.
• Feature engineering incorporated demographic, clinical (IBD subtype, medications, comorbidities), and laboratory data to predict rCDI within 60 days of the initial episode.

Main Results:

• The RecurCDI-IBD model demonstrated strong predictive performance with 80.05% accuracy and an Area Under the Curve of 0.88.
• Significant predictors of rCDI included IBD subtype, gender, use of steroids and anti-TNF agents, and comorbidities like chronic pulmonary and renal disease.

Conclusions:

• The RecurCDI-IBD model effectively identifies IBD patients at higher risk for recurrent CDI, showing balanced precision and recall.
• These data-driven insights can enhance clinical risk assessment, although further validation in larger, diverse cohorts is recommended.