Passive Screening for Depressive Symptoms Using Daily Wrist Actigraphy and Deep Learning: Model Development and Validation Study.

Background: Depressive symptoms are common yet often underrecognized in routine care, underscoring the need for scalable screening approaches beyond episodic self-report assessments. Wearable actigraphy can passively and continuously capture daily activity and 24-hour rest-activity rhythms associated with depressive symptom burden. However, the performance of artificial intelligence (AI) models that leverage actigraphy data for depressive symptom screening remains insufficiently established. Objective: This study aimed to develop and evaluate AI-based models for passive screening of depressive symptoms from daily wrist actigraphy data. Methods: We analyzed actigraphy recordings from 1160 Hispanic/Latino adults in the Hispanic Community Health Study/Study of Latinos who completed the 10-item Center for Epidemiologic Studies Depression scale (CESD-10), a self-reported depressive symptom screening scale. Multichannel actigraphy data, including activity counts, light exposure, and wake status, were used as inputs to 5 deep learning architectures to classify CESD-10-defined depressive symptom groups, comparing mild and higher symptoms with the normal group. Results: Actigraphy-derived behavioral markers differed across depressive symptom groups, showing lower daytime activity and altered circadian rest-activity organization with increasing symptom burden. Among the 5 deep learning architectures evaluated, the long short-term memory model consistently demonstrated the strongest overall discrimination. In held-out testing, the long short-term memory model achieved a macro-averaged area under the receiver operating characteristic curve of 0.80, with the strongest discrimination observed for the higher depressive symptom group (area under the receiver operating characteristic curve 0.889). These findings indicate improved model discrimination with increasing symptom severity, although false-positive rates remained notable across both classification tasks. Conclusions: Our study suggests that actigraphy-derived data can support AI-based classification of depressive symptoms. An actigraphy-based AI model may serve as a scalable, passive, and noninvasive complementary signal to aid early screening alongside traditional depressive symptom assessments before clinical diagnosis.