In F1000Research
Background: Patients with Crohn's disease (CD) have an altered intestinal microbiome, which may facilitate novel diagnostic testing. However, accuracy of microbiome classification models across geographic regions may be limited. Therefore, we sought to examine geographic variation in the microbiome of patients with CD from North America and test the performance of a machine learning classification model across geographic regions. Methods: The RISK cohort included 447 pediatric patients with CD and 221 non-inflammatory bowel disease controls from across North America. Terminal ileum, rectal and fecal samples were obtained prior to treatment for microbiome analysis. We divided study sites into 3 geographic regions to examine regional microbiome differences. We trained and tested the performance of a machine learning classification model across these regions. Results: No differences were seen in the mucosal microbiome of patients with CD across regions or in either the fecal or mucosal microbiomes of controls. Machine learning classification algorithms for patients with CD performed well across regions (area under the receiver operating characteristic curve [AUROC] range of 0.85-0.91) with the best results from terminal ileum. Conclusions: This study demonstrated the feasibility of microbiome based diagnostic testing in pediatric patients with CD within North America, independently from regional influences.
Shah Rajesh, Hoffman Kristi, Denson Lee, Kugathasan Subramaniam, Kellermayer Richard
2022
Crohn’s disease, inflammatory bowel disease, machine learning, microbiome
