Ethics
All research studies conducted by Sahha, either independently or in collaboration, have undergone rigorous ethical review process. This study was reviewed and approved by the University of Otago Human Ethics Committee, under Reference Number 21/074.
Executive Summary
This study aims to explore the reported associations between activity patterns and depression level as measured by the PHQ-9 using smartphone-based activity tracking.
Our results are consistent with reported associations between depression level and the activity patterns derived from wrist-worn trackers.
We observed sex-specific differences in the associations: activity regularity was statistically significant in males but not females, while activity in the early morning was statistically significant in females but not males.
Our findings indicate that the inference of depression levels from activity data captured from smartphones and wearable data is scientifically feasible.
Introduction
Background
Our mission at Sahha is to empower technologies with robust and scalable health analytics solutions using passively collected data. We aim to provide organizations with reliable measurements and inferences of their user’s health parameters. To achieve this, strong scientific bases behind our methods are necessary.
Mental wellness is a key aspect of overall health, and the drive to monitor and enhance it is rapidly gaining traction in the health-tech sector. A growing array of solutions is emerging to help users improve their mental well-being. Examples of this trend include guided meditation and mindfulness apps, which have become increasingly popular among users seeking to enhance their mental well-being.
With the rising adoption of wearable technology and the greater precision of smartphone sensor suites, it is now possible to capture a user’s activity and sleep patterns with impressive accuracy. A recent meta-analysis found that activity tracking improves activity levels equivalent to an additional 1,800 steps per day and reductions of approximately 1 kg in bodyweight. Another study on the use of sleep tracker apps found that the use of sleep tracker apps is associated with improved perceived mental well-being. These findings suggest that activity and sleep tracking have verifiable impacts on a user’s physical and mental wellbeing.
A natural progression in the technology would be the inference of a user’s mental states using activity and sleep pattern data. While wearables still tend to be more accurate than smartphone-based methods, the activity tracking capabilities and accuracies of smartphone-based sensors are rapidly catching up.
Tracking Mental State Using Activity Patterns
Multiple studies examining the associations between activity and mental states have emerged over recent years. While the methodologies used vary, a trend of participants with higher levels of depressive symptoms exhibiting lower overall activity levels and a higher proportion of their activity at night is consistent across studies.
A study by Banihashemi et al. (2016) recruited 168 participants with a history of affective disorders and 68 control participants. Diagnosis was determined using DSM-IV criteria. Activity was measured via wrist-worn actigraphy over a mean period of 14 days. Using functional smoothing and functional linear analysis, the investigators found that higher depression severity was associated with higher nighttime activity level.
A study by Difrancesco et al. (2019) recruited 359 participants with a history of affective disorders from the Netherlands Study of Depression and Anxiety (NESDA) cohort. Activity was measured via wrist-worn actigraphy over 14 days. Depression and anxiety levels were assessed using the inventory of depressive symptomatology (IDS) and Beck anxiety inventory (BAI) respectively. Using a linear regression analysis, the investigators found that higher levels of depressive and anxiety symptoms were associated with lower gross motor activity and a greater proportion of daily activity taking place at night.
A follow-up study utilizing 121 participants (63 controls) from the NESDA cohort examined depressive symptoms specifically. Using a linear mixed effect model, they found that the depression group exhibited a reduced difference between peak and mean activity level, and a preference for nighttime activity.
Another study utilized 1,800 participants from the 2011-2014 National Health and Nutrition Examination Survey (NHANES). Activity was measured using wrist-worn actigraphy and depression levels using the PHQ-9. The investigators performed clustering analysis and found that the clusters containing the highest proportion of participants with high PHQ-9 scores (greater than 10) are characterized by lower and later activity patterns along with more fragmented and unstable patterns.
Thus, there is growing literature indicating an individual’s mental state is often reflected in their activity. These findings suggest that inferring a user’s mental state using activity is possible. Existing studies examine this association using wrist-worn activity trackers. It is unknown if this association can be detected using smartphone-based activity trackers.
Study Aim
The aim of this study is to examine whether mental state, specifically depression level, is associated with activity pattern data tracked using smartphone-based methods. Depression level was measured via the PHQ-9. The PHQ-9 (Patient Health Questionnaire-9) is a self-report psychometric instrument for screening and measuring the level of depression experienced by the individual. It consists of nine items that assess various symptoms of depression, such as mood, interest in activities, and sleep pattern. Each item is scored based on the frequency of symptoms on a scale of 0-3, where 0 is “not at all” and 3 is “nearly every day”. A final score is calculated by summing the individual items. A score of 0-4 indicates minimal depression, 5-9 mild depression, 10-14 moderate depression, 15-19 moderately severe depression, and 20-27 severe depression.
While commonly used to screen for clinical depression in a primary care and research setting, the PHQ-9 score has also been shown to be a reliable dimensional measure of depression level. Through this study, we seek to examine the scientific basis of using smartphone sensors to infer depression level. We believe that the findings of this study will contribute materially to the field of mood tracking using smartphones.
Methodology
Participants Recruitment
Participants were recruited from participant recruitment services. Inclusion criteria include:
- Age from 18-65
- Has a smartphone (at least Android version 8+ or iPhone with internet connection)
- Proficient in English
Participants were recruited across 13 recruitments over a one-year period from February 2022 to February 2023.
Data Collection
After informed consent and acknowledgement of our data collection and privacy policies, participants were required to install the Sahha Research App on their smartphone. Participants then completed the enrollment procedure by providing information on age, sex, ethnicity, education attainment, income, country, and living arrangement.
The Sahha Research App collects information provided by Apple HealthKit (iOS) or Health Connect (Android), including step counts and sleep duration. Screen lock and unlock events were logged for Android users. The study period was one month, during which participants completed the PHQ-9 weekly. All data collected in this study were anonymized in accordance with the data collection and privacy policies.
Activity Pattern Measurements
Using step count data, we derived seven variables describing an individual’s daily activity pattern:
- Average active hours based on duration (hours with more than 1 minute of physical activity)
- Average active hours based on steps (hours with more than 100 steps)
- Activity regularity (consistency in active hours across days)
- Average number of steps in the early morning (5 am to 9 am)
- Average number of steps in the day (9 am to 6 pm)
- Average number of steps in the evening (6 pm to 9 pm)
- Average number of steps at night (9 pm to 5 am)
Statistical Analysis
We performed multivariate linear regression analysis with activity pattern measurements as predictor variables and PHQ-9 scores as the response variable. Distributions of variables were checked for normality using QQ plots. Linearity and homoskedasticity were checked using predicted vs residual plots. Residual normality, linearity, and homoskedasticity were within acceptable ranges.
To account for potential sex-specific differences, we performed the analysis separately for male and female participants. Age was included as a covariate in the models to account for reported associations between age and depression levels. Statistical significance was defined at α = 0.05. All analysis was performed on Python 3.10 with the statsmodels package 0.14.1.
Results
Participant Demographics
A total of 2,864 participants were recruited and participated in the study. After filtering out those with insufficient activity data and those who did not complete at least one PHQ-9 survey, measurements from 2,068 participants were retained for analysis. Across these participants, a total of 26,888 PHQ-9 surveys were collected. The median number of surveys per participant was 12.
Table 1: Participant demographics
| Demographic | n [%] | Mean age [SD] | Mean PHQ-9 score [SD] |
|---|---|---|---|
| Male | 798 [38.6] | 36.5 [11.4] | 32.9 [9.81] |
| Female | 1270 [61.4] | 33.8 [10.1] | 8.336 [5.79] |
| White/Caucasian | 1697 [81.2] | 35.3 [10.9] | 7.87 [5.88] |
| African descent | 83 [4.01] | 33.3 [8.75] | 5.2 [5.23] |
| Hispanic | 58 [2.8] | 31.7 [10.6] | 7.59 [6.43] |
| South Asian | 46 [2.22] | 29.2 [7.30] | 7.66 [6.99] |
| East Asian | 31 [1.5] | 29.2 [6.63] | 6.68 [6.45] |
| Middle Eastern | 16 [0.77] | 28.9 [8.94] | 8.52 [5.7] |
| Other Ethnicity | 201 [9.72] | 32.9 [9.81] | 7.57 [6.07] |
Statistical Analysis
Table 2: Multivariate regression results - Male
| Variable | Coefficient | Std. Error | t value | Pr(>|t|) |
|---|---|---|---|---|
| (Intercept) | -3.10E-17 | 0.035 | -8.91E-16 | 1.03E-12 |
| Average active hours (duration) | -0.1401 | 0.07 | -2.011 | 0.045 |
| Average active hours (steps) | 0.0273 | 0.074 | 0.369 | 0.712 |
| Activity regularity | -0.0756 | 0.038 | -2.012 | 0.045 |
| Average daytime activity | -0.0607 | 0.065 | -0.937 | 0.349 |
| Average evening activity | 0.0499 | 0.059 | 0.852 | 0.394 |
| Average night activity | 0.107 | 0.046 | 2.319 | 0.021 |
| Average early activity | -0.0949 | 0.051 | -1.858 | 0.064 |
| Age | -0.0361 | 0.035 | -1.027 | 0.305 |
Table 3: Multivariate regression results - Female
| Variable | Coefficient | Std. Error | t value | Pr(>|t|) |
|---|---|---|---|---|
| (Intercept) | 1.03E+01 | 1.48E+00 | 6.97E+00 | 5.00E-12 |
| Average active hours (duration) | -1.64E-01 | 7.43E-02 | -2.203 | 0.02781 |
| Average active hours (steps) | 5.54E-02 | 1.12E-01 | 0.493 | 0.62199 |
| Activity regularity | 3.69E-01 | 1.92E+00 | 0.193 | 0.84735 |
| Average daytime activity | -1.73E-05 | 4.55E-04 | -0.038 | 0.96964 |
| Average evening activity | -5.60E-04 | 5.72E-04 | -0.98 | 0.32747 |
| Average night activity | 4.26E-03 | 9.22E-04 | 4.616 | 4.31E-06 |
| Average early activity | -2.13E-03 | 8.05E-04 | -2.649 | 0.00817 |
| Age | -2.62E-02 | 5.91E-03 | -4.433 | 1.01E-05 |
Discussion
To our knowledge, this is the first study examining an association between activity patterns tracked via smartphones and depression levels within the general population. These associations are consistent with those found by studies utilising wrist-worn trackers.
For males, the coefficients of average active hours (duration) and activity regularity were negative while the coefficient for average night activity was positive. These results suggest that higher average active hours (duration) and activity regularity are associated with a lower depression level, while a higher level of night-time activity is associated with a higher depression level.
For females, the coefficients of average active hours (duration), average early activity, and age were negative, while the coefficient of average night activity was positive. These results suggest that higher average active hours (duration), activity in the early hours, and older age are associated with a lower depression level, while higher night-time activity is associated with a higher depression level.
The negative association between age and depression level is consistent with findings of a greater prevalence of depressive disorders within younger women. The different significant activity markers between male and female participants suggest that sex-specific models are necessary when designing mental health tracking models.
Conclusion
Our study found that the reported associations observed between depression levels and activity patterns are also present in the general population. Additionally, this association can be observed using smartphone-based data. The findings support a move toward smartphone-based technology in mental health tracking. The observed sex differences in these associations emphasize the importance of tailoring mental health strategies to account for gender-specific patterns.
References
- American Psychiatric Association. (1994). Diagnostic and statistical manual of mental disorders: DSM-IV. American Psychiatric Association.
- Attie, E., & Meyer-Waarden, L. (2023). How Do You Sleep? The Impact of Sleep Apps on Generation Z’s Well-Being. Journal of Interactive Marketing, 58, 222–247. https://doi.org/10.1177/10949968221142806
- Banihashemi, N., Robillard, R., Yang, J., Carpenter, J. S., Hermens, D. F., Naismith, S. L., Terpening, Z., White, D., Scott, E. M., & Hickie, I. B. (2016). Quantifying the effect of body mass index, age, and depression severity on 24-h activity patterns in persons with a lifetime history of affective disorders. BMC Psychiatry, 16, 1–11.
- Beck, A. T., Epstein, N., Brown, G., & Steer, R. A. (1988). An inventory for measuring clinical anxiety: Psychometric properties. Journal of Consulting and Clinical Psychology, 56, 893–897. https://doi.org/10.1037/0022-006X.56.6.893
- Bianchi, R., Verkuilen, J., Toker, S., Schonfeld, I. S., Gerber, M., Brähler, E., & Kroenke, K. (2022). Is the PHQ-9 a unidimensional measure of depression? A 58,272-participant study. Psychological Assessment, 34, 595–603. https://doi.org/10.1037/pas0001124
- Caputo, E. L., Feter, N., Alberton, C. L., Leite, J. S., Rodrigues, A. N., Dumith, S. de C., & Silva, M. C. da. (2022). Reliability of a smartphone application to measure physical activity. Research in Sports Medicine, 30, 264–271.
- Chen, T.-C., Parker, J. D., Clark, J., Shin, H.-C., Rammon, J. R., & Burt, V. L. (2018). National health and nutrition examination survey: estimation procedures, 2011–2014.
- Difrancesco, S., Lamers, F., Riese, H., Merikangas, K. R., Beekman, A. T., van Hemert, A. M., Schoevers, R. A., & Penninx, B. W. (2019). Sleep, circadian rhythm, and physical activity patterns in depressive and anxiety disorders: A 2-week ambulatory assessment study. Depression and Anxiety, 36, 975–986.
- Difrancesco, S., Riese, H., Merikangas, K. R., Shou, H., Zipunnikov, V., Antypa, N., van Hemert, A. M., Schoevers, R. A., Penninx, B. W., & Lamers, F. (2021). Sociodemographic, health and lifestyle, sampling, and mental health determinants of 24-hour motor activity patterns: observational study. Journal of Medical Internet Research, 23, e20700.
- Ferguson, T., Olds, T., Curtis, R., Blake, H., Crozier, A. J., Dankiw, K., Dumuid, D., Kasai, D., O’Connor, E., Virgara, R., & Maher, C. (2022). Effectiveness of wearable activity trackers to increase physical activity and improve health: A systematic review of systematic reviews and meta-analyses. The Lancet Digital Health, 4, e615–e626. https://doi.org/10.1016/S2589-7500(22)00111-X
- Hawthorne, G., Goldney, R., & Taylor, A. W. (2008). Depression Prevalence: Is it Really Increasing? Australian & New Zealand Journal of Psychiatry, 42, 606–616. https://doi.org/10.1080/00048670802119788
- Kroenke, K., Spitzer, R. L., & Williams, J. B. (2001). The PHQ-9: validity of a brief depression severity measure. Journal of General Internal Medicine, 16, 606–613.
- Martin, A., Rief, W., Klaiberg, A., & Braehler, E. (2006). Validity of the Brief Patient Health Questionnaire Mood Scale (PHQ-9) in the general population. General Hospital Psychiatry, 28, 71–77. https://doi.org/10.1016/j.genhosppsych.2005.07.003
- Minaeva, O., Booij, S. H., Lamers, F., Antypa, N., Schoevers, R. A., Wichers, M., & Riese, H. (2020). Level and timing of physical activity during normal daily life in depressed and nondepressed individuals. Translational Psychiatry, 10, 259.
- Modave, F., Guo, Y., Bian, J., Gurka, M. J., Parish, A., Smith, M. D., Lee, A. M., & Buford, T. W. (2017). Mobile device accuracy for step counting across age groups. JMIR mHealth and uHealth, 5, e7870.
- Neary, M., & Schueller, S. M. (2018). State of the Field of Mental Health Apps. Cognitive and Behavioral Practice, 25, 531–537. https://doi.org/10.1016/j.cbpra.2018.01.002
- Penninx, B. W., Beekman, A. T., Smit, J. H., Zitman, F. G., Nolen, W. A., Spinhoven, P., Cuijpers, P., De Jong, P. J., Van Marwijk, H. W., & Assendelft, W. J. (2008). The Netherlands Study of Depression and Anxiety (NESDA): rationale, objectives and methods. International Journal of Methods in Psychiatric Research, 17, 121–140.
- Piccinini, F., Martinelli, G., & Carbonaro, A. (2020). Accuracy of Mobile Applications versus Wearable Devices in Long-Term Step Measurements. Sensors, 20, 6293. https://doi.org/10.3390/s20216293
- Rush, A. J., Gullion, C. M., Basco, M. R., Jarrett, R. B., & Trivedi, M. H. (1996). The inventory of depressive symptomatology (IDS): psychometric properties. Psychological Medicine, 26, 477–486.
- Seabold, S., & Perktold, J. (2010). statsmodels: Econometric and statistical modeling with python. In Proceedings of the 9th Python in Science Conference.
- Smagula, S. F., Zhang, G., Gujral, S., Covassin, N., Li, J., Taylor, W. D., Reynolds, C. F., & Krafty, R. T. (2022). Association of 24-hour activity pattern phenotypes with depression symptoms and cognitive performance in aging. JAMA Psychiatry, 79, 1023–1031.