Development and Validation of Predictive Models for Inflammatory Bowel Disease Diagnosis: A Machine Learning and...

Area of Science:

• Gastroenterology and Hepatology
• Medical Informatics
• Biostatistics

Background:

• Inflammatory bowel disease (IBD) encompasses chronic, incurable gastrointestinal conditions like Crohn's disease (CD) and Ulcerative colitis (UC).
• Current diagnostic methods for IBD lack a definitive gold standard, necessitating improved predictive tools.

Purpose of the Study:

• To develop and validate predictive models for diagnosing IBD, CD, and UC.
• To compare the efficacy of machine learning (ML) and traditional nomogram models in IBD diagnosis.

Main Methods:

• Utilized three cohorts with initial laboratory test data from UK Biobank, the First Hospital of Jilin University, and a Chinese tertiary hospital.
• Developed ML models using LightGBM and XGBoost algorithms, and nomogram models via Logistic regression.
• Validated model performance using Area Under the Curve (AUC) metrics in independent cohorts.

Main Results:

• Machine learning models demonstrated exceptional discrimination across cohorts, with AUCs ranging from 0.772 to 0.932 for CD and UC predictions.
• Nomogram models exhibited good diagnostic capability, with validated AUCs for IBD, CD, and UC ranging from 0.758 to 0.817 in the external testing cohort.
• Both ML and nomogram approaches showed promising predictive performance for IBD subtypes.

Conclusions:

• Developed risk prediction models for IBD, CD, and UC using conventional laboratory data and integrated ML and nomogram techniques.
• The models demonstrated good diagnostic capability and were successfully validated in an independent cohort, suggesting clinical utility.
• This study offers a novel, data-driven approach to enhance the diagnostic accuracy of inflammatory bowel diseases.