| dc.description.abstract | Background: Clinicians and researchers often use subjective measures of anxiety at isolated timepoints (e.g., retrospective self-report questionnaires), which are useful for capturing meaningful clinical differences between people and across time. However, such subjective measures are limited by recall biases and an inability to capture real-time fluctuations in anxiety-related behaviours. Smartphone mobile sensors (e.g., GPS; call logs; screen usage logs) have emerged as a method for measuring mental health-related behaviours in real-time. In adults, smartphone mobile sensors have been found to be useful for predicting symptom course and outcomes for internalizing disorders (e.g., depression; anxiety). However, research in youth with anxiety remains scarce, particularly amongst samples with clinically significant anxiety collected after the COVID-19 pandemic.
Objective: The present study sought to use the Predicting Risks and Outcomes of Social InTeractions (PROSIT) app to track anxiety-related behaviours in a sample of youth and adults with clinically significant anxiety, collected after the COVID-19 pandemic.
Methods: 65 youth (aged 15-21 years) and 96 adults (aged 26-40 years) with clinically significant anxiety (as assessed at baseline) were recruited via online advertising for a two-week study. Participants first were asked to complete baseline assessments of anxiety, depression, substance use, and demographics. Then, they were asked to use the PROSIT app for 14 days, including passive sensing (i.e., without them having to do anything) of their mobility, sociability, physical activity, screen usage, and sleep behaviours, as well as answering daily ratings of anxiety and weekly questions about sociability, physical activity, and sleep behaviours. Features were engineered from the PROSIT app and summarized into two weekly timepoints. Mobile sensing features were used in Spearman correlations, structural equation models, and random forest machine learning models to predict four outcomes: momentary anxiety (i.e., weekly summaries of the daily anxiety ratings), momentary anxiety variance (variance in weekly summaries of daily anxiety ratings), baseline anxiety, and baseline depression.
Results: Overall, mobile sensing features related to internalizing symptoms, with the associations being outcome-specific and sometimes different between age groups: for example, at the bivariate level, screen usage features predicted greater momentary and baseline anxiety in youth only; greater mobility in a structural in a structural equation model predicted greater momentary anxiety variance in youth, and lesser momentary anxiety variance in adults. Random forest models indicated that mobile sensing features assessing all five behavioural categories offered some degree of utility in predicting anxiety in youth and adults.
Conclusions: The present study indicates that mobile sensing features can predict anxiety and anxiety variance in youth and adults, and that there are measurable differences in the expression of anxiety between youth and adults, many of which are consistent with prior literature. Future studies may benefit from larger sample sizes, as the present study was too small to rigorously assess the contribution of demographic moderators, and some effects may have failed to approach statistical significance due to small sample size. | en_US |
