Machine learning to phenotype pain and predict response to pain interventions among young adults with irritable bowel syndrome.

Introduction: Irritable bowel syndrome (IBS) is a prevalent disorder whose most debilitating symptom is pain. The complex, multifactorial nature of IBS pain leads to highly variable and often inadequate responses to self-management, underscoring the urgent need for personalized prediction models. Methods: This ancillary analysis of a randomized controlled trial (NCT03332537) utilized data from 80 young adults with IBS. We applied the Bayesian Additive Regression Trees machine learning algorithm to develop 27 distinct predictive models for pain severity, pain interference, and quality of life (QOL) at baseline and post-intervention. Predictors included a comprehensive, multi-domain set of variables spanning genetics, quantitative sensory testing, gut microbiota, psychosocial factors, and food intake. Results: Model performance was strong, with area under the curve (AUC) values ranging from 0.753 to 0.981. A consistent hierarchy of predictors emerged. The Discussion: This study establishes a comprehensive, multi-omics framework that explains individual differences in IBS pain and treatment response. The identified predictors provide a practical tool for advancing precision medicine. By classifying patients based on their distinct profiles, clinicians can proactively customize self-management strategies, potentially transforming care for this complex condition.