Psychological Predictors of Symptom Severity in Disorders of Gut–Brain Interaction: A Machine Learning Approach

Background/Aims: : Disorders of gut-brain interaction (DGBI) exhibit heterogeneous symptom burdens and psychosocial variability. We aimed to identify severity-based phenotypes via unsupervised clustering and evaluate the psychological factors independently associated with symptom severity using machine learning. Methods: : We analyzed 480 patients with DGBI who completed validated gastrointestinal symptom scales and a psychological battery assessing anxiety, depression, childhood trauma, social support, resilience, and Type D personality. K-means clustering (k = 2) yielded low-severity groups (n = 281) and high-severity groups (n = 199). Between-group differences were examined using Welch's t-test with false discovery rate (FDR) correction and Cohen's d. Five logistic regression models with different feature sets were compared under stratified 5-fold cross-validation. Feature importance was assessed via bootstrap odds ratios with 95% confidence intervals. Results: : The high-severity group showed significantly elevated anxiety and depressive symptoms (Cohen's d = 0.55-0.92 for the Beck Anxiety Inventory [BAI]; d = 0.56-0.63 for the Beck Depression Inventory) and higher childhood trauma scores on physical abuse, sexual abuse, and physical neglect subscales after FDR correction. The BAI-only model (area under the receiver operating characteristic curve [AUC] = 0.750) nearly matched the full 17-feature model (AUC = 0.753; accuracy = 71.3%; balanced accuracy = 69.9%), suggesting that the incremental contribution of multidimensional psychological profiling beyond anxiety assessment alone was limited. Conclusions: : Anxiety symptoms, particularly somatic and physiological dimensions, are the primary psychological determinants of DGBI severity. Childhood trauma associations are likely to operate through anxiety and depression pathways. These findings support routine somatic anxiety screening in gastroenterology to identify candidates for brain-gut behavioral therapies.