Real-world smartphone data predicts mood after cerebrovascular symptoms and may constitute digital endpoints

Though depression is prevalent in patients with cerebrovascular dysfunction, screening for symptoms is not routine and is often limited via subjective patient-reported surveys. Using smartphone sensors, we sought to evaluate the performance of objective behavior measures and self-report surveys at predicting depression severity in patients with cerebrovascular syndromes. Among enrolled participants (n = 54), 35 patients with ischemic stroke or transient ischemic attack symptoms were monitored in real-world settings using the Beiwe app for 8 or more weeks with adequate compliance. Depression symptoms were tracked via weekly Patient Health Questionnaire-8 (PHQ-8) surveys, monthly personnel-administered Montgomery-Asberg Depression Rating Scale (MADRS) assessments, and passive smartphone sensors. Across weeks, several passive measures were significantly associated with PHQ-8 scores. Personnel-assessed depression severity moderately correlated with self-reported scores. To estimate MADRS, we applied linear mixed models using passive data and PHQ-8 scores. Using antecedent PHQ-8 scores and demographic data, average root-mean-squared error (RMSE) for depression severity prediction across models was 1.54 with accelerometer data, 1.40 also including global position system (GPS) data, and 1.33 also including PHQ-8 open survey duration. Though future research should validate this decentralized approach in a larger cohort, real-world monitoring with active and passive data may triage cerebrovascular patients for efficient depression screening and provide novel mobility and response time outcome measures.