Accurate Classification of Pediatric Colonic Inflammatory Bowel Disease Subtype Using a Random Forest Machine...

Area of Science:

• Gastroenterology
• Pediatric Medicine
• Computational Biology

Background:

• Existing pediatric inflammatory bowel disease (PIBD) classification algorithms lack consistency and focus on features atypical for ulcerative colitis (UC).
• Distinguishing between pediatric UC and colonic Crohn disease (CD) is crucial for appropriate treatment and management.

Purpose of the Study:

• To develop a novel algorithm for accurate differentiation between pediatric UC and colonic CD.
• To identify specific features that effectively discriminate between these two conditions in pediatric patients.

Main Methods:

• Utilized a dataset of 74 pediatric colonic IBD patients (56 UC, 18 CD) including clinical, endoscopic, radiologic, and histologic data.
• Employed similarity network fusion (SNF) for initial clustering and trained a Random Forest (RF) classifier.
• Validated the model using leave-one-out cross-validation and tested on a separate cohort of 15 patients.

Main Results:

• The RF classifier achieved 97% accuracy in distinguishing UC from CD in the initial cohort.
• Identified 7 key features (3 histologic, 4 endoscopic) critical for differentiation.
• A refined RF classifier using these 7 features demonstrated 100% accuracy in an independent validation set.

Conclusions:

• A combination of supervised and unsupervised analyses successfully identified a concise set of features to reliably distinguish pediatric UC from colonic CD.
• The developed algorithm and identified features offer a promising tool for consistent diagnosis in pediatric IBD.
• Further validation in diverse patient populations is recommended for broader clinical application.