Key result
Five-variable model distinguishes ulcerative colitis from Crohn's disease with an external AUC of ~0.83.
Why the study?
Does a machine learning model incorporating high-density lipoprotein cholesterol (HDL-C) and four routine clinical/laboratory measures accurately differentiate ulcerative colitis from Crohn's disease in treatment-naive adults presenting for diagnostic evaluation?
Cohort (n=783)
Yes
Does a machine learning model incorporating high-density lipoprotein cholesterol (HDL-C) and four routine clinical/laboratory measures accurately differentiate ulcerative colitis from Crohn's disease in treatment-naive adults presenting for diagnostic evaluation?
Answers are grounded in this paper and cite their sources.
A machine learning model combining HDL-C with four routine clinical and blood measures (age, stool frequency, WBC, and platelets) provides high specificity (95.0%) and good overall discrimination (AUC 0.834) for early triage differentiation of ulcerative colitis from Crohn's disease while definitive endoscopic and histological work-up is pending.
Distinguishing ulcerative colitis (UC) from Crohn’s disease (CD) early in the diagnostic pathway remains difficult when a full assessment is not yet available. This study assessed whether high-density lipoprotein cholesterol (HDL-C), in combination with routine clinical and laboratory measures, could aid in distinguishing ulcerative colitis from Crohn’s disease. This retrospective study included 783 patients from three centers, with 502 in the development cohort and 281 in two external validation cohorts. Candidate variables were narrowed using recursive feature elimination with cross-validation (RFECV). The support vector machine (SVM) classifier was evaluated for discrimination, calibration, and clinical net benefit, with feature contributions assessed using Shapley additive explanations (SHAP). The final panel comprised age, daily stool frequency, white blood cell (WBC) count, platelet count, and HDL-C. In the pooled external cohorts, SVM discrimination was 0.834 by area under the receiver operating characteristic curve; accuracy, sensitivity, and specificity were 0.769, 0.525, and 0.950, respectively. SHAP values generally shifted predictions toward UC with increasing stool frequency, age, WBC, and HDL-C, and toward CD with increasing platelet count. HDL-C contributed information beyond the other four variables. A model built from five readily available measures distinguished UC from CD across the participating centers and may assist with early triage. Its HDL-C component may reflect metabolic and inflammatory differences, although the prediction is an adjunct to, rather than a substitute for, endoscopy, histopathology, and imaging. The online calculator is available at https://uc-vs-cd-predictor-a3fq3ocovi7pagqpwjwh4m.streamlit.app/ .
No takes yet. Share an insight, caveat, or question.
Liu et al. (2026) conducted a cohort in Inflammatory bowel disease (ulcerative colitis and Crohn's disease) (n=783). Five-variable support vector machine (SVM) diagnostic model incorporating HDL-C, age, daily stool frequency, WBC count, and platelet count vs. Four-variable SVM model excluding HDL-C (and reference standard multimodal clinical/endoscopic/histological diagnosis) was evaluated on Diagnostic discrimination between ulcerative colitis and Crohn's disease in pooled external validation cohorts (ROC-AUC) (AUC 0.834, 95% CI 0.785-0.880). A five-variable SVM model incorporating HDL-C, age, stool frequency, WBC, and platelets differentiated ulcerative colitis from Crohn's disease with an external AUC of 0.834.
Synapse has enriched 10 related papers. Consider them for comparative context: