Multimodal phenotypic classification of generalized anxiety and panic using structural MRI data and psychosocial factors: machine learning results from the German National Cohort (NAKO) study

Abstract Anxiety disorders are common and impairing mental health conditions. Using data from 26,378 adults in the German National Cohort Study (NAKO), we investigated psychosocial and neuroimaging predictors of generalized anxiety disorder (GAD) symptoms and panic attacks. We conducted machine-learning analyses of 246 regions of interest from whole-brain imaging data in combination with psychosocial variables. Neuroimaging data alone showed suboptimal classification performance, whereas psychosocial variables alone - particularly depressive symptoms, stress, and childhood trauma - achieved the strongest discrimination for GAD symptoms and panic attacks. Adding neuroimaging features to psychosocial models modestly improved unbalanced accuracy and specificity by reducing false-positive classifications, indicating a conditional and complementary contribution of neuroanatomical information. Within the multivariate models, features from anxiety-related circuits, including the amygdala and superior parietal lobule, were consistently selected. Overall, these findings suggest that psychosocial factors dominate classification of anxiety outcomes, while structural MRI measures may provide complementary information within multimodal frameworks aimed at refining classification and supporting the development of individualized risk profiles to guide tailored therapeutic and preventive strategies.