P0309 Machine learning for predicting Crohn’s disease from routine blood tests years before diagnosis: results from the epi-IIRN

Abstract Background Various biomarkers may help predict inflammatory bowel disease (IBD) in advance of diagnosis, but these are mostly not routinely tested and thus are more challenging for screening. In this nationwide study, we used the epi-IIRN validated cohort to explore the utility of routine blood tests as markers for predicting IBD occurrence. Methods We used the epi-IIRN cohort to collect results of all blood tests performed up to 15 years before diagnosis, from all IBD patients diagnosed between 2005-2020 and insured by Israeli health maintenance organizations (HMOs); each patient was individually matched to two non-IBD controls. Means were compared using Welch’s t-test with false discovery rate correction to account for multiple comparisons. We then used machine learning to build a model using the 15 most statistically significant blood tests to predict Crohn’s disease. Results 46,223 individuals were included – 8,630 with Crohn’s disease (CD) (43% males, mean age at diagnosis 38±19 years) and 6,791 with ulcerative colitis (UC) (41% males, mean age at diagnosis 42±10 years) with their 2:1 matched controls. For adults with CD, 29 tests showed differences between cases and controls earlier than 1 year pre-diagnosis, including 3 with consistent differences >10 years pre-diagnosis (white blood cell, neutrophil and platelet counts). In children, 17 blood tests showed consistent differences earlier than 1 year pre-diagnosis. No test showed significant differences between UC cases and controls. The machine-learning model had an area under the curve (AUC) for predicting future CD 1 year pre-diagnosis of 0.70 in adults and 0.68 in children. Conclusion We were able to detect significant changes in routinely collected blood tests long before CD diagnosis and to predict future CD using a machine learning model. This opens the possibility of detecting early signals of future CD in patients undergoing routine blood tests, which may be used for developing screening and prediction models for prevention strategies.