Machine learning classification of inflammatory bowel disease activity using white blood cell subsets

Area of Science:

• Immunology
• Computational Biology
• Gastroenterology

Background:

• Current inflammatory bowel disease (IBD) monitoring lacks rapid, validated tests.
• Existing biomarkers like fecal calprotectin are slow, and bloodwork lacks specificity.
• The relationship between white blood cell (WBC) subsets and IBD activity is understudied.

Purpose of the Study:

• To classify IBD disease activity using routine bloodwork.
• To apply machine learning (ML) to identify complex patterns in WBC subsets.
• To develop a novel approach for tracking IBD activity.

Main Methods:

• Analysis of 1458 bloodwork measurements from 108 IBD patients.
• Disease activity classified by physician's global assessment.
• Four ML models trained on complete blood count, differential, albumin, ESR, and CRP.

Main Results:

• The optimal ML model achieved an AUC of 0.882.
• Neutrophils, C-reactive protein, and albumin were key predictive features.
• The ML model outperformed individual biomarkers and was minimally impacted by medications.

Conclusions:

• ML analysis of routine bloodwork can effectively classify IBD activity.
• This approach can aid in treatment planning and improve patient outcomes.
• This method offers a more efficient alternative to current IBD monitoring tools.