Abstract
Psychosocial stress contributes to heart disease in part by adversely affecting maintenance of health behaviors, while exercise can reduce stress. Assessing the bi-directional relationship between stress and exercise has been limited by lack of real-time data and theoretical and statistical models. This lack may hinder efforts to promote exercise maintenance.
We test the bi-directional relationship between stress and exercise using real-time data for the average person and the variability—individual differences—in this relationship.
An observational study was conducted within a single cohort randomized controlled experiment. Healthy young adults, (n = 79) who reported only intermittent exercise, completed 12 months of stress monitoring by ecological momentary assessment (at the beginning of, end of, and during the day) and continuous activity monitoring by Fitbit. A random coefficients linear mixed model was used to predict end-of-day stress from the occurrence/non-occurrence of exercise that day; a logistic mixed model was used to predict the occurrence/non-occurrence of exercise from ratings of anticipated stress. Separate regression analyses were also performed for each participant. Sensitivity analysis tested all models, restricted to the first 180 days of observation (prior to randomization).
We found a significant average inverse (i.e., negative) effect of exercise on stress and of stress on exercise. There was significant between-person variability. Of N = 69, exercise was associated with a stress reduction for 15, a stress increase for 2, and no change for the remainder. We also found that an increase in anticipated stress reported the previous night or that morning was associated with a significant 20–22% decrease (OR = 0.78–0.80) in the odds of exercising that day. Of N = 69, this increase in stress reduced the likelihood of exercise for 17, increased the odds for 1, and had no effect for the remainder. We were unable to identify psychosocial factors that moderate the individual differences in these effects.
The relationship of stress to exercise can be uni- or bi-directional and varies from person to person. A precision medicine approach may improve exercise uptake.
Compliance with Ethical Standards
Authors' Statement of Conflict of Interest and Adherence to Ethical Standards Karina Davidson is the co-owner of MJBK, a small business that provides mhealth technology solutions to consumers and the co-owner of IOHealthWorks, a small consulting services company. No other potential conflicts of interest relevant to this article were reported.
Funding This work was supported by a grant from the National Institutes of Health (R01HL115941) to Drs. Burg and Davidson.
Ethical Approval All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards.
Informed Consent Informed consent was obtained from all individual participants included in the study.
Introduction
There are long-held beliefs that psychosocial stress contributes to diseases of the heart [c.f., 1]. While the contribution of stress may be through direct effects on cardiac physiology, another key mechanism may be through the adverse effects of stress on the likelihood of engaging in health behaviors [c.f., 2, 3]. Failure to maintain health behaviors such as regular physical activity accounts for up to 40% of the risk of incident cardiovascular disease (CVD) [4]. While prior approaches to understanding the impact of stress on health behaviors have been informative, significant gaps remain in our understanding of the pathways linking stress exposure and stress perceptions to CVD-relevant health behaviors. Furthermore, our ability to assess complex, potentially bi-directional relationships between stress and health behaviors has been limited by dated theoretical, measurement, and statistical models.
The benefits of regular physical activity on health are widely recognized [5–9], with research also demonstrating benefits for emotional well-being, and positive effects on brain, cardiovascular, and autonomic nervous system function, with particular positive effects on stress sensitivity [c.f., 9–13]. A review of research concerning physical activity and stress [14] revealed a consistent reduction in stress reactivity among individuals who achieved a level of aerobic conditioning through regular exercise, and a recent meta-analysis [15] reported that a single bout of aerobic exercise of moderate to high intensity significantly reduced the hemodynamic response to a wide range of laboratory stressors. Regular physical activity can also reduce anger and hostility [16], each linked to more frequent and more exaggerated reactivity to momentary stress [17–18].
The importance of a healthy lifestyle that includes regular physical activity is largely understood; yet, the broad adoption and maintenance of this lifestyle remain elusive [19]. While regular physical activity is widely recommended for improving health outcomes, clinical trials focused on implementation of a regular program of exercise are most clearly defined by failure of long-term maintenance [20]. Thus, research is increasingly focused on methods to improve exercise maintenance [21–24]. Chronic and acute stress, and the demanding features of daily life, can serve as a barrier to good health practices and the perpetuation of an unhealthy lifestyle [25]. Indeed, two recent reviews [26, 27] found that episodic stress—for example, during periods of varying stress such as final examinations and work demands—predicts less physical activity/intentional exercise and more sedentary behavior. In one study reviewed [28], it was found that a single stressful episode reduced subsequent total energy expenditure. The authors concluded that the experience of stress impairs efforts to be physically active.
Given that regular physical activity can, for some, mitigate the physiological impact of chronic and acute stress while improving aspects of stress resilience [16], the relationship of these factors may shed light on how best to approach the issue of maintenance. The nature of this relationship may, however, be notably complex. For example, this relationship for some may be bi-directional in nature, while for others, it may be uni-directional, and yet for others, there may be no association. Unfortunately, our ability to assess such complex, potentially bi-directional associations on a per person basis—in this case, between stress exposure and a daily bout of exercise—has been limited by our reliance on nomothetic or between-person methods, which involve the study of classes or cohorts of individuals. Using these methods, we have not been able to discern the elements of the environment-person milieu that contribute to adoption and maintenance of regular exercise vs. failure.
Idiographic—or within-subject—methods describe the study of the individual as the sole population to which we want to generalize. These methods are based on the assumption that unique within-subject associations and causal factors may exist and cannot easily be discovered when between-subject designs are employed. The use of an idiographic or within-subject method to understand the unique association(s) of daily stress and daily exercise may reveal interesting and differing patterns for different persons. This within-subject approach is now possible with ecological momentary assessment (EMA) and continuous measurement of physical activity, e.g., with an activity monitor. With EMA, the individual reports on current and recent aspects of the environment, including the associated person-level experience and behaviors engaged in. EMA allows for repeated and random monitoring of stress as it occurs in ecologically valid settings, such as home and work, and all in real time [29–31]. Coupling EMA with a monitoring device that objectively measures both the occurrence of physical activity and the level of exertion during that activity provides for prospective ascertainment of the effect that stress has on whether a given individual engages in exercise and how engaging in exercise influences the subsequent experience of stress. Powerful statistical modeling methods that allow for the testing of complex and bi-directional relationships among these factors at the level of the single subject have also recently become available. These methods can model this complex set of continuous data collected over months to discern, prospectively, the diverse and bi-directional nature of the stress-(exercise) behavior relationship in general and at the level of the individual.
The topic of stress was among those identified in 2010 as high priority by the National Institutes of Health (NIH) Basic Behavioral and Social Sciences Research Opportunity Network (OppNet) to support research into the mechanisms, processes, and pathways of behavior related to health and underlying multiple diseases and conditions [32]. The topics included identification of bi-directional pathways between stress and health-relevant behavior. In this paper based on an OppNet-funded investigation, we report on the bi-directional relationship between stress and exercise for the average person and at the level of the individual.
Methods
Overview
This was an observational study conducted within a single cohort, 12-month randomized controlled experiment of personal daily stress and exercise. Potentially eligible participants were identified and screened. Study methods were described, and potential participants were asked to consider their ability to complete study procedures over a 1-year period. Those remaining eligible who agreed to the study procedures completed an assessment battery and initiated daily assessments by EMA and actigraphy.
Study Participants
Participants were healthy individuals aged 18 or older who responded to fliers posted throughout the buildings of Columbia University Medical Center and who on phone screening reported only intermittent engagement in exercise, having access to a personal computer, and having an iPhone or Android phone. Excluded were individuals who had previously been told by a healthcare professional to restrict physical activity, were deemed unable to comply with the protocol (either self-selected, by indicating during screening that s/he could not complete all requested tasks), were unavailable during the following continuous 12 months, had serious medical comorbidity that would compromise their ability to engage in usual physical activity, had occupational work demands that required rigorous activity or would make responding to the EMA dangerous, or were unable to read and speak English.
Measures
Activity Monitoring
Physical activity was continuously and objectively monitored for the 12 months of participation using a wrist worn Fitbit activity monitor (http://www.fitbit.com). The Fitbit device tracks the wearer's daily physical activity, including steps, distance walked, stairs climbed, and calories burned. The Fitbit contains an accelerometer, which provides for tracking of activity more accurately than pedometers, plus an altimeter for tracking stairs and hills climbed, and we have previously validated the device for measuring physical activity [33]. Fitbit also classifies the intensity of physical activity for each minute throughout the day. Data from the device are automatically uploaded to the Fitbit website (www.Fitbit.com) whenever the device was within 15 ft of the base station, which was plugged into the participant's own Mac or PC. Participants were instructed to sync and charge their device at least every 5–7 days to ensure no loss of activity data.
The Fitbit was chosen for the study because it is simple to use and more convenient than other activity monitoring devices, which require the participant to return to the study office for syncing and battery charging. With Fitbit, participants could sync data and charge the device on their own. With additional programming, we were able to extract data from the backend of the Fitbit.com website for analysis. To use the device properly, health information of a very limited nature (e.g., gender, weight, height, date of birth) was entered on the Fitbit website during the initialization process. These data were necessary for estimating user-specific activity calculations, such as calories burned.
The device was initialized, and a profile was created during the first study visit. Log in credentials were not provided to the participant, thus ensuring that the participant remained blinded and did not have access to their own data. Fitbit data were downloaded to a server managed by the Columbia University Medical Center IT group. After the study, participants were given their Fitbit to keep and given their Fitbit.com log in credentials so they could continue using the activity tracker device if they chose to.
Ecological Momentary Assessments
An electronic diary that used the participant's own iPhone or Android phone was used to capture momentary and summary aspects of their stress experience. Each morning upon rising, the participant responded to questions on the browser asking them, (1) “How stressful do you expect today to be?” and 2) “How likely are you to exercise today?”, each answered on a scale from 0 (not at all) to 10 (extremely). Similarly, each evening, the participant responded to questions on the browser asking them, (1) “Overall, how stressful was your day?” and (2) “Overall, how stressful do you think tomorrow will be?”, again with each question answered on a scale from 0 (not at all) to 10 (extremely). Lastly, they were asked whether they had exercised for 30 min or more that day (yes/no). The current report describes the relationship(s) of these beginning and end-of-day reports to objectively and subjectively assessed physical activity.
In addition, the system was programmed to prompt the participant via text message or email three random times per day during their preset hours of wakefulness (e.g., 7 a.m.–10 p.m.), with the specific time period programmed individually according to the participant's own sleep schedule. Daily notifications were separated by at least 1 h. Each text message or email contained a randomly generated URL linked to a mobile web browser page that expired 70 min after notification was sent. The system tracked when the notification went out and when the participant started and completed the assessment. If an instance of the survey was closed before finishing, participants were able to return to the survey via the URL to complete it, unless they had already been prompted for completion of their next survey. Each of the three daytime momentary assessments included questions concerning (1) key sources of stress (one screen listing possible sources of stress including work, home, financial, inter-personal, time/scheduling pressure, daily hassle, general/other—with the participant checking all that apply) and (2) stress appraisal, using the four-item Perceived Stress Scale [34]. The surveys could be completed on smartphone/mobile web browser or on a computer web browser. Data transmission was secured via SSL (i.e., https) and sent to a server managed by the Columbia University Medical Center IT group.
Additional Measures
A number of potential psychosocial moderators of the relationship between perceived stress and physical activity were assessed. These included chronic perceived stress (Perceived Stress Scale (PSS Scale)) [34], optimism (Life Orientation Test (LOT)) [35], perceptions of and satisfaction with social support [36], sense of mastery [37], anxiety sensitivity [38], and demographics.
All self-report instruments including EMA were administered via Qualtrics.
Randomized Controlled Experiment
This was a 12-month randomized controlled experiment. After 6 months of observation, half of the participants received either their personalized “stress-exercise fingerprint” or general descriptive information about their exercise and reported stress. The “fingerprint” message started with the template sentence “Looking at a number of daily combinations, we analyzed your physical activity and smartphone survey responses for the last six months to explore any relationships between your activity and experiences of stress.” The second sentence conveyed a personalized predictive message: “We found that your [personalized predictors of exercise] have an effect on how much you exercise [personalized exercise behavior]. This is followed by the last sentence restating the previous sentence; “In other words, [personalized predictor phrases], the more likely you are to exercise [personalize].” Control participants received a similar three-sentence message but without a description of the stress-exercise relationship. Observation continued for an additional 6 months after receipt of this message.
Statistical Analyses
Data Management
The analyses reported the following utilized three streams of data.
First, there was person-level data, e.g., demographic characteristics and scale scores computed from psychosocial questionnaires administered at baseline.
Second, there was the Fitbit data, in which the activity level of every minute of every day was categorized as sedentary, light, moderate, or vigorous. We defined a “social day” as the period from 3 a.m. 1 day until 2:59 a.m. the next day. To define a bout of exercise as any consecutive 30-min period within which 24 or more minutes were classified as moderate or vigorous intensity, we followed the recommendations of Ward et al. [39] regarding best practices for the use of accelerometer data in research on physical activity. Specifically, physical activity guidelines recommend exercise in bouts of 10 min or more for at least 30 min a day, while accommodating interruptions. Thus, when analyzing accelerometer data, the approach is to quantify exercise in bouts of 10 min with allowances for 2 min of interruption (e.g., 8 out of 10 min). Extrapolating the definition based on 10 to 30 min, this yielded 24 out of 30 min. Software was written to determine, for each day, whether there was any 30-min period within which at least 24 min were classified as moderate or vigorous activity; this was our objectively assessed measure of a 30-min period of exercise. We also calculated the number of hours (3:00–3:59 a.m., 4:00–4:59 a.m., 2:00–2:59 a.m.) in which the Fitbit recorded 10 or more steps; hours with fewer than 10 steps were treated as non-wear periods. Only days with a minimum of 10 h of wear time were included in the analyses.
Third, there was the EMA data, specifically the morning EMA report of anticipated stress level for the day, the evening EMA report of anticipated stress level for the next day, and the evening EMA report summarizing the actual stress for that day.
The analysis file had one observation for each person-day, containing these EMA reports, the number of hours the Fitbit was worn that day and whether a 30-min bout of exercise had occurred, and the person-level measures. Access to the physical activity study dataset and information about the study's execution and materials are available publicly at https://osf.io/kmszn/.
Test of Hypothesis that Exercise Affects Stress
All analyses were performed in SAS (version 9.4). The random coefficient linear regression models were estimated using the MIXED procedure, and the random coefficient linear regression models were estimated using the GLIMMIX procedure with a logit link function. All model estimates and hypothesis tests were based on the results from the multilevel analyses.
A random coefficients linear mixed model predicting the EMA end-of-day stress rating from the occurrence/non-occurrence of a bout of exercise that day was estimated. Both the intercept and the coefficient of exercise were treated as random effects, with an unstructured covariance matrix. This analysis yielded an estimate of the average within-person effect of exercise on end-of-day stress, a test of whether this average effect was significantly different from zero, an estimate of person-to-person variability (i.e., individual differences) in this effect, and a test of whether this variability was statistically significant.
We also performed a separate regression analysis for each participant, obtaining an estimate (and 95% confidence interval) of the effect of exercise on stress for each person. These estimates are portrayed graphically in a forest plot, and the number of participants exhibiting a significant positive association, significant negative association, and non-significant association is reported. In addition, a finding of significant individual differences raises the obvious question of whether there are person characteristics (e.g., gender, social support) that might account for these differences, i.e., whether we can identify person characteristics that moderate the effect of exercise on the end-of-day stress rating. Such effect modification was tested by the addition of person-level characteristics and their interaction with exercise to the original random coefficients linear mixed model analysis.
Test of Hypothesis that the Morning EMA Report of Anticipated Stress for the Day Affects the Likelihood of Exercising that Day
The analysis of this hypothesis is essentially the same, except that the outcome is binary. Accordingly, a random coefficients logistic mixed model predicting the occurrence/non-occurrence of a bout of exercise from the morning EMA rating of anticipated stress was estimated. Given significant individual differences, a separate logistic regression was estimated for each participant, and a forest plot showing the odds ratio of exercise per five-point change in morning stress rating was generated. Again, we tested for potential effect modification with respect to a select number of possible moderators.
Because of the randomization that occurred at 180 days in the context of the randomized controlled experiment, sensitivity analyses were used to test all models, restricted to the first 180 days of observation.
Results
Overall, 194 individuals completed online screening of whom 61 were found ineligible. Of the remaining 133, 80 individuals completed the baseline run-in period, with 1 being administratively removed, leaving a final study sample of 79 healthy individuals, 6 of whom withdrew over the first 6 months of the study. A total of 63 participants provided at least 11 months of data for analyses (see CONSORT diagram, Fig. 1). Mean (±SD) age was 31.9 (±9.5 years), 43% were male, 14% were African American, and 28% were Hispanic. In addition, 85.7% reported at least a college degree, 58.8% were single, and 14.3% were living alone (see Table 1).
Consort diagram
Baseline demographic characteristics of 79 participants
| Characteristics | N (%) | Mean (SD) |
|---|---|---|
| Average age (years) | 31.9 (9.5) | |
| Height (cm) | 168.5 (8.5) | |
| Weight (kg) | 75.3 (17.4) | |
| BMI (kg/m2) | 26.4 (5.3) | |
| Gender | ||
| Men | 34 (43.0%) | |
| Women | 45 (57.0%) | |
| Race | ||
| American Indian/Alaska Native | 0 | |
| Asian | 15 (19.0%) | |
| Black/African American | 11 (13.9%) | |
| Native Hawaiian/Pacific Islander | 3 (3.8%) | |
| White | 33 (41.8%) | |
| Two or more | 4 (5.1%) | |
| Unknown/declined (mostly Hispanic) | 13 (16.5%) | |
| Ethnicity | ||
| Hispanic | 22 (27.8%) | |
| Non-Hispanic | 56 (70.9%) | |
| Unknown/declined (multiple races) | 1 (1.3%) | |
| Education | ||
| Less than high school | 0 | |
| Some high school | 0 | |
| High school diploma/GED | 1 (1.3%) | |
| Some college | 12 (15.2%) | |
| College degree | 34 (43.0%) | |
| Graduate/professional school | 32 (40.5%) | |
| Unknown/declined | 0 | |
| Partner status | ||
| Single | 45 (57.0%) | |
| Partner/spouse | 32 (40.5%) | |
| Separated | 0 | |
| Widowed | 0 | |
| Divorced | 2 (2.5%) | |
| Living situation | ||
| Live alone | 9 (11.4%) | |
| Live with roommates | 11 (13.9%) | |
| Live with family | 39 (49.4%) | |
| Live with partner/significant other | 20 (25.3%) | |
| Number of days participated | 331.1 (76.2) | |
| Fitbit worn ≥10 h | ||
| Number of days | 228.4 (89.2) | |
| % of days | 67.9 (18.8) | |
| Morning EMA report completed | ||
| Number of days | 201.1 (87.7) | |
| % of days | 59.5 (20.8) | |
| Evening EMA report completed | ||
| Number of days | 213.7 (90.9) | |
| % of days | 63.4 (21.2) | |
| Characteristics | N (%) | Mean (SD) |
|---|---|---|
| Average age (years) | 31.9 (9.5) | |
| Height (cm) | 168.5 (8.5) | |
| Weight (kg) | 75.3 (17.4) | |
| BMI (kg/m2) | 26.4 (5.3) | |
| Gender | ||
| Men | 34 (43.0%) | |
| Women | 45 (57.0%) | |
| Race | ||
| American Indian/Alaska Native | 0 | |
| Asian | 15 (19.0%) | |
| Black/African American | 11 (13.9%) | |
| Native Hawaiian/Pacific Islander | 3 (3.8%) | |
| White | 33 (41.8%) | |
| Two or more | 4 (5.1%) | |
| Unknown/declined (mostly Hispanic) | 13 (16.5%) | |
| Ethnicity | ||
| Hispanic | 22 (27.8%) | |
| Non-Hispanic | 56 (70.9%) | |
| Unknown/declined (multiple races) | 1 (1.3%) | |
| Education | ||
| Less than high school | 0 | |
| Some high school | 0 | |
| High school diploma/GED | 1 (1.3%) | |
| Some college | 12 (15.2%) | |
| College degree | 34 (43.0%) | |
| Graduate/professional school | 32 (40.5%) | |
| Unknown/declined | 0 | |
| Partner status | ||
| Single | 45 (57.0%) | |
| Partner/spouse | 32 (40.5%) | |
| Separated | 0 | |
| Widowed | 0 | |
| Divorced | 2 (2.5%) | |
| Living situation | ||
| Live alone | 9 (11.4%) | |
| Live with roommates | 11 (13.9%) | |
| Live with family | 39 (49.4%) | |
| Live with partner/significant other | 20 (25.3%) | |
| Number of days participated | 331.1 (76.2) | |
| Fitbit worn ≥10 h | ||
| Number of days | 228.4 (89.2) | |
| % of days | 67.9 (18.8) | |
| Morning EMA report completed | ||
| Number of days | 201.1 (87.7) | |
| % of days | 59.5 (20.8) | |
| Evening EMA report completed | ||
| Number of days | 213.7 (90.9) | |
| % of days | 63.4 (21.2) | |
Baseline demographic characteristics of 79 participants
| Characteristics | N (%) | Mean (SD) |
|---|---|---|
| Average age (years) | 31.9 (9.5) | |
| Height (cm) | 168.5 (8.5) | |
| Weight (kg) | 75.3 (17.4) | |
| BMI (kg/m2) | 26.4 (5.3) | |
| Gender | ||
| Men | 34 (43.0%) | |
| Women | 45 (57.0%) | |
| Race | ||
| American Indian/Alaska Native | 0 | |
| Asian | 15 (19.0%) | |
| Black/African American | 11 (13.9%) | |
| Native Hawaiian/Pacific Islander | 3 (3.8%) | |
| White | 33 (41.8%) | |
| Two or more | 4 (5.1%) | |
| Unknown/declined (mostly Hispanic) | 13 (16.5%) | |
| Ethnicity | ||
| Hispanic | 22 (27.8%) | |
| Non-Hispanic | 56 (70.9%) | |
| Unknown/declined (multiple races) | 1 (1.3%) | |
| Education | ||
| Less than high school | 0 | |
| Some high school | 0 | |
| High school diploma/GED | 1 (1.3%) | |
| Some college | 12 (15.2%) | |
| College degree | 34 (43.0%) | |
| Graduate/professional school | 32 (40.5%) | |
| Unknown/declined | 0 | |
| Partner status | ||
| Single | 45 (57.0%) | |
| Partner/spouse | 32 (40.5%) | |
| Separated | 0 | |
| Widowed | 0 | |
| Divorced | 2 (2.5%) | |
| Living situation | ||
| Live alone | 9 (11.4%) | |
| Live with roommates | 11 (13.9%) | |
| Live with family | 39 (49.4%) | |
| Live with partner/significant other | 20 (25.3%) | |
| Number of days participated | 331.1 (76.2) | |
| Fitbit worn ≥10 h | ||
| Number of days | 228.4 (89.2) | |
| % of days | 67.9 (18.8) | |
| Morning EMA report completed | ||
| Number of days | 201.1 (87.7) | |
| % of days | 59.5 (20.8) | |
| Evening EMA report completed | ||
| Number of days | 213.7 (90.9) | |
| % of days | 63.4 (21.2) | |
| Characteristics | N (%) | Mean (SD) |
|---|---|---|
| Average age (years) | 31.9 (9.5) | |
| Height (cm) | 168.5 (8.5) | |
| Weight (kg) | 75.3 (17.4) | |
| BMI (kg/m2) | 26.4 (5.3) | |
| Gender | ||
| Men | 34 (43.0%) | |
| Women | 45 (57.0%) | |
| Race | ||
| American Indian/Alaska Native | 0 | |
| Asian | 15 (19.0%) | |
| Black/African American | 11 (13.9%) | |
| Native Hawaiian/Pacific Islander | 3 (3.8%) | |
| White | 33 (41.8%) | |
| Two or more | 4 (5.1%) | |
| Unknown/declined (mostly Hispanic) | 13 (16.5%) | |
| Ethnicity | ||
| Hispanic | 22 (27.8%) | |
| Non-Hispanic | 56 (70.9%) | |
| Unknown/declined (multiple races) | 1 (1.3%) | |
| Education | ||
| Less than high school | 0 | |
| Some high school | 0 | |
| High school diploma/GED | 1 (1.3%) | |
| Some college | 12 (15.2%) | |
| College degree | 34 (43.0%) | |
| Graduate/professional school | 32 (40.5%) | |
| Unknown/declined | 0 | |
| Partner status | ||
| Single | 45 (57.0%) | |
| Partner/spouse | 32 (40.5%) | |
| Separated | 0 | |
| Widowed | 0 | |
| Divorced | 2 (2.5%) | |
| Living situation | ||
| Live alone | 9 (11.4%) | |
| Live with roommates | 11 (13.9%) | |
| Live with family | 39 (49.4%) | |
| Live with partner/significant other | 20 (25.3%) | |
| Number of days participated | 331.1 (76.2) | |
| Fitbit worn ≥10 h | ||
| Number of days | 228.4 (89.2) | |
| % of days | 67.9 (18.8) | |
| Morning EMA report completed | ||
| Number of days | 201.1 (87.7) | |
| % of days | 59.5 (20.8) | |
| Evening EMA report completed | ||
| Number of days | 213.7 (90.9) | |
| % of days | 63.4 (21.2) | |
The mean number of days with 10+ h of Fitbit wear time was 228 ± 89 days; the mean number of EMA morning reports of anticipated stress for that day was 201 ± 88 days; the mean number of evening EMA reports of anticipated stress for the next day and actual stress for that day were 214 ± 91 days. Overall, 76% of the participants had EMA data for at least 50% of the days, while 82% had 10 or more hours of Fitbit wear for at least 50% of the days. We examined whether EMA stress ratings predicted failure to wear the Fitbit device for 10 h on a given day, the threshold required for ascertainment of exercise by Fitbit for that day. We found that none of the stress rating variables, e.g., stress rating in the evening regarding stress for that day or stress anticipated for the next day; morning report of stress anticipated for that day—predicted availability, e.g., “missingness”—of this threshold of Fitbit wear. We also tested whether Fitbit wear at the required threshold predicted availability, e.g., missingness, of any of the morning or evening EMA stress ratings. Here too, we found no relationship. Thus, missing data—either Fitbit data at the required threshold or EMA stress reports—are unlikely to bias the estimate of the bi-directional associations between reported/anticipated stress and Fitbit-defined exercise.
We found only a moderate correlation between self-reported exercise and objectively assessed exercise. When examined at the subject level (i.e., between-subject correlation of proportion of days exercised according to Fitbit with proportion of days exercised based on self-reports), the correlation was r = 0.404; when examined within subject (e.g., person-days), the correlation between Fitbit-assessed exercise days and self-reported exercise days was r = 0.344. We also found a very strong correlation for anticipated stress, between the levels reported the preceding night for the next day and the levels reported in the morning for that day (r = 0.975); the pooled within-person correlation between these two self-reports was r = 0.607, indicating a strong tendency for the morning report of anticipated stress to be higher than usual when the prior evening report of anticipated stress had been higher than usual.
Relationship of Objectively Assessed Exercise to End-of-Day Stress Report
The mean end-of-day stress rating for that day across all participant days was 3.28 (±2.58), on the 0–10 scale. In the random coefficients regression analysis, the occurrence of a 30-min bout of objectively assessed moderate/vigorous physical activity was associated, on average, with a quarter-point lower end-of-day stress report (B = −0.25, p = 0.0002), and there were extensive individual differences in the size of this effect. Specifically, the estimated standard deviation of the Bs is 0.43 (p < 0.0001). This implies that approximately 95% of the Bs lie between −1.01 to +0.60. Thus, while for the average person a day with moderate/vigorous exercise was associated with a small end of day decrease in stress, there were individuals who showed a substantially stronger negative association and some who showed a sizeable positive association; Fig. 2 shows a forest plot of the estimated effect of a 30-min bout of exercise on end-of-day stress rating for the 69 participants who had 50 or more observations for their person-specific regression. Exercise was associated with a significantly lower end-of-day stress rating for several (n = 15), a significantly higher end-of-day stress rating for others (n = 2), and with no significant difference for the remainder (n = 62). In the sensitivity analysis, restricted to the first 180 days of monitoring, a 30-min period of exercise was again associated with a modestly lower end-of-day stress report (average B = −0.20, p = 0.001), and the estimated standard deviation of the Bs was 0.26 (p = 0.06).
Forest plot showing the effect of a continuous 30-min or greater bout of continuous moderate to vigorous physical activity on the evening stress report that day, for up to 365 days (effect size with 95% confidence interval for each participant; black vertical line represents “no association”; red vertical line shows average effect across participants) (color figure online)
Relationship of Daily Stress Report to Objectively Assessed Exercise
A 30-min bout of moderate/vigorous exercise occurred on 31.9% of the days in which the Fitbit was worn for at least 10 h. The random coefficients logistic regression analysis indicated that for the average person, a five-point increase in the anticipated stress level reported for the next day and recorded the previous night was associated with a 22% decrease (OR = 0.78, p = 0.009) in the odds of exercising that next day; the corresponding effect for a five-point increase in anticipated stress recorded in the morning of that day was a 20% (OR = 0.80, p = 0.06) decrease in the odds of exercising that day. Once again, there was a high degree of inter-individual variability in this relationship (p < 0.0001), with 95% of the participants predicted to have odds ratios between 0.25 and 2.44. Figure 3 shows the forest plot of the estimated effect for each participant with 50 or more observations, based on performing a separate logistic regression for each. According to this, stress was associated with a significantly decreased likelihood of exercise for several participants (n = 17), a significantly increased likelihood of exercise for only one participant, and was not significantly associated with exercise for the remainder of participants (n = 61). When the analysis was restricted to the first 180 days of monitoring, a five-point increase in anticipated stress reported the night before was associated with an average reduction in the odds of exercising that day of 17% (OR = 0.83, p = 0.05), with a similar degree of inter-individual variability in this relationship (p = 0.01).
![Forest plot showing the effect of morning report of anticipated stress for that day on the occurrence of a continuous 30-min or greater bout of continuous moderate to vigorous physical activity that day, for up to 365 days (odds ratio (OR) with 95% confidence interval for each participant; black vertical line represents “no association [OR = 1.0]”; red vertical line shows average effect across participants) (color figure online)](https://oup.silverchair-cdn.com/oup/backfile/Content_public/Journal/abm/51/6/10.1007_s12160-017-9902-4/3/m_12160_2017_9902_fig3_html.gif?Expires=1712687259&Signature=tdwt0jS~rE3IFWaX~fx~7cgLsbYsasHUk3qqs7cgK5vw8ySoVjcWjPiuCA~SUXVqxKYCbd6OVyReZIzA1cfLHtQiDrrenJU1uWFWGX7WOJJygBLz5cfi72mrLB89e1k-z3M-DbrXzvn~Kyg0JPyCt0qkQwDmcyCYPeHs95-as1F32q4U2lq~aHZ05ppLxuSJBec5z-KW-wUGfXxghCL1yaFJUVaqT7XOHtryekEXPJjX1QqA4A-W4fLAcN7UE-V4OJTgj9W-sdPeDPDQkBK0SUYd8Vgv1L7sFYqHmtSeRZlFRM66ujy3awNQdvXiYM8Q1RsxdglI1wBMHRACcWKU7Q__&Key-Pair-Id=APKAIE5G5CRDK6RD3PGA)
Forest plot showing the effect of morning report of anticipated stress for that day on the occurrence of a continuous 30-min or greater bout of continuous moderate to vigorous physical activity that day, for up to 365 days (odds ratio (OR) with 95% confidence interval for each participant; black vertical line represents “no association [OR = 1.0]”; red vertical line shows average effect across participants) (color figure online)
Examining the bi-directional effects concurrently revealed that (a) for n = 5 participants, there was only a significant negative effect of anticipated stress on the likelihood of exercise (anticipated stress decreasing the likelihood of exercise); (b) for n = 4 participants, there was only a significant effect of exercise on end-of-day stress rating, with 3 showing a negative effect (exercise associated with lower stress rating) and 1 showing a positive effect (exercise associated with higher stress rating); and (c) for n = 13, there was a significant bi-directional effect, with 12 showing that anticipated stress was associated with a reduced likelihood exercise and exercise was associated with a reduced end-of-day stress rating, while 1 person showing that anticipated stress was associated with an increased likelihood of exercise and exercise was associated with increased end-of-day stress rating.
Moderation of Effects
Analyses to test for moderation revealed no effect on the bi-directional association between stress and exercise for any of the additional variables—dispositional optimism, perceived chronic stress, sense of mastery, anxiety sensitivity, gender, or partner status.
Discussion
In what we believe to be the first study of its kind, we objectively monitored physical activity along with anticipated and perceived stress for 1 year among healthy young adults and determined the strength of the effects of each—stress and exercise—on the other for the average person and at the level of the individual. We found that the average effect of a 30-min bout of exercise on stress reported at the end of that day was negative and highly significant. Of note, we also found a large degree of variability among individuals in this relationship, with exercise being associated with lower subsequent stress ratings for several, a higher subsequent stress rating for others, and having no significant effect on subsequent stress rating for many. We also found a highly significant negative average effect of anticipated stress for a given day—whether reported the evening before or that morning—on the likelihood of exercise that day, again with a high degree of variability at the level of the individual. When we examined the individual person effects, some demonstrated only a significant effect of one—stress or exercise—on the other, and this effect was negative for some and positive for others. Furthermore, a substantial minority of participants demonstrated significant bi-directional effects, whereby anticipated stress was associated with a lower likelihood of exercise and exercise was associated with a lower end-of-day report of stress. Yet again, most demonstrated no significant effect in either direction. These findings highlight the limitations of the nomothetic or between-subject methods that predominate in research and the potential of idiographic—or within-subject—methods that assume the existence of unique, person-level associations and factors that may be highly relevant to among other things, the adoption and maintenance of health practices at the level of the individual. This study reveals the promise of idiographic research for understanding complex relationships among factors that are essential for the maintenance of health, and the promise as well for “precision medicine,” whereby disease prevention and treatment take into account individual variability in genes, environment, and lifestyle for each person [40].
While revealing the complexity of the stress-exercise relationship at the level of the individual, we found no effect modification by any of the additional measures—of chronic perceived stress, dispositional optimism, social support, sense of mastery, or anxiety sensitivity—that we administered. Prior research has found that dispositional optimism has an effect on stress and on the likelihood of maintaining an exercise program [41, 42], while social support has long been thought to buffer the effects of stress on health outcomes [43] and has previously been found to influence stress experiences and adoption of a healthy lifestyle [44–46]. It may be that we did not find effect moderation because we used only the baseline measure of these potentially moderating variables. An alternate interpretation is that the EMA stress ratings—whether at the end of the day reporting on the stressfulness of the day, the evening report of anticipated stress for the next day, or the morning report of anticipated stress for that day—implicitly incorporated, at the level of the individual, the moderating effect of these variables, e.g., that a person's dispositional optimism, their sense of being socially supported, their sense of mastery over their circumstances, and/or their sensitivity to anxiety experiences were incorporated into their anticipation of how stressful their day was going to be. While speculative, there is support for this hypothesis in experimental studies showing, for example, that the presence of a supportive other, or the self-recitation of personal affirmations of self-worth, reduces acute stress reactivity [43, 47]. Future research should address this important question, as these findings may again indicate the promise of idiographic approaches for discerning person-level factors in health behavior.
An additional and potentially important finding was the only moderate correlation between self-reported and objectively assessed bouts of exercise. This finding mirrors that for many health-relevant behaviors—such as medication adherence—and highlights the importance of objective assessment where possible [48–49]. Yet in the case here, it is possible that participants were doing something that they considered exercise—such as yoga or swimming—and that they self-reported this as exercise; yet, the nature of the activity was such that it was not captured by the objective Fitbit measurement. Overall, of those person-days for which there were both an end-of-date self-report and 10+ h of Fitbit wear time, the percentage of exercise days was slightly higher by self-report than by Fitbit (37.5 vs 35.5%). That said, when examined at the person level, slightly more than half of the participants had a higher percentage of days exercised by Fitbit than by self-report. Thus, future research should incorporate objective assessment methods, e.g., by including objective measures of physical exertion such as heart rate, that can account for physical activity that does not include a substantial degree of physical movement.
In contrast to this finding on physical activity, there was a high correlation on the self-report assessments of anticipated stress for a given day, e.g., between the assessment made the night before that day and the morning of that day. We believe this indicates that the expectation of stress may be highly stable, and this is mirrored in the stability of measures of chronic perceived stress as shown by others [34, 50].
While the findings of the current study provide new and important insights concerning the bi-directional nature of the stress-exercise relationship at the level of the individual, it is not without limitations. The study cohort was young and healthy. While this is an important age for the establishment of life-long health practices, the application of these findings to older populations—when the presence of chronic disease places greater importance on the maintenance of exercise regimens and on the potential effects of health-related stress—cannot be assumed. In addition, participants were English speaking and were not engaged in employment that limited completion of EMA reports. The sample size was also small, though the comprehensive data for the sample are noteworthy and, to our knowledge, unique. Furthermore, data were not available every day for all participants, and there was notable variability in the data available, for example, in the number of days with 10+ h of Fitbit wear. Yet, the great majority of participants provided data on at least 50% of the days, and we found that the availability of one data stream, e.g., Fitbit wear or stress rating, did not affect the availability of the other data stream. In addition, there are few studies that have undertaken the collection of data (e.g., EMA) daily for a full year such as we have undertaken here, making this study quite unique in the literature; those few studies that have undertaken similar—though less comprehensive—EMA data collection [51–52] also report comparable data completion. Nonetheless, we cannot assume that findings for days with data would be mirrored for days without data.
Conclusions
Regular exercise remains a key element in the effort to reduce risk for incident cardiovascular disease [53]; yet, efforts to enhance this have met mixed success. This study was designed to examine the bi-directional relationship between daily stress experience and exercise behavior in those who are intermittent exercisers. We believe that by examining this daily process within subjects, we have the potential to discover new and possibly potent influences that stress may have on behavior and that behavior may have on stress, for each person individually, thereby providing a vehicle for increasing personal regular exercise or decreasing personal daily stress. The expansion of this idiographic, N-of-1 approach to other areas of health risk and maintenance is consistent with the emerging precision medicine movement, and future research should be directed toward this effort.
