Artificial Intelligence-Driven Multimodal Prediction of Disease Severity and Biologic Therapy Response in Inflammatory Bowel Disease Using Integrated Clinical Endoscopic and Biomarker Data

Abstract

Inflammatory bowel disease (IBD), encompassing Crohn’s disease and ulcerative colitis, is characterized by substantial heterogeneity in clinical manifestations, disease progression, treatment responsiveness, and long-term outcomes. Despite considerable advances in biologic therapies, therapeutic decision-making remains challenging because patients frequently exhibit variable responses to treatment and unpredictable disease trajectories. Conventional clinical assessment approaches rely heavily on isolated biomarkers, endoscopic findings, or physician experience, often limiting the ability to accurately identify patients at risk of severe disease progression or treatment failure. Recent developments in artificial intelligence (AI) and machine learning have created opportunities to integrate diverse clinical, endoscopic, and molecular datasets into predictive frameworks capable of supporting personalized disease management. This study presents a multimodal artificial intelligence framework for predicting disease severity and biologic therapy response in patients with inflammatory bowel disease through the integration of clinical characteristics, endoscopic activity indices, laboratory biomarkers, inflammatory markers, and multi-omics-derived indicators. The proposed framework combines supervised machine learning algorithms, feature selection techniques, and explainable artificial intelligence methods to identify the most informative predictors associated with disease activity and therapeutic outcomes. Variables incorporated into the predictive model include demographic characteristics, disease duration, C-reactive protein levels, fecal calprotectin concentrations, endoscopic severity scores, microbiome-associated signatures, and prior treatment histories. Model performance was evaluated using multiple classification metrics, including accuracy, sensitivity, specificity, area under the receiver operating characteristic curve, and predictive value assessments. Comparative analyses demonstrated that multimodal models consistently outperformed single-domain approaches by capturing complex interactions among clinical, endoscopic, and biomarker variables. Feature importance analysis further revealed that inflammatory biomarkers, endoscopic severity measurements, and microbiome-derived indicators contributed substantially to prediction accuracy. The findings support the growing role of artificial intelligence in precision gastroenterology and suggest that integrated predictive systems may facilitate earlier identification of high-risk patients, optimization of biologic treatment strategies, reduction of unnecessary therapeutic escalation, and improvement of long-term clinical outcomes. The proposed framework provides a clinically interpretable and scalable approach for advancing individualized care in inflammatory bowel disease and offers a foundation for future implementation of data-driven decision support systems in routine clinical practice.