Modeling the Transdiagnostic Impact of Childhood Trauma on Adult Psychiatric Symptoms Using Interpretable Machine Learning

Purpose: This study aimed to examine the transdiagnostic impact of childhood trauma subtypes on adult psychiatric symptom severity and to evaluate their predictive value using interpretable machine learning models beyond conventional statistical approaches. Methods: In this cross-sectional study, 1,192 adults from clinical and community settings completed the Childhood Trauma Questionnaire–Short Form (CTQ-28) and the Symptom Checklist-90-Revised (SCL-90-R). Hierarchical multiple regression analyses assessed the incremental contribution of trauma subtypes after controlling for sociodemographic and clinical variables. In parallel, supervised machine learning models (Random Forest and XGBoost) were developed and evaluated using cross-validation and independent test sets. Explainable artificial intelligence techniques were applied to enhance interpretability. Results: Childhood trauma significantly predicted psychiatric symptom severity across all SCL-90-R domains. Emotional abuse and emotional neglect consistently emerged as the strongest transdiagnostic predictors. Among machine learning approaches, Random Forest demonstrated the best predictive performance based on test-set RMSE and R² values. Explainability analyses identified emotional abuse, emotional neglect, and age as the most influential variables across symptom domains. Conclusion: Emotional forms of childhood trauma represent key transdiagnostic determinants of adult psychiatric symptoms. Interpretable machine learning approaches provide a robust and clinically meaningful framework for modeling complex trauma–symptom relationships and may facilitate trauma-informed risk stratification and early identification strategies.