Longitudinal Changes in Activity and Sleep Behaviors from Activity ... · Longitudinal activity and...
1
Methods STUDY POPULATION ◆ In 2017, members of the Achievement online health community (a product of Evidation Health) were invited to participate in a 1-year observational cohort study on mental health. ◆ Participants were 18 years or older, lived in the U.S., and had a self- reported diagnosis of anxiety or depression. BEHAVIORAL AND HEALTH OUTCOMES DATA ◆ Participants tracked their activity and sleep levels using consumer wearable activity trackers, and recorded weight and food logs using health and fitness apps. ◆ Surveys were completed at baseline, and quarterly thereafter to assess depression as measured by the Patient Health Questionnaire 9-item Scale (PHQ-9). ◇ PHQ-9 scores range from 0 to 27. An Increase in score indicates worsening depression. STATISTICAL ANALYSIS ◆ Steps and sleep behavior data were examined for intraday data availability and completeness for each 90-day window prior to the first four survey timepoints (baseline, months 3, 6, and 9). ◆ Behavioral features were constructed for each 90-day panel window. ◆ Using a random effects framework, we conducted univariate regressions to identify associations between behavioral data features and change in PHQ-9. Longitudinal Changes in Activity and Sleep Behaviors from Activity Trackers Are Associated with Change in Depression Severity Ernesto Ramirez, PhD; Jennifer L. Tran, PharmD; Benjamin Bradshaw, BS; Shefali Kumar, MPH; Caroline Tai, PhD; Luca Foschini, PhD; Jessie L. Juusola, PhD Background ◆ Consumer digital health tools have created new opportunities to explore relationships between everyday behavior and mental health status. ◆ Intense longitudinal objective data related to physical activity and sleep may provide insights related to changes in behavior and changes in mental health status over time. Evidation Health, San Mateo, CA evidation.com Objectives ◆ We sought to examine the relationship between change in activity and sleep behaviors and change in self-reported depression status over a 9-month period. Results STUDY POPULATION UNIVARIATE REGRESSION ANALYSIS ◆ Multiple features were found to be significantly associated with increased PHQ-9 scores over time: ◇ More time spent in bed on days with the longest sleep periods. ◇ More time spent awake in bed on the most restless nights. ◇ Later time of day of the highest 3-minute rolling average of steps per minute. ◇ Greater variability in the time spent being active. ◆ A greater number of minutes per day with nonzero step counts was associated with decreased PHQ-9 scores. Limitations ◆ Limiting analysis to participants with dense interday data reduced the sample size from the ~1,300 enrolled to 639. ◆ The study population was largely female and young, limiting generalizability of the results. Conclusions ◆ For a given participant, spending more time in bed (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 throughout the day over time was significantly associated with improvement in depressive symptoms. ◆ Longitudinal activity and sleep data from consumer wearables may serve as a potential indicator of worsening depression in individuals with depression or anxiety. ◇ Future research is needed to validate these findings and to explore the utility of wearable tracker data as potential digital biomarkers for mental health status. Behavioral Data Feature Regression Coefficient p-value 1 Total minutes in bed on 95 th percentile days 0.002 0.001 Total minutes awake while in bed on 95 th percentile days 0.005 0.005 Time of day of 95 th percentile for rolling 3-minute average step count 0.212 0.026 Skewness of the fraction of the day with step counts > 100 steps per minute 0.169 0.035 Fraction of the day with nonzero step counts -8.784 0.026 1 P-values were corrected for false discovery rates (FDR) to minimize the Type 1 error rate from multiple comparisons using the 2-stage Benjamini-Hochberg method. Demographics N = 639 Age, mean years (SD) 33.1 (8.8) Female, n (%) 537 (84.0) Baseline PHQ-9 score, mean (SD) 9.0 (5.7) Tracker Data Coverage, median Panel 0 (-90 days to baseline) Panel 1 (baseline to month 3) Panel 2 (month 3 to month 6) Panel 3 (month 6 to month 9) 89 days 90 days 89 days 86 days ◇ Minute-level tracker data were used to construct over 300 data features capturing behavioral aspects of daily living. ◇ Daily features were aggregated using the mean, 5th/95th percentile, standard deviation, and other distribution characteristics. Feature Creation Conceptual Flow
Transcript of Longitudinal Changes in Activity and Sleep Behaviors from Activity ... · Longitudinal activity and...
MethodsSTUDY POPULATION◆ In 2017, members of the Achievement online health community (a
product of Evidation Health) were invited to participate in a 1-year observational cohort study on mental health.
◆ Participants were 18 years or older, lived in the U.S., and had a self-reported diagnosis of anxiety or depression.
BEHAVIORAL AND HEALTH OUTCOMES DATA◆ Participants tracked their activity and sleep levels using consumer
wearable activity trackers, and recorded weight and food logs using health and fitness apps.
◆ Surveys were completed at baseline, and quarterly thereafter to assess depression as measured by the Patient Health Questionnaire 9-item Scale (PHQ-9).
◇ PHQ-9 scores range from 0 to 27. An Increase in score indicates worsening depression.
STATISTICAL ANALYSIS◆ Steps and sleep behavior data were examined for intraday data
availability and completeness for each 90-day window prior to the first four survey timepoints (baseline, months 3, 6, and 9).
◆ Behavioral features were constructed for each 90-day panel window.
◆ Using a random effects framework, we conducted univariate regressions to identify associations between behavioral data features and change in PHQ-9.
Longitudinal Changes in Activity and Sleep Behaviors from Activity Trackers Are Associated with Change in Depression SeverityErnesto Ramirez, PhD; Jennifer L. Tran, PharmD; Benjamin Bradshaw, BS; Shefali Kumar, MPH; Caroline Tai, PhD; Luca Foschini, PhD; Jessie L. Juusola, PhD
Background◆ Consumer digital health tools have created new opportunities to explore relationships between everyday behavior and mental health status.◆ Intense longitudinal objective data related to physical activity and sleep may provide insights related to changes in behavior and changes in mental
health status over time.
Evidation Health, San Mateo, CA evidation.com
Objectives◆ We sought to examine the relationship between change in activity and sleep behaviors and change in self-reported depression status over a 9-month
period.
ResultsSTUDY POPULATION
UNIVARIATE REGRESSION ANALYSIS◆ Multiple features were found to be significantly associated with
increased PHQ-9 scores over time:◇ More time spent in bed on days with the longest sleep periods.◇ More time spent awake in bed on the most restless nights.◇ Later time of day of the highest 3-minute rolling average of steps
per minute. ◇ Greater variability in the time spent being active.
◆ A greater number of minutes per day with nonzero step counts was associated with decreased PHQ-9 scores.
Limitations◆ Limiting analysis to participants with dense interday data reduced the
sample size from the ~1,300 enrolled to 639.◆ The study population was largely female and young, limiting
generalizability of the results.
Conclusions◆ For a given participant, spending more time in bed (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 throughout the day over time was significantly associated with improvement in depressive symptoms.◆ Longitudinal activity and sleep data from consumer wearables may serve as a potential indicator of worsening depression in individuals with
depression or anxiety.◇ Future research is needed to validate these findings and to explore the utility of wearable tracker data as potential digital biomarkers for mental
health status.
Behavioral Data Feature Regression Coefficient p-value1
Total minutes in bed on 95th percentile days 0.002 0.001
Total minutes awake while in bed on 95th percentile days 0.005 0.005
Time of day of 95th percentile for rolling 3-minute average step count 0.212 0.026
Skewness of the fraction of the day with step counts > 100 steps per minute 0.169 0.035
Fraction of the day with nonzero step counts -8.784 0.0261 P-values were corrected for false discovery rates (FDR) to minimize the Type 1 error rate from multiple comparisons using the 2-stage Benjamini-Hochberg method.
Demographics N = 639Age, mean years (SD) 33.1 (8.8)Female, n (%) 537 (84.0)Baseline PHQ-9 score, mean (SD) 9.0 (5.7)Tracker Data Coverage, median
Panel 0 (-90 days to baseline)Panel 1 (baseline to month 3)Panel 2 (month 3 to month 6) Panel 3 (month 6 to month 9)
89 days90 days89 days86 days
◇ Minute-level tracker data were used to construct over 300 data features capturing behavioral aspects of daily living.
◇ Daily features were aggregated using the mean, 5th/95th percentile, standard deviation, and other distribution characteristics.
Feature Creation Conceptual Flow
