Development and External Validation of a Clinically Interpretable Prediction Model for Differentiating Crohn’s Disease from Intestinal Tuberculosis

Abstract Background Crohn's disease (CD) and intestinal tuberculosis (ITB) have overlapping clinical, endoscopic, radiologic, and histopathologic features, complicating diagnosis when confirmatory microbiologic or characteristic histopathologic evidence is unavailable. We developed a practical prediction model using routinely available diagnostic information. Methods A multivariable logistic regression model was developed in 132 complete cases (82 with CD and 50 with ITB) and evaluated using a fixed formula and cutoff in a same-center temporal validation cohort (n = 117) and an independent external validation cohort (n = 65). Discrimination, calibration, and classification performance were assessed. Results The final model included age at symptom onset, interferon-gamma release assay, and the numbers of endoscopic features supporting CD and ITB. The AUC was 0.970 (95% CI, 0.947–0.994) in development, 0.9994 (95% CI, 0.9980-1.0000) in temporal validation, and 0.9995 (95% CI, 0.9981-1.0000) in external validation. Accuracy was 97.4% and 95.4% in the two validation cohorts, respectively. The model generally underestimated ITB probability. Conclusions The concise model combining age at symptom onset, IGRA, and structured endoscopic features showed promising discrimination and classification performance in the validation cohorts and may assist CD-ITB differentiation. Prospective multicenter studies should evaluate calibration, clinical utility, and generalizability.