TOPLINE:

A supervised machine learning model developed using routinely available data from electronic health records achieved 80.05% accuracy and showed strong overall performance in predicting which patients with inflammatory bowel disease (IBD) were at a high risk for recurrent Clostridioides difficile infection (CDI).

METHODOLOGY:

• Recurrent CDI is a significant challenge in IBD, yet its prediction remains limited due to the complexity and heterogeneity of both diseases. Researchers developed and internally validated a machine learning model (RecurCDI-IBD) to identify patients at a high risk for recurrent CDI.
• They included 2495 adults with confirmed IBD (median age, 50.4 years; 55.1% women) who developed CDI between 2013 and 2021.
• Recurrent CDI was defined as either a positive repeat test along with the initiation of a new antibiotic course or the initiation of CDI-directed antibiotics alone within 60 days of completing initial therapy. Overall, 654 patients experienced recurrent CDI and 1841 did not.
• A supervised machine learning model was developed using available data such as demographics, IBD subtype, comorbidities, medications, and lab parameters extracted from electronic health records.
• Researchers trained the model on 80% of the patients and tested it on the remaining 20%. They addressed a balanced representation of recurrence cases and provided insights into which clinical factors most strongly influenced each patient’s risk.

TAKEAWAY:

• The model achieved an accuracy of 80.05% and an area under the receiver operating characteristic curve (AUC) of 0.88; sensitivity, 0.76; specificity, 0.84; and precision, 0.83.
• The model showed a balanced performance across both recurrent and nonrecurrent CDI groups, with a precision of 0.83 vs 0.78 and recall of 0.76 vs 0.84.
• Recent hospitalization within 3 months, male sex, initial metronidazole therapy, and comorbidities were associated with an increased risk for recurrent CDI; however, initial therapy with fidaxomicin or vancomycin and prior exposure to 5-aminosalicylic acid were linked to a lower risk.
• Internal cross-validation showed consistent performance of the model, with a mean accuracy of 72.75% and an AUC of 0.82.

IN PRACTICE:

“In this study, we developed a machine learning model that accurately predicts the risk of CDI recurrence in IBD patients based on routinely collected clinical data. Our model offers a new tool to proactively identify high-risk patients and enable earlier, more tailored care,” the authors of the study wrote.

SOURCE:

This study was led by Sahil Khanna, MBBS, MS, of Mayo Clinic in Rochester, Minnesota. It was published online in The American Journal of Gastroenterology.

LIMITATIONS:

The dataset lacked racial and ethnic diversity and was derived from a single academic center. The exclusion of imaging data prevented insights into the severity of IBD, complications, and treatment response. Additionally, the researchers noted that although the model’s performance was consistent, its accuracy metrics should be interpreted cautiously given the study’s exploratory and retrospective design.

DISCLOSURES:

This study received funding from the National Institute of Diabetes and Digestive and Kidney Diseases of the National Institutes of Health. The authors reported having no conflicts of interest.

This article was created using several editorial tools, including AI, as part of the process. Human editors reviewed this content before publication.