Machine Learning Improves Accuracy of Inflammatory Bowel Disease Identification Beyond ICD-10 Codes (Houston Methodist)

Houston Methodist researchers developed an explainable machine-learning model that distinguishes Crohn’s disease, ulcerative colitis and patients without IBD using electronic medical record data, addressing the limitations of relying on ICD-10 codes alone. In a manually validated cohort, the model increased the positive predictive value from 65% to 90% by combining 198 clinical variables—including medications, imaging, pathology, specialist involvement and procedures. A random-forest model performed best, while SHAP-based explainability showed which clinical features drove each prediction. The approach could help researchers build more accurate IBD cohorts and strengthen the reliability of large-scale EMR-based studies.