ROUTINELY COLLECTED BLOODWORK REVEALS IMMUNE CELL PATTERNS THAT REFLECT DISEASE ACTIVITY IN PATIENTS WITH INFLAMMATORY BOWEL DISEASE

Abstract

Abstract OBJECTIVE The standard procedure for evaluation of inflammatory bowel disease (IBD) diagnosis and activity requires endoscopy, but there have been many biomarkers identified that are associated with IBD and are collected much less invasively. White blood cell subsets, also known as “the differential”, is a commonly used test for patients with IBD, but the relationship between the differential and disease activity is unknown. Understanding the connection between the differential and disease activity could provide insight to a patient’s clinical presentation without requiring more invasive testing. The aim of this study was to identify patterns present in the differential of IBD patients and use machine learning to determine how these parameters change with IBD activity. METHODS A total of 373 differential counts from 43 patients with IBD were evaluated. All IBD patients with adequate clinical information were included. Disease activity was determined by chart review using the Physician’s Global Activity Score. Binary classification prediction was performed on the training set using multiple supervised machine learning models. RESULTS The outcome was balanced, with 56% of the observations experiencing IBD activity at the time of blood collection. A random forest (RF) model performed best with a final fit accuracy of 0.70 and an ROC-AUC score of 0.75. The RF classifier ranked monocytes and neutrophils as having the highest relative importance to the model. CONCLUSION The optimal RF model demonstrated an overall correctness in its predictions and as reflected in the ROC-AUC score, can discriminate between active disease state and remission in IBD patients. These results indicate that a pattern is present between the differential and IBD activity and with further exploration, may be used to create a predictive model for IBD diagnosis.