Identifying candidates for smartphone monitoring: a cross-sectional and longitudinal analysis of depression phenotypes associated with cerebrovascular disease in UK Biobank (Preprint)

BACKGROUND Interest in using smartphones to monitor depression, a risk factor for cerebrovascular disease (CeVD), has grown rapidly; however, little is known about behavioral phenotypes related to mood and how phenotypes may vary based on symptom severity in patients before CeVD. OBJECTIVE We aimed to examine the relationships between depression frequency and lifestyle factors, able to be captured by smartphone sensors, to assess the feasibility of using smartphones for monitoring mood in incident CeVD cohorts outside of clinical settings. METHODS We retrospectively identified patients who suffered a CeVD after baseline (n = 14,508) and conducted cross-sectional analyses with patients in the UK Biobank (UKBB) observational cohort study at baseline. Longitudinal analyses were performed in those (n = 603) who completed a follow-up. RESULTS In the cross-sectional analysis in diagnosed depression (DDs) and control cohorts, optimal sleep (OR = 0.58-0.71; P < .001) was associated with decreased frequency depressed mood while former/current smoker status (OR = 1.146-1.151; P < .05) and daily screen time (OR = 1.039-1.058; P < .007) were associated with increased frequency. In both cohorts, older age (> 60 y) was protective (OR = 0.52-0.67; P <.001) while social deprivation (OR = 1.041-1.048; P < .002) was linked with higher frequency. Specific to controls, male sex (OR = 0.713; P <.001) and increased daily physical activity duration (OR = 0.989; P = .001) were protective. Longitudinal analysis revealed that older age’s protective effect persisted in controls (OR = 0.552; P = .02). At follow-up, baseline depressed mood frequency (OR = 5.897; P <.001) was associated with increased depressed mood frequency in controls while prolonged daily screen time (OR = 1.379; P <.015) was associated with higher frequency in DDs. Sensitivity analyses stratified by time-to-diagnosis suggest that associations between lifestyle factors and depressed mood in incident CeVD may be transient and time-dependent. CONCLUSIONS Multiple behaviors observable via smartphone sensors are associated with depression in patients before a CeVD diagnosis. Clinicians monitoring this mood phenotype should pay close attention to screen time and sleep duration in at risk patients. For patients at risk of CeVD with no history of depression, screening for depression may provide insights into possible mood changes emergent before CeVD. Given that lifestyle behaviors linked to depression may evolve in the years before a CeVD diagnosis, a robust approach, incorporating both passive and active smartphone sensors, is needed when monitoring patients at risk of CeVD.