BACKGROUND
Unobtrusively collected objective sensor data from everyday devices like smartphones provide a novel paradigm to infer mental health symptoms. This process called smart sensing allows a fine-grained assessment of various features (e.g., time spent at home based on the GPS sensor). Based on its prevalence and impact depression is a promising target for smart sensing. However, currently it is unclear which sensor- based features should be used in depression prediction and if they hold an incremental benefit over established fine-grained assessments like Ecological Momentary Assessment (EMA).
OBJECTIVE
Hence, the present study investigated various features based on the smartphone screen, app usage, and call sensor alongside EMA to infer to depression severity. Bivariate, cluster-wise, and cluster-combined analysis were conducted to determine the incremental benefit of smart sensing features among each other and over EMA in parsimonious regression models for depression severity.
METHODS
In this exploratory observation study participants were recruited from the general population. Participants needed to be 18 years of age, provide written informed consent, and own an Android-based smartphone. Sensor data and EMA were collected via the INSIGHTS app. Depression severity was assessed by the PHQ-8 questionnaire. Missing data was handled by multiple imputations. Correlation analyses were conducted for bivariate associations, and stepwise linear regression analyses to find the best prediction models for depression severity. Models were compared by adjusted R2. All analyses were pooled across the imputed data sets according to Rubin’s rule.
RESULTS
A total of N=107 participants were included in the study. Age ranged from 18 to 56 years (M=22.81, SD=7.32) and 78% of the participants identified themselves as female. Depression severity was subclinical on average (M=5.82, SD=4.44, PHQ-8 ≥10: 18.7%). Small to medium correlations were found for depression severity and EMA (e.g., valence: r = -.55, 05%-CI: -.67 to -.41) and small correlations with sensing features (e.g., screen duration: r = .37, 95%-CI: .20 to .53). EMA features could explain 35.38% (95%-CI: 20.73% to 49.64%) of variance and sensing features adj. R2 = 20.45% (95%-CI: 7.81% to 35.59%). The best regression model contained EMA and sensing features (R2 = 45.15%, 95%-CI: 30.39% to 58.53%).
CONCLUSIONS
Our findings underline the potential of smart sensing and EMA to infer to depression both as isolated paradigms and especially when combined. While these could become important parts in clinical decision support systems for depression diagnostics and treatment in future, confirmatory studies are highly needed before an application in routine care. Furthermore, privacy, ethical, and acceptance issues need to be addressed.