Integrating wearable biosensing with clinical interviews for suicide risk detection in adolescents.

Background: Adolescent suicide is a critical public health challenge. Traditional risk screening relies on self-report measures limited by various biases, while passive EDA monitoring during daily activities, often yields limited predictive validity due to environmental confounding such as motion artifacts, thermoregulatory sweating, and ambient temperature fluctuations. This study proposes an innovative "interview-embedded" framework to capture physiological signatures of suicide ideation (SI) during a standardized clinical probe. Methods: A total of 151 adolescents (102 with active suicide ideation, 49 matched controls) were enrolled. Their electrodermal activity signals were continuously recorded throughout clinical interview (Mini International Neuropsychiatric Interview for Children and Adolescents) and analyzed through advanced machine learning approaches. Results: XGBoost model achieved superior classification performance (AUC=0.802, sensitivity=0.857, specificity=0.8) compared to CatBoost, and Balanced Random Forest, significantly outperforming resting-state models and clinical symptom baselines (Resting EDA Model: AUC = 0.598, sensitivity = 0.571, specificity = 0.4; Full Clinical Interview Baseline: AUC = 0.926, sensitivity = 0.476, specificity = 1; Symptom-only Clinical Baseline Model: AUC = 0.712, sensitivity = 0.619, specificity = 0.8). Feature importance analysis revealed that dynamic features reflecting physiological reactivity were the most discriminative markers, providing predictive information potentially beyond self-report. EDA features did not significantly correlate with continuous BSS severity scores (all ρ < 0.15, all p > 0.05; regression R Conclusions: This study demonstrates that a task-embedded biosensing framework, integrating wearable biosensing into standardized clinical interviews is a feasible and effective approach for adolescent suicide risk detection. Embedding physiological data collection within established clinical workflows offers a scalable solution to improve early suicide risk screening in real-world settings, such as schools and primary care.