Leveraging low-cost app-based step count data to assess depression and anxiety in university students: A cross-sectional mobile health study.

BACKGROUND: The prevalence of depression and anxiety among college students worldwide is on the rise, significantly impacting their health and quality of life. Traditional mental health screening scales have some limitations, including subjectivity and recall bias. Therefore, developing rapid, low-cost objective detection technologies is needed. METHODS: We designed a low-cost, app-based data collection system to collect step count data and self-reported depression and anxiety questionnaires. This cross-sectional study included 578 participants with 28 days of step count data. Beyond basic statistical step metrics, we extracted frequency domain and non-linear features to capture lifestyle periodicity and regularity, and we novelly introduced a weekday-weekend activity fluctuation feature to quantify behavioral variability. We examined the associations between these features and depression/anxiety severity, and further evaluated machine learning models for depression/anxiety status classification. RESULTS: Several step count features showed significant associations with higher depression scores, including lower overall steps, reduced lifestyle regularity, and greater weekday-weekend fluctuations. Anxiety severity was also associated with lower regularity in step counts. In binary classification, the depression model achieved acceptable performance (AUC = 0.802, 95 % CI: 0.766-0.838; accuracy = 71.1 %, 95 % CI: 69.8 %-75.5 %). LIMITATIONS: The cross-sectional study cannot explain causal relationships. Additionally, unmeasured confounding factors may influence the observed associations. CONCLUSIONS: This study demonstrates that step count-derived features are significantly associated with depression and anxiety severity, highlighting the potential of passive sensing data as a scalable, low-cost screening approach for mental health assessment.