The needle study: Machine learning as a new method for case-finding in celiac disease.

Area of Science:

• Pediatric Gastroenterology
• Medical Informatics
• Computational Biology

Background:

• Celiac disease (CeD) diagnosis can be challenging in children with nonspecific symptoms.
• Machine learning (ML) offers potential for improving diagnostic accuracy using clinical data.

Purpose of the Study:

• To develop and evaluate ML prediction models for identifying children requiring CeD screening.
• To identify uncommon clinical features indicative of CeD in pediatric patients.

Main Methods:

• Utilized a discovery cohort of children with CeD and matched controls.
• Collected demographic, symptom, laboratory, and family history data, excluding specific antibody levels.
• Applied various supervised ML models, including Ridge Classifier and LASSO, with 10-fold cross-validation.

Main Results:

• A Ridge Classifier model achieved an area under the ROC curve of 0.763.
• The LASSO model identified 40 predictive features, including muscle pain, reflux-like symptoms, and fatigue.
• Performance metrics included F1-score of 0.662, sensitivity of 0.652, and specificity of 0.689.

Conclusions:

• ML models accurately identified CeD in children with nonspecific clinical features.
• The identified 40 features can enhance case-finding strategies for improved CeD detection.