Abstract
PURPOSE: Cumulative bowel damage (BD) critically influences the progression and prognosis of Crohn's disease (CD). Although the Lémann Index (LI) remains the standard for BD assessment, its clinical utility is limited by heavy reliance on extensive clinical data. Multiparametric magnetic resonance enterography (MRE) provides noninvasive macroscopic evaluation of BD severity, however, it fails to characterize microscopic alterations. We therefore integrated MRE with gut microbiome and metabolomic data to uncover mechanistic insights and develop a comprehensive model for better prediction of BD. METHODS AND RESULTS: In this prospective two-center study, 309 CD patients were stratified into BD and non-BD groups using LI. Patients underwent MRE, fecal 16S rRNA gene sequencing, and fecal/serum metabolomic analysis. Thirty healthy controls were included for comparison. The relationships between microbial/metabolic factors and MRE features were explored using correlation and mediation analyses. Seven machine learning algorithms, each paired with seven distinct combinations of multi-omics features, were evaluated using nested 5-fold cross-validation to construct an optimal prediction model. BD patients exhibited reduced gut microbial diversity (P<0.05), with Erysipelatoclostridium and [Ruminococcus]_gnavus_group as key discriminators. Metabolomics revealed elevated fecal aromatic amino acids and depleted serum glycerophospholipids/sphingolipids (P<0.05) linked to MRE-quantified features through mediation by microbial pathways (eg, 22.8% mediation effect of Prevotella_9 on penetration, P (ACME)=0.022). The optimal Xtreme Gradient Boosting Classifier (XGBC) model integrating three microbial genera, six fecal metabolites, three serum metabolites, and three MRE features achieved superior performance (AUC 0.857 and 0.829 in the derivation and external validation cohorts, respectively). SHapley Additive exPlanations (SHAP) analysis prioritized perianal diseases, Erysipelatoclostridium, and fecal alanine as key contributors. CONCLUSION: Our study underscores the interplay between gut microbial dysbiosis, metabolic alterations, and MRE-quantified structural changes in BD patients. The integrated multi-omics model provides a promising tool for BD prediction, enabling precise CD severity stratification and personalized clinical decision-making.