What question did this study set out to answer?

This study aims to develop and validate a machine learning model to assist in the early diagnosis of Crohn's disease.

May 5, 2026

Development and validation of a machine learning model and a web tool for predicting the risk of Crohn's disease diagnosis in Chinese patients.

Key Points

This study aims to develop and validate a machine learning model to assist in the early diagnosis of Crohn's disease.
Conducted a retrospective cohort study across two hospitals in China from 2020 to 2024.
Analyzed electronic medical records of patients with a first-time Crohn's disease diagnosis and disease duration ≤24 months, excluding certain demographics.
Trained and validated nine algorithms, primarily focusing on gradient boosting machine for model performance evaluation.
The final gradient boosting machine model achieved an AUROC of 0.998 in the training cohort and 0.954 in the test cohort.
External validation showed an AUROC of 0.944 in a different cohort with a calibrated Brier score of 0.094.
Ten predictive features were selected from 28 candidates based on logistic regression.

Abstract

OBJECTIVES: Diagnostic delay in Crohn's disease (CD) may increase the risk of poor prognosis. This study developed and validated a machine learning (ML) model for clinicians to conduct preliminary diagnosis and refer suspected patients. STUDY DESIGN: Retrospective cohort study. METHODS: This retrospective dual-center study analyzed electronic medical records (EMRs) from two Chinese tertiary hospitals (2020-2024). Patients with a first-time CD diagnosis and disease duration ≤24 months were included, while those aged <18 years, with incomplete medical histories, or prior use of aminosalicylates, glucocorticoids, or biologics were excluded. Nine algorithms were trained, including logistic regression (LR), support vector machine (SVM), gradient boosting machine (GBM), and neural network (NN). Model performance was evaluated using the area under the receiver operating characteristic curve (AUROC), calibration curves, and decision curve analysis. The interpretability of the optimal model was enhanced by SHapley Additive exPlanations (SHAP) values. RESULTS: The final model was developed using data from 526 patients at Jinling Hospital (276 CD and 250 non-IBD), with 10 predictive features selected from 28 candidate variables through univariate and multivariate logistic regression. The GBM model showed superior performance, achieving an AUROC of 0.998 (95% CI: 0.995-1.000) in the training cohort and 0.954 (95% CI: 0.923-0.985) in the test cohort. External validation in 196 patients at Zhongda Hospital (100 CD and 96 non-IBD) yielded an AUROC of 0.944 and a well-calibrated Brier score of 0.094. CONCLUSIONS: Using dual-center data, an interpretable ML model was developed and validated to support early CD diagnosis and referral decisions.

Ask AI

Helpful

Bookmark