Introduction The widespread availability of consumer digital health tools has created opportunities to explore the relationship between everyday behavior, changes in behavior, and mental health status using intensive time series data that may more accurately depict physical activity and sleep behaviors over time compared to self-reported behavioral data. We conducted a virtual, one-year observational study to explore associations between longitudinal objective measures of health behaviors and mental health status. Methods Adults with self-reported anxiety or depression were invited to enroll in the study and asked to link their activity tracker accounts for data access. Depression severity as measured by the Patient Health Questionnaire-9 (PHQ-9) was collected quarterly. For this analysis, behavioral data features were constructed for each 90-day period prior to the baseline, months 3, 6, and 9 surveys. Minute-level steps and sleep tracker data were used to construct 343 data features capturing behavioral aspects of daily living. The daily behavioral data features were aggregated over each distinct 90-day panel period using the mean, 5th/95th percentile, standard deviation, and other distribution characteristics. Using a random effects framework, we conducted univariate regressions to identify associations between behavioral data features and change in PHQ-9. P-values were adjusted for multiple testing via false discovery rate correction. Results A total of 639 participants had minute-level tracker data and completed the PHQ-9 at each of the 4 time points. The following behavioral data features were significantly associated with increased PHQ-9 scores (i.e. worsening depression) over time: more time spent in bed on days with the longest sleep periods (𝝱=0.002, p<0.05), more time spent awake in bed on the most restless nights (𝝱=0.005, p<0.05), later time of day of the highest 3-minute rolling average of steps per minute (𝝱=0.21, p<0.01), and greater variability in the time spent being active (𝝱=0.17, p<0.05). Alternatively, a greater number of minutes per day with non-zero step counts was associated with decreased PHQ-9 scores (𝝱=-8.784, p<0.01). Conclusions Longitudinal activity and sleep data from consumer wearables may serve as a potential indicator of worsening depression in individuals with depression or anxiety. We found that within-subject changes related to spending more time in bed (either awake or asleep) over time, becoming more active later in the day, and having more fluctuations in activity over time were all significantly associated with worsening depression. Becoming more active in the day over time was significantly associated with improvement in depressive symptoms. Further research is warranted to validate these findings as potential digital biomarkers, and to explore how real-time behavioral information can be leveraged to improve treatment.