Predicting generalized anxiety disorder among Chinese depressed adolescents: an explanable machine learning approach (Preprint)

Abstract

BACKGROUND Anxiety disorders have high lifetime comorbidity (45.7%) with depression and the co-occurrence of depression and anxiety constitutes greater mental health burden and impact on quality of life than depression symptoms alone. This necessitates the prediction of anxiety disorders in depressed adolescents, and the identification of key risk factors as well as their interactions for early prevention. OBJECTIVE The current research leverages an explainable machine learning approach to develop a predictive model that evaluates the predictive pertinence of diverse factors. Meanwhile, the interactions between different risk factors and validity of factor significance measures were investigated. METHODS We recruited 2316 depressed adolescents through the Chinese Adolescent Depression Cohort (CADC) and collected 34 predictive factors for model construction. The Light Gradient Boosting Machine (LightGBM) prediction model and Shapley Additive Explanations (SHAP) algorithm were implemented for in-depth interpretation of the predictive importance of different factors. Furthermore, Chi-square Automatic Interaction Detection (CHAID) and ordinal logistic regression were used to explore the factor interactions and validate the SHAP value-based factor importance, respectively. RESULTS Nine key risk factors were identified. Besides the depressive symptoms, rumination, perceived stress, sleep quality, alexithymia, peer victimization, academic stress level, emotion-focused coping, and parental overprotect were recognized as key risk factors for onset of anxiety. In addition, resilience was recognized as the protective factor. Interaction analysis results indicate that adolescents experiencing severe depression symptoms are more vulnerable to alexithymia and severe anxiety. For adolescents with low depressive symptoms but high alexithymia, they are more likely to develop into severe anxiety and depression when their alexithymia exceeds a threshold of 3.5. Three high-risk subgroups for severe anxiety were detected by the CHAID decision tree: (1) adolescents with severe depression symptoms, (2) with moderate depression symptoms and high rumination, (3) with severe depression symptoms and high alexithymia. Three low-risk subgroups are: (1) adolescents with low depression and rumination, (2) with low depression, low alexithymia, and more parental care, (3) with low depression, moderate rumination, and moderate academic stress. CONCLUSIONS Utilizing an explainable machine learning approach enables us to identify the risk and protective factors of anxiety disorders among depressed adolescents. While depressive symptoms play a critical role in the prediction of anxiety, the comorbidity of depression and anxiety is driven by additional factors such as alexithymia and rumination. These findings suggest the clinical workers take into consideration of the above risk and protective factors as well as their interactions to develop appropriate therapy for the prevention of comorbid anxiety with depression.