0350 Leveraging Sleep-Related Cognitive Susceptibility and Polysomnography-Based Sleep Features to Detect Depression in Adolescents Using Machine Learning

Abstract Introduction Depression is common and often associated with adverse psychosocial outcomes among adolescents. Early and accurate identification of depression is therefore crucial for this vulnerable population. However, current screening and diagnostic methods for depression often rely only on subjective assessments and clinical interviews. This study aimed to determine whether sleep-related cognitive susceptibility and polysomnography-based sleep features could detect depression among adolescents using machine learning (ML). Methods A total of 100 adolescents (Mean age = 17.4 ± 2.0; range 12-20 years; 54% female), including 50 diagnosed with DSM-5 depressive disorders[SL1.1] and 50 controls, were recruited. Participants completed a diagnostic interview and self-report questionnaires assessing insomnia severity, eveningness, and sleep-focused rumination. Participants completed a two-night laboratory-based PSG, with the first night serving as the adaptation night. Sleep data on the second night were extracted to assess sleep architecture and spectral power. Depression diagnosis served as the model output (e.g., Yes/No). The dataset was split into two, with 80% of the data randomly allocated to the training set and 20% to the validation set. A 10-fold cross-validation was used to evaluate five algorithms, namely linear discriminant analysis (LDA), classification and regression trees (CART), k-Nearest Neighbors (kNN), Random Forest (RF), and Logistic Regression (LR), which were specified to predict depression. Results The best-performing algorithm is RF, followed by CART, kNN, LDA, and LR. RF achieved an accuracy of 80.0%, sensitivity of 90%, specificity of 70%, and an AUC of 0.83 in the validation subsample. Conclusion This study demonstrates that leveraging sleep-related cognitive susceptibility, objective sleep parameters, and sleep EEG spectral power in trained ML models can effectively identify depression among adolescents. The findings support the feasibility of developing comprehensive, sleep-focused, data-driven screening methods to facilitate early detection and intervention in adolescent depression. Future research could validate these findings with existing depression screening tools and in larger, more diverse populations. Support (if any) This work was supported by Hong Kong Research Grants Council under the General Research Fund (Ref. 17615722, awarded to SX Li) and the Seed Fund for Basic Research from the University Research Committee, awarded to SX Li.