BSc Psychobiology
Bachelorproject
Physical activity and sleep ameliorate perceived stress
levels in healthy participants: an accelerometer-based
pilot study
By
Melissa Stoel
11653590
January 2020 -
June 2020
30 EC
Supervisor:
Assessor:
Examiner:
Physical activity and sleep ameliorate perceived stress
levels in healthy participants: an accelerometer-based
pilot study
Melissa R. Stoel
INTRODUCTION
Psychiatric disorders place a burden on a large number of individuals in the population. The most prevalent forms of psychiatric disorders are depressive disorders and anxiety disorders (Kessler, Petukhova, Sampson, Zaslavsky & Wittchen, 2012). In addition to the individual burden, these disorders place a problematic social and economic burden on society (Kessler & Greenberg, 2002, Simon, 2003). World Health Organization has predicted that by the year 2030, depression will be the most common cause of burden by disease (World Health Organization, 2008). Depression and anxiety disorders are not
only present in adulthood but are also a common problem during adolescence. One of the factors that contributes to the development of depression in adolescence is stress. Research has shown that stress and major life events in adulthood can lead to the development of depression (Pine, Cohen, Johnson & Brook, 2002). Furthermore, exposure to stress early in life can cause structural brain changes that contribute to the occurrence of depression in adolescence (Andersen & Teicher, 2008). The high prevalence of depression in adolescents is problematic, as nearly one third of all depressed adolescents will attempt to commit suicide and 3-4% of all adolescents succeed in Depression and anxiety are common psychiatric disorders that are responsible for a large
individual and socio-economic burden on society. Because physical activity levels and sleep quality are often lower in depressed subjects, potential treatment objectives for depression include
increasing physical activity and improving sleep quality. However, the exact way in which physical activity and sleep modulate the development or duration of depression remains largely unknown. An important factor in the interaction between depression, activity and sleep is perceived stress, which is influenced by both activity and sleep and can cause development of depression early in life. In this pilot study, a potential association between 21 accelerometer-derived physical activity, sleep parameters and daily perceived stress scores were assessed in 9 healthy participants with a mean age of 28.1 years. Analysis revealed a significant negative correlation between time spend in moderate activity and levels of perceived stress, moderate-to-vigorous physical activity and levels of perceived stress and onset of waking and levels of perceived stress. In addition, there were significant differences between measurement times (week, weekdays and weekend) for onset of sleep and onset of waking. Our results indicate that higher levels of physical activity and late determination of sleep can contribute to a decrease in perceived stress levels in healthy participants, suggesting a possible role for physical activity and stress in reducing depressive symptoms via a decrease in perceived stress. Taken together, these findings provide a first step into exploring possible predictive functions of accelerometer-derived physical activity and sleep
parameters in the prevention of depression.
their attempt (Hazell, 2011). This high rate of suicide attempts continues into adulthood, where the amount of suicide attempts amongst depressed patients is nearly three times higher than in the general population (Weissman et al., 1999). Depression and anxiety are highly comorbid. This is mainly caused by the overlap of symptoms and increased risk for the other disorder over time (Garber & Weersing, 2010). In addition to comorbidity, both depression and anxiety are continuous throughout life. In fact, 4% of all people suffering from anxiety disorders will remain ill for over 8 years (Keller et al., 1992). This causes depression and anxiety in adolescence to continue into adulthood psychopathology as well. The constant continuation of mood and anxiety disorders is mostly established by recurrence of depressive episodes and chronicity of anxiety (Ormel et al., 2015). Recurrence and relapse are a problem mainly in mood disorders. During this paper, relapse is defined as establishment of a new episode of depression or anxiety after remission of the disorder. Research has shown that adolescents with Major Depressive Disorder have a 33% chance of relapse within 4 years (Lewinsohn, Clarke, Seeley, & Rohde, 1994). Relapse numbers differ per study and per treatment provided. Research has found extremely high relapse numbers of 61%, even after 60 days of remission and pharmacotherapy (Emslie et al., 1997). These numbers emphasize the cruciality of effective relapse prevention therapies in adolescents to prevent psychopathology in adulthood.
A promising behavioral therapy for depression in adolescents is physical activity. Indeed, physical activity can ameliorate the symptoms of both depression and anxiety (Ströhle, 2009). According to a review by Lawlor & Hopker (2001), multiple studies indicate that regular exercise can reduce the symptoms of depression. Exercise is effective, even in treatment resistant patients with Major Depressive Disorder (Mota-Pereira et al., 2011). In addition, both acute and regular physical activity have a beneficial
effect on sleep (Kredlow, Capozzoli, Hearon, Calkins & Otto, 2015). This is especially relevant given that 90% of all depressive patients complain about poor sleep quality (Tsuno, Besset & Ritchie, 2005). Physical activity does not only improve sleep and ameliorate symptoms of depression and anxiety, but also influences the perception of psychological stress. As mentioned before, stress is a key contributor to the development of psychopathology in adolescents. Several different forms of stress, including work-related stress, work-family conflicts and stress of major life events cause a decrease in levels of physical activity (Stults-Kolehmainen & Sinha, 2014). Mainly high levels of perceived stress are associated with a decline in levels of physical activity (Norris, Carroll. & Cochrane, 1992). Higher levels of perceived stress are also connected to shorter sleep duration and lower sleep quality (Charles et al., 2011). Not only is exercise considered an effective treatment for psychopathology, research has proven that physical activity is also a predictor of treatment outcomes in depression (Hallgren et al., 2016). Physical activity has protective effects on the development and relapse of depression in both adults and adolescents (Schuch & Stubbs, 2019).
In order to determine levels of physical activity, a wrist-worn activity tracker can be used. Research has shown that wearable activity trackers (i.e. accelerometers) are a valuable method of measuring human physical activity (Wright, Hall Brown, Collier & Sandberg, 2017). Wearable activity trackers are widely available for consumers and currently produced by different brands. However, most wearables are designed mainly for consumer purposes and not for research purposes. The easy accessibility of consumer-based wearables makes for one disadvantage. The common use of such wearables for research causes this research to fail in noticing many physical activity parameters, as consumer wearables often only focus on broad measurements such as “calories burned” or “total step count”. Nowadays, there are accelerometers
on the market that are especially designed for research. Examples of such accelerometers are Axivity’s AX3, ActivInsights’ GeneActiv or the ActivInsights band. Factors like battery life, availability, facile data access and current article count have to be taken into consideration when deciding which wearable to use for research purposes (Henriksen et al., 2018). Research-oriented accelerometers are specialized in performing high-frequency measurements of acceleration and provide raw data which can be analyzed. This provides insights in more detailed physical activity and sleep parameters. Such research-oriented accelerometers have been used in earlier studies to investigate the role of physical activity in the development of depression (Vallance et al., 2011).
In order to deepen the understanding of depression and anxiety in adolescents, the StayFine project has been established. The StayFine project provides adolescents who have a history of depression or anxiety with a personalized intervention that aims to prevent the occurrence of relapse. The StayFine study is a longitudinal project with multiple assessments during which depressive and anxious symptoms are monitored. One of the monitored parameters includes perceived stress. In addition to screening for symptoms, one of the measurements that is used to determine relapse behavior is physical activity and sleeping behavior. The StayFine project assesses physical activity and sleep using the research-oriented accelerometer Axivity AX3. To examine the working mechanism and quality of the AX3 accelerometer, we designed a pilot study. This pilot aims not only to investigate the functionality of the accelerometer but also aims to expand current knowledge regarding the relationship between stress, physical activity and sleep. The earlier mentioned review of Stults-Kolehmainen & Sinha (2014) showed that stress is likely to cause a decrease in physical activity, although it should be mentioned that it can also cause an increase of activity in some participants. This is likely due to physical activity serving as a
coping mechanism for stress. Thus, the exact manner how stress, sleep and physical activity interact is still incompletely understood. Broadly investigating the more detailed sleep and physical activity parameters that can be measured using a research-oriented accelerometer may provide better insights in the relationship between these parameters and stress. During this pilot, healthy participants wore an accelerometer and were asked to fill out a questionnaire regarding perceived stress at the end of each day, for a consecutive period of two weeks. A questionnaire regarding demographics and coping mechanisms was conducted at the beginning of the pilot, in order to evaluate whether physical activity might increase due to coping with stress. Correlational analyses were used to assess in which way physical activity and sleep are associated with daily levels of perceived stress. We hypothesized that the Axivity AX3 accelerometer is a valid and objective method of measuring physical activity and sleep. Expectations are that the physical activity parameters such as mean acceleration, total inactivity and total moderate activity will be predictors of stress, since previous research has shown a connection between these parameters and stress (Norris, Carroll. & Cochrane, 1992). Sleep parameters such as sleep duration and sleep efficiency are also expected to be good predictors of perceived stress (Charles et al., 2011).
MATERIALS AND METHODS
Healthy participants were recruited at the University of Amsterdam. Current diagnosis of any DSM-V mental disorder was an exclusion criterium. In addition, participants were excluded if participants had any exercise-disabling injuries during the physical activity tracking period. All participants provided informed consent (appendix 1) and the pilot was approved by the Faculty Ethics Committee of the University of Amsterdam. Data was collected between April 3rd
and April 16th, 2020. During data collection,
were asked to continue their normal daily activities.
Questionnaires
Two questionnaires were used in order to determine baseline demographics, coping behavior and perceived stress. An intake questionnaire was conducted at the beginning of the two-week study period. This questionnaire concerned personal information such as gender, age, family relations and included a question regarding ways of coping with stress after a stressful day (appendix 2). The perceived stress scale (PSS) was conducted daily over the entire two-week study period. Participants received a link via messenger service WhatsApp to the online questionnaire every day at approximately 21:30 and were asked to fill out this questionnaire as soon as possible that same evening before the onset of sleep. The Dutch translation of the PSS that is used in the StayFine project was also used for this pilot study. This PSS was originally obtained from Cohen, Kamarck and Mermelstein (1983) and was adjusted to fit the StayFine project (appendix 3). Participants were asked to rate the amount of times they have felt a certain way during the day on a 5-point Likert scale, ranging from 0 (never) to 5 (fairly often). Both questionnaires were constructed and completed in University of Amsterdam’s online survey software Qualtrics and were filled out via a secured internet connection.
Accelerometer
The Axivity AX3 accelerometer (from hereon termed AX3) was used to determine levels of physical activity. Detailed information on all considered wearables is provided in appendix 4. All participants wore an AX3 on their non-dominant wrist. Axivity’s AX3 contains the ADXL345 triaxial accelerometer puck. The X, Y and Z axis of this accelerometer measure two types of acceleration, namely dynamic acceleration (as caused by motion) and static acceleration (as caused by gravity). The dynamic
range of sensitivity was set to ±8 gravitational units (g). The accelerometers were set to acquire data at a frequency of 50 Hz. AX3 data was downloaded using Axivity’s OmGui open-source software (version 1.0.0.43). The downloaded raw data was saved as .cwa file for further analysis. GGIR and outcome measures
All raw data files were analyzed using R Studio (version 1.2.5033; http://cran.r-project.org) with R package GGIR version 1.11-0 (Migueles, Rowlands, Huber, Sabia, & van Hees, 2019, Sabia et al., 2014, van Hees et al., 2013, van Hees et al., 2015). GGIR autocalibration was used to utilize local gravity and temperature as reference point (Van Hees et al., 2014). Euclidian Norm Minus One (ENMO) was used to express acceleration in mg, averaged over 5-second epochs. Negative ENMO values were automatically rounded up to zero by GGIR, to reduce possible error and bias. The full R-script that was used to run GGIR can be read in appendix 6.1.
Inactivity was qualified as acceleration below 30 milli gravitational units (mg). Light activity was qualified as acceleration between 30 mg and 100 mg. Activity between 100 mg and 400 mg was labeled as moderate activity and activity above 400 mg as vigorous activity. Both bouted and un-bouted activity was used for further analysis, bouted activity being specified as sustained activity for a period >10 minutes. Unbouted activity and total activity were compared to determine differences in statistical relevance between these parameters. GGIR sleep analysis was based on calculation of the sleep period time window (SPT-window), without use of a sleep diary. Research has shown that use of the heuristic algorithm behind the SPT-window provides a valid estimation of sleep parameters (Van Hees et al., 2018). All outcome values were calculated by using default GGIR settings. Detailed description of the GGIR values that were used for further analysis can be found in
Statistical analysis
Statistical analysis was performed using R (version 1.2.5033; http://cran.r-project.org). GGIR values were transformed to fit normal distribution using R-package BestNormalize. Depending on the GGIR value, this package applied a square-root transformation, arcsinh transformation, BoxCox transformation (Box & Cox, 1964), Yeo-Johnson transformation (Yeo & Johnson, 2000) or the ordered quantile technique (Van der Waerden, 1952) to normalize this value. Repeated measures correlation between daily GGIR outcome values and PSS scores for each participant was assessed using R-package rmcorr (Bakdash & Marusich, 2017). Assumptions of equal variance and normal distribution of residuals were tested using a Levene’s test and a Shapiro-Wilk Normality test. To determine differences in GGIR values throughout the week, during weekdays and during weekend days, a one-way repeated measures ANOVA was used. Assumptions of normality and sphericity were tested using a Shapiro-Wilk Normality test and Mauchly’s test. If needed, the Greenhouse-Geisser sphericity correction was applied to values violating the assumption of sphericity. A post-hoc Pairwise T-test with Bonferroni correction was conducted to determine the time-points between which GGIR values differed. The R-scripts that were used to perform statistical analyses can be read in appendix 6.2 and 6.3.
RESULTS
Nine healthy participants (5 male, 4 female) with a mean age of 28.1 ±8.3 years were included in this pilot. All participants filled out the daily perceived stress scale accordingly. PSS scores were calculated based on the PSS point-system and held a mean score of 8.7 ±5.3 points. Only one participant indicated that they used some for as physical activity to cope with stress. Other participants preferred relaxation after a stressful day. The AX3 .cwa files were successfully downloaded using OmGui software and were
processed in R using R-package GGIR. There were no disfunctions in any of the accelerometers. One participant indicated 2 hours of non-wear time, which matched the non-wear time generated by GGIR. There was one unexpected outlier detected in the acceleration between 1 AM and 6 AM. This outlier proved to be valid after consulting with the concerned participant.
Repeated measures correlation
A separate file was created linking the daily PSS scores for each participants to the associated GGIR value. Assumptions of repeated measures correlation were tested for twenty different GGIR outcome values (appendix 5). Assumptions of equal variance were not violated by any of the GGIR values. However, the assumptions of normal distribution of the residuals were violated for all GGIR values. R-package BestNormalize was used to transform each GGIR value to better fit the normal distribution. Different transformations were used for each GGIR value, in accordance to the best fit of each transformation. Despite transforming the GGIR values, the assumption of normal distribution of residuals remained violated. Separate repeated measures correlations for each GGIR value showed a significant negative correlation between daily PSS score and total duration spend in moderate activity (figure 1A). There was also a significant negative correlation between PSS score and time spend in moderate-to-vigorous activity (MVPA) (figure 1B). The participant that indicated to prefer physical activity as a mechanism of coping with stress, showed an increase of stress was indeed associated with an increase of MVPA. However, this was also the case for several participants who indicated relaxation was their way of coping with stress. In addition, there was a significant negative correlation between daily PSS score and onset of waking time (figure 1C). Correlations between other GGIR values and daily PSS scores were not significant (appendix 7.1).
Figure 1. Repeated measures correlation between daily perceived stress score and GGIR outcome values. (A) Levene’s test
of equal variance was not violated for the total duration spend in moderate activity (F22, 103 = 0.57, P = 0.94). Shapiro-Wilk Normality test showed that the residuals were not normally distributed (W = 1, P = 0.0007). Repeated measures correlation showed a significant correlation between PSS score and total duration spend in moderate activity (r116 = -0.262, P = 0.00415). (B) Levene’s test was not violated for MVPA (F22, 103 = 0.62, P = 0.90). The Shapiro-Wilk Normality test showed that residuals were not normally distributed (W = 0.96, P = 0.002). Repeated measures correlation showed a significant correlation between PSS score and MVPA (r116 = -0.249, P = 0.0065). (C) Levene’s test showed that assumptions of equal variance were not violated for onset of waking time (F22, 103 = 0.87, P = 0.64). According to Shapiro-Wilk Normality test, residuals were not normally distributed (W = 1, P = 0.001). Repeated measures correlation showed a significant correlation between PSS score and onset of waking time (r116 = -0.264, P = 0.00386).
A B
Both unbouted physical activity and total physical activity (bouted + unbouted) were included in the GGIR outcomes. Bouted activity was defined as a sustained period of that activity lasting longer than 10 minutes. Correlations between both total physical activity parameters and unbouted parameters were plotted next to
each other to assess differences and to determine dissimilarities in statistical relevance.
Correlations of minutes spend in activity with PSS scores were stronger for total physical activity parameters than for parameters that only included unbouted physical activity (figure 2).
Figure 2. Dissimilarities between total physical activity and unbouted physical activity in correlation to perceived stress.
Repeated measures correlation coefficients were higher for total inactivity (r116 = 0.175, P = 0.0584) than for unbouted inactivity (r116 = -0.0899, P = 0.333). Correlation coefficients were higher for total light activity (r116 = -0.144, P = 0.119) than for unbouted light activity (r116 = -0.114, P = 0.221). Correlation coefficients were also higher for total moderate activity (r116 = -0.262, P = 0.00415) than for unbouted moderate activity (r116 = -0.0768, P = 0.408).
Repeated measures ANOVA
Differences in PSS score and 12 GGIR outcome values between different measurement times (weeks, weekends and weekdays) were assessed using a one-way repeated measures ANOVA. Six out of nine participants (66.6%) indicated that they worked from home on at least one weekend day. Assumptions of normal distribution were not violated for the PSS score and the majority of the GGIR values. The data showed significant
differences between measurement times for onset of sleep and onset of waking. Post-hoc testing proved that these significant differences were present between all measurement times for both onset of sleep and onset of waking (figure 3). There were no significant differences between measurement times for the PSS score (figure 4) or for any other GGIR outcome values (appendix 7.2).
Figure 3. One-way repeated measures ANOVA between different measurement times for onset of sleep and onset of waking. (A) The Shapiro-Wilk Normality test showed that the assumption of normal distribution was not violated for onset of
sleep measured over weeks (W = 0.874, P = 0.137), over weekdays (W = 0.855, P = 0.084) or over weekends (W = 0.904, P = 0.273). Repeated measures ANOVA showed that there were significant differences between measurement times for onset of sleep (F2, 16 = 10.4, P = 0.001). Post-hoc pairwise T-tests with Bonferroni correction showed that there was a significant difference in onset of sleep between weeks and weekdays (T8 = 3.23, P = 0.036). There also was a significant difference in onset of sleep between weeks and weekends (T8 = -3.23, P = 0.036) and between weekends and weekdays (T8 = -3.23, P = 0.036). (B) The Shapiro-Wilk Normality test proved that the weeks of onset of waking followed a normal distribution (W = 0.962, P = 0.818). The Shapiro-Wilk Normality test also showed a normal distribution for weekdays (W = 0.956, P = 0.757) and for weekends (W = 0.956, P = 0.75) of onset of waking. The repeated measures ANOVA proved there were significant differences between measurement times for onset of waking (F2, 16 = 19.4, P = 0.00006). Post-hoc Pairwise T-tests with Bonferroni correction showed there was a significant difference in onset of sleep between week and weekdays (T8 = 4.4, P = 0.007). There also was a significant difference in onset of waking between weeks and weekends (T8 = -4.4, P = 0.007) and between weekends and weekdays (T8 = -4.4, P = 0.007).
Figure 4. One-way repeated measures ANOVA between different measurement times for PSS score. The Shapiro-Wilk
Normality test showed that the assumption of normal distribution was not violated for PSS score measured over weeks (W = 0.959, P = 0.789). Assumptions of normal distribution were also met for PSS score measured over weekdays (W = 0.973, P = 0.919) and over weekends (W = 0.982, P = 0.973). The repeated measures ANOVA showed no significant differences between weeks, weekdays and weekends of measuring PSS score (F2, 16 = 0.99, P = 0.393).
Of all GGIR outcome values, extreme outliers were found in L5 onset, M5 acceleration, 1AM to 6AM acceleration and duration of sustained inactivity bouts during the day (DURSIBD). In addition to these extreme outliers, assumptions of normal distribution were violated for these GGIR values. R-package BestNormalize was used to transform these GGIR values. An arcsinh transformation was applied to the L5 onset and M5 acceleration values, after which the normal distribution improved. However, there were still no significant differences between weeks, weekends and weekdays for these GGIR values (appendix 7.2, figure 10). Performing the arcsinh transformation on 1AM to 6AM acceleration and DURSIBD values did not improve normal distribution for these values. Thus, non-parametric test was used to determine significant differences for 1AM to 6AM acceleration and DURSIBD. This showed no significant differences between measurement times for 1AM to 6AM acceleration and DURSIBD (appendix 7.2, figure 11A & 12C).
DISCUSSION
Physical activity and sleep are promising treatment-targets and predictors of depression. Both physical activity and sleep also have beneficial effects on stress. This pilot study demonstrates that more time spend in moderate activity is associated with a lower score on a perceived stress scale. In addition, higher levels of MVPA are associated with a lower PSS score. Therefore, it can be assumed that moderate physical activity contributes to lower levels of perceived stress. Furthermore, later onset of waking is associated with a lower PSS score, indicating that sleeping in also contributes to lower levels of stress during the day. Significant differences between measurement times of outcome values were only found for onset of sleep and onset of waking, showing that participants tend to fall asleep later and wake up later during weekends compared to weekdays. Data also showed that correlations with perceived stress
were stronger for total time spend in activity than for unbouted time only, suggesting that total time spend in activity is a more valid measure than unbouted time spend in activity. To conclude, these data prove that both moderate physical activity and late onset of waking can contribute to lower levels of perceived stress in healthy participants throughout the day.
Physical activity and sleep have been linked to stress in many previous studies. However, this pilot study aimed to incorporate more detailed physical activity and sleep parameters in order to investigate the association of these parameters with perceived stress. Earlier research has shown that a decline in physical activity is associated with higher levels of perceived stress (Norris, Carroll. & Cochrane, 1992). The finding of a negative correlation between PSS score and both total time spend in moderate activity and MVPA is coherent with this earlier research. On the contrary, no positive correlation between perceived stress score and time spend in inactivity was shown in this pilot (appendix 7, figure 8A). Even though this would be expected based on previous research and the finding that moderate activity is associated with lower levels of stress. Similar to time spend in total inactivity, several other GGIR outcome values such as DURSIBD and duration of the SPT-window appear to have some correlation with stress based on the regression lines (appendix 7). This would be expected as previous research has shown that both sustained inactivity and short sleep duration can contribute to higher levels of perceived stress (Charles et al., 2011). Nevertheless, these correlations were not significant. A general limitation which can be accountable for this lack of significance is the small sample size of this pilot study. It is expected that a larger number of participants would provide enough data for these correlations to be significant. The significant negative correlation found between wake onset and PSS score somewhat coheres with previous research. Later onset of waking is associated with a lower PSS
score (figure 1C). Earlier research showed that longer sleep duration is associated with less perceived stress (Charles et al., 2011). However, no significant correlation between PSS score and sleep duration was shown. Therefore, late onset of waking may influence stress in a way that does not alter sleep duration. It can be assumed that late onset of waking is accompanied by not having to set an alarm and waking up in a more natural way. This might contribute to less perceived stress throughout the day and could explain the negative correlation found in this pilot’s data. The finding of significant differences between measurement times of onset of sleep and waking was not unforeseen. During the weekends, participants tended to go to bed later and stay in bed longer in the morning. This is likely as most participants did not have any work-related obligations in the weekend and were able to adjust their sleep schedule to their own wishes without the need for an early alarm.
The principal limitation of this pilot study comes from the current world pandemic caused by the Corona virus disease 2019 (COVID-19). COVID-19 led to a government-directed intelligent lockdown in which schools, universities, shops and gyms were closed. People were encouraged to work from home and leave the house for necessary purposes only. This intelligent lockdown caused a decrease in this pilot’s sample size as researchers were no longer allowed to deliver AX3’s to participants in person. Furthermore, the lockdown drastically influenced physical activity, stress and perhaps even sleep. Upon completion of the pilot study, 7 out of 9 participants indicated that the situation regarding the Corona virus had influenced their stress levels. An increase of stress may very well be a result of social isolation, as research has shown that solitude can lead to increased levels of stress (Campagne, 2019). On the other hand, the intelligent lockdown may have also had a beneficial effect on stress. For example, exams were postponed, working hours were shortened and social obligations were no longer present.
This may have caused a decrease in perceived stress for some participants. Furthermore, 7 out of 9 participants declared that they had lower levels of physical activity than usual due to the Corona virus. Despite having more leisure time due to the lockdown, participants may have preferred staying at home and sleeping in over exercising.
The fact that this pilot was conducted during a lockdown can also explain the results which were found regarding different measurement times of perceived stress and GGIR outcome values. It was expected that perceived stress would differ between weekdays and weekends, as most people encounter less work-related stressful situations during the weekend than during weekdays. However, due to Corona virus participants were forced to work from home, which resulted in frequent working during weekends as well. This same situation occurred for GGIR values regarding physical activity. During a normal situation participants would engage in some form of travelling from home to work, which was not possible during the lockdown period. This caused a decrease in physical activity levels throughout the weekdays. The division between weekdays and weekends was beclouded by the intellectual lockdown, which led to little differences between different measurement times of GGIR values and perceived stress score. Another restraint of this pilot is the fact that the assumption of normal distribution of residuals was violated for all GGIR values in the repeated measures correlation analysis. This violation could not be improved by any data transformations and is therefore considered a limitation in data-analysis. According to Bakdash & Marusich (2017), an abnormal distribution of residuals does not prohibit the use of repeated measures correlation. However, it can result in a biased repeated measures correlation model and may cause misleading results (Bakdash & Marusich, 2017). This has to be taken into account when interpreting the results of the repeated measures correlation.
One final limitation of this pilot concerns the use of the Dutch translation of the perceived stress scale adjusted for StayFine. According to the original PSS from Cohen, Kamarck and Mermelstein (1983), the Likert scale of question 7 has to be reversed when calculating a PSS score for participants. However, in the Dutch translation of the PSS this question was formulated differently in such way that reversing the Likert scale would have caused misinterpretation of the results. Therefore, it was decided to not revert the Likert scale for question 7 when calculating PSS scores with the Dutch translation of the perceived stress scale in this pilot study. In addition, one participant indicated that they found question 8 of the PSS difficult to interpret due to vague formulation. This is something that has to be looked into and can be adjusted before the start of the StayFine study in order to prevent further difficulties.
When compared to similar research in this field, a clear asset of this pilot study is the usage of the research-oriented Axivity AX3 accelerometer. Whereas most research focusses on broad physical activity parameters such as step count per day, the use of the AX3 provided the possibility to investigate more detailed parameters such as L5/M5 onset and acceleration, duration of the SPT-window and sleep efficiency. Even though there were no significant correlations found between any of these detailed GGIR outcome values and perceived stress, it remains a fact that using the AX3 provides an advantages over the usage of consumer-based wearables. Rather than analyzing data based on averages per day, this pilot included raw data measured at a frequency of 50 Hz. This drastically
increases the amount of data points and therefore increases reliability and validity of this study.
Altogether, this pilot’s findings show that there is a clear association between physical activity, sleep and perceived stress. It has been proven that high levels of stress play a leading role in the development of depression during adolescence. Ameliorating these levels of perceived stress via physical activity or exercise and improvement of sleep may contribute to effective treatment and perhaps even prevention of depression. This can help lift some of the problematic burdens that depression causes off of our society. However, this pilot study requires further research and validation in order to expand current knowledge regarding the associations between physical activity, sleep and depression. For example, future research may focus on assessing similarities between accelerometer measurements in stressed humans and patterns of physical activity and sleep in stressed rodents. It is highly likely that such translational research contributes to a better understanding of the interaction between stress, physical activity and sleep. Furthermore, future research could utilize the AX3 accelerometer to determine associations between stress and physical activity in patients suffering from depression or anxiety. When future research is conducted, physical activity and sleep could be used as predictors for the development of a depressive episode and for the prediction of relapse in depressed patients. The StayFine project will aim to further investigate these predictive values of physical activity and sleep on depression.
REFERENCES
1. Andersen, S.L., & Teicher, M.H. (2008). Stress, sensitive periods and maturational events in adolescent depression. Trends in neurosciences, 31(4), 183-191. DOI: 10.1016/j.tins.2008.01.004.
2. Bakdash, J.Z. & Marusich, L.R. (2017). Repeated Measures Correlation. Frontiers in Psychology, 8, 456. DOI: 10.3389/fpsyg.2017.00456.
3. Box, G.E., & Cox, D.R. (1964). An analysis of transformations. Journal of the Royal Statistical Society: Series B (Methodological), 26(2), 211-243. DOI: 10.1111/j.2517-6161.1964.tb00553.x.
4. Campagne, D. M. (2019). Stress and perceived social isolation (loneliness). Archives of gerontology and geriatrics. DOI: 10.1016/j.archger.2019.02.007.
5. Charles, L.E., Slaven, J.E., Mnatsakanova, A., Ma, C., Violanti, J.M., Fekedulegn, D., ... & Burchfiel, C.M. (2011). Association of perceived stress with sleep duration and sleep quality in police
officers. International journal of emergency mental health, 13(4), 229.
6. Cohen, S., Kamarck, T., & Mermelstein, R. (1983). A global measure of perceived stress. Journal of Health and Social Behavior, 24(4), 386-396. DOI: 10.2307/2136404. 7. Emslie, G.J., Rush, A.J., Weinberg, W.A., Gullion, C.M.,
Rintelmann, B.S. & Hughes, C.W. (1997). Recurrence of Major Depressive Disorder in Hospitalized Children and Adolescents. Journal of the American Academy of Child & Adolescent Psychiatry, 36(6), 785-792. DOI: 10.1097/00004583-199706000-00015.
8. Hallgren, M., Nakitanda, O. A., Ekblom, Ö., Herring, M. P., Owen, N., Dunstan, D., ... & Forsell, Y. (2016). Habitual physical activity levels predict treatment outcomes in depressed adults: A prospective cohort study. Preventive medicine, 88, 53-58. DOI: 10.1016/j.ypmed.2016.03.021.
9. Hazell P. (2011). Depression in children and adolescents. BMJ clinical evidence, 2011, 1008.
10. Henriksen, A., Mikalsen, M. H., Woldaregay, A. Z., Muzny, M., Hartvigsen, G., Hopstock, L. A., & Grimsgaard, S. (2018). Using fitness trackers and smartwatches to measure physical activity in research: analysis of consumer wrist-worn wearables. Journal of medical Internet research, 20(3), e110. DOI: 10.2196/jmir.9157.
11. Garber, J., & Weersing, V.R. (2010). Comorbidity of anxiety and depression in youth: Implications for treatment and prevention. Clinical Psychology: Science and Practice, 17(4), 293-306. DOI: 10.1111/j.1468-2850.2010.01221.x.
12. Keller, M.B., Lavori, P.W., Wunder, J., Beardslee, W.R., Schwartz, C.E., & Roth, J. (1992). Chronic course of anxiety disorders in children and adolescents. Journal of the American Academy of Child & Adolescent Psychiatry, 31(4), 595-599. DOI: 10.1097/00004583-199207000-00003.
13. Kessler, R.C., & Greenberg, P.E. (2002). The economic burden of anxiety and stress disorders. Neuropsychopharmacology: The fifth generation of progress, 67, 982-92. DOI: 10.4088/jcp.v60n0702. 14. Kessler, R.C., Petukhova, M., Sampson, N.A., Zaslavsky,
A.M., & Wittchen, H.U. (2012). Twelve-month and lifetime prevalence and lifetime morbid risk of anxiety and mood disorders in the United States. International journal of methods in psychiatric research, 21(3), 169-184. DOI: 10.1002/mpr.1359.
15. Kredlow, M.A., Capozzoli, M.C., Hearon, B.A., Calkins, A.W., & Otto, M.W. (2015). The effects of physical activity on sleep: a meta-analytic review. Journal of behavioral medicine, 38(3), 427-449. DOI: 10.1007/s10865-015-9617-6.
16. Lawlor, D. A., & Hopker, S. W. (2001). The effectiveness of exercise as an intervention in the management of depression: systematic review and meta-regression analysis of randomised controlled trials. Bmj, 322(7289), 763. DOI:10.1136/bmj.322.7289.763.
17. Lewinsohn, P.M., Clarke, G.N., Seeley, J.R., & Rohde, P. (1994). Major depression in community adolescents: age at onset, episode duration, and time to recurrence. Journal of the American Academy of Child & Adolescent Psychiatry, 33(6), 809-818. DOI: 10.1097/00004583-199407000-00006.
18. Migueles, J.H., Rowlands, AV., Huber, F., Sabia, S., & van Hees, V.T. (2019). GGIR: a research community– driven open source R package for generating physical activity and sleep outcomes from multi-day raw accelerometer data. Journal for the Measurement of
Physical Behaviour, 2(3), 188-196. DOI: 10.1123/jmpb.2018-0063.
19. Mota-Pereira, J., Silverio, J., Carvalho, S., Ribeiro, J. C., Fonte, D., & Ramos, J. (2011). Moderate exercise improves depression parameters in treatment-resistant patients with major depressive disorder. Journal of psychiatric research, 45(8), 1005-1011. DOI: 10.1016/j.jpsychires.2011.02.005.
20. Norris, R., Carroll, D., & Cochrane, R. (1992). The effects of physical activity and exercise training on psychological stress and well-being in an adolescent population. Journal of psychosomatic research, 36(1), 55-65. DOI: 10.1016/0022-3999(92)90114-h.
21. Ormel, J., Raven, D., Van Oort, F., Hartman, C., Reijeneveld, S., Veenstra, R., … Oldehinkel, A. (2015). Mental Health in Dutch adolescents: A TRIALS report on prevalence, severity, age of onset, continuity and co-morbidity of DSM disorders. Psychological medicine, 45(2), 345-360. DOI: 10.1017/S0033291714001469. 22. Pine, D.S., Cohen, P., Johnson, J.G., & Brook, J.S. (2002).
Adolescent life events as predictors of adult depression. Journal of affective disorders, 68(1), 49–57. DOI: 10.1016/s0165-0327(00)00331-1.
23. Rock, P., Goodwin, G., Harmer, C., & Wulff, K. (2014). Daily rest-activity patterns in the bipolar phenotype: a controlled actigraphy study. Chronobiology international, 31(2), 290-296. DOI: 10.3109/07420528.2013.843542. 24. Sabia, S., van Hees, V.T., Shipley, M.J., Trenell, M.I.,
Hagger-Johnson, G., Elbaz, A., ... & Singh-Manoux, A. (2014). Association between questionnaire-and accelerometer-assessed physical activity: the role of sociodemographic factors. American journal of epidemiology, 179(6), 781-790. DOI: 10.1093/aje/kwt330. 25. Schuch, F. B., & Stubbs, B. (2019). The role of exercise in
preventing and treating depression. Current sports medicine reports, 18(8), 299-304. DOI: 10.1249/JSR.0000000000000620.
26. Simon, G. E. (2003). Social and economic burden of mood disorders. Biological psychiatry, 54(3), 208-215. DOI: 10.1016/S0006-3223(03)00420-7.
27. Ströhle, A. (2009). Physical activity, exercise, depression and anxiety disorders. Journal of neural transmission, 116(6), 777. DOI: 10.1007/s00702-008-0092-x.
28. Stults-Kolehmainen, M.A., & Sinha, R. (2014). The effects of stress on physical activity and exercise. Sports medicine, 44(1), 81-121. DOI: 10.1007/s40279-013-0090-5.
29. Tsuno, N., Besset, A., & Ritchie, K. (2005). Sleep and depression. The Journal of clinical psychiatry. DOI: 10.4088/JCP.v66n1008.
30. Vallance, J.K., Winkler, E.A.H., Gardiner, P.A., Healy, G.N., Lynch, B.M., Owen, N. Associations of objectively-assessed physical activity and sedentary time with depression: NHANES (2005-2006). Preventive Medicine. 2011;53(4-5):284-288. DOI: 10.1016/j.ypmed.2011.07.013. 31. Van der Waerden, B.L. (1952). Order tests for the two-sample problem and their power. In Indagationes Mathematicae (Proceedings) (Vol. 55, pp. 453-458). North-Holland.
32. Van Hees, V.T., Fang, Z., Langford, J., Assah, F., Mohammad, A., da Silva, I.C., ... & Brage, S. (2014). Autocalibration of accelerometer data for free-living physical activity assessment using local gravity and temperature: an evaluation on four continents. Journal of Applied Physiology, 117(7), 738-744. DOI: 10.1152/japplphysiol.00421.2014.
33. Van Hees, V.T., Gorzelniak, L., Leon, E.C.D., Eder, M., Pias, M., Taherian, S., ... & Brage, S. (2013). Separating movement and gravity components in an acceleration signal and implications for the assessment of human daily physical activity. PloS one, 8(4). DOI: 10.1371/journal.pone.0061691.
34. Van Hees, V.T., Sabia, S., Anderson, K.N., Denton, S.J., Oliver, J., Catt, M., ... & Singh-Manoux, A. (2015). A
novel, open access method to assess sleep duration using a wrist-worn accelerometer. PloS one, 10(11). DOI: 10.1371/journal.pone.0142533
35. van Hees, V.T., Sabia, S., Jones, S.E., Wood, A.R., Anderson, K.N., Kivimäki, M., ... & Mazzotti, D.R. (2018). Estimating sleep parameters using an accelerometer without sleep diary. Scientific reports, 8(1), 1-11. DOI: 10.1038/s41598-018-31266-z.
36. Weissman, M.M., Wolk, S., Wickramaratne, P., Goldstein, R.B., Adams, P., Greenwald, S., ... & Steinberg, D. (1999). Children with prepubertal-onset major depressive disorder and anxiety grown up. Archives of general psychiatry, 56(9), 794-801. DOI: 10.1001/archpsyc.56.9.794.
37. World Health Organization. (2008). The global burden of disease: 2004 update. World Health Organization. 38. Wright, S.P., Hall Brown, T.S., Collier, S.R., & Sandberg,
K. (2017). How consumer physical activity monitors could transform human physiology research. American Journal of Physiology-Regulatory, Integrative and Comparative Physiology, 312(3), R358-R367. DOI: 10.1152/ajpregu.00349.2016.
39. Yeo, I. K., & Johnson, R. A. (2000). A new family of power transformations to improve normality or symmetry. Biometrika, 87(4), 954-959. DOI: 10.1093/biomet/87.4.954.
APPENDICES
1. Informed consent Informed consent proefpersoon
Betreft: onderzoek in het kader van fysieke activiteit en stress in gezonde proefpersonen. Informatie met betrekking tot het onderzoek:
U neemt deel aan een onderzoek over de invloed van stress op dagelijkse fysieke activiteit. U ontvangt een AX3 accelerometer om voor een periode van 14 dagen te dragen. De accelerometer moet ten alle tijden (zowel gedurende dag als nacht) gedragen worden. Tevens ontvangt u een korte vragenlijst aan het begin van de studie, en een dagelijkse vragenlijst gedurende de 14 dagen van het onderzoek. Alle verzamelde data zal anoniem blijven.
Ik verklaar hierbij op duidelijke wijze te zijn ingelicht over de aard en methode van het onderzoek. Ik stem geheel vrijwillig in met deelname aan dit onderzoek en ik besef dat ik op elk moment mag stoppen met deelname. Indien mijn onderzoeksresultaten gebruikt zullen worden in wetenschappelijke publicaties, onderzoeksverslagen, presentaties dan wel op een andere manier openbaar worden gemaakt, zal dit volledig geanonimiseerd gebeuren. Mijn persoonsgegevens zullen niet door derden worden ingezien zonder mijn uitdrukkelijke toestemming en zullen vernietigd worden na afloop van het onderzoek. Als ik nog verdere informatie over het onderzoek zou willen krijgen of eventuele klachten heb, nu of in de toekomst, kan ik me wenden tot:
M. (Melissa) Stoel
E-mail adres: m.r.stoel@amsterdamumc.nl T: +31 (0) 612064335
Handtekening:
4. Wearable comparison
Table 1. Comparison of all considered waterproof wearables.
Y = Yes, N = No, NAV = Not available.
Wearable FitBit
Inspire HR FitBit Versa AX3 GENEActiv Activinsights Band
Batterylife 5 days 4 days 14 days (at
100 Hz) 7 days (at 100 Hz) 365 days
Raw data N N Y Y N
Hz range NAV NAV 12.5-3200 10-100 NAV
HR Y Y N N Y
HRV Y N N N N
Forced
notification N N N N N
Design Band Watch Band Watch Band
Price €89.00 €159.00 €224.00 €192.00 €234.00
Wearable Misfit Vapor Xiaomi Mi
Band Garmin Vivosport HR Polar A370 Polar M430
Batterylife 2 days 20 days 7 days 4 days 5 days
Raw data N N N N N
Hz range NAV NAV NAV NAV NAV
HR Y Y Y Y Y
HRV N N N N N
Forced
notification N N N Y N
Design Watch Band Band Band Watch
5. GGIR outcome values Table 2. GGIR - Outcome values
Variable GGIR code Definition
L5 onset L5hr_ENMO_mg_0-24hr Starting time of the five least active hours from midnight on, on a scale from 0 to 24.
L5 acceleration L5_ENMO_mg_0-24hr Average acceleration over the five least active hours.
M5 onset M5hr_ENMO_mg_0-24hr Starting time of the five most active hours from midnight on, on a scale from 0 to 24.
M5 acceleration M5_ENMO_mg_0-24hr Average acceleration over the five most active hours.
1am-6am acceleration mean_ENMO_mg_1-6am Average acceleration between 1 a.m. and 6 a.m.
Mean acceleration mean_ENMO_mg_0-24hr Average acceleration over the entire day. UBIN duration dur_day_OIN_min Unbouted time spent in
inactivity.
UBLIG duration dur_day_LIG_min Unbouted time spent in light activity.
UBMOD duration dur_day_MOD_min Unbouted time spent in moderate activity. UBVIG duration dur_day_VIG_min Unbouted time spent in
vigorous activity. TIN duration dur_TINday_min Unbouted and bouted
time spent in inactivity. TLIG duration dur_TLIGday_min Unbouted and bouted
time spent in light activity.
TMOD duration dur_TMODday_min Unbouted and bouted time spent in moderate activity.
TVIG duration dur_TVIGday_min Unbouted and bouted time spent in vigorous activity.
MVPA
MVPA_E5S_B10M80%_T100_ENMO_0-24hr MVPA calculated based on 5 second epoch setting bout duration 10 minute and inclusion criterion of more than 80 percent. Sleep onset acc_onset Detected onset of sleep
the midnight of the previous night.
Wake onset acc_wake Detected onset of waking expressed in hours since the midnight of the previous night.
SPT duration acc_Sptduration Difference between the sleep onset and waking onset.
Sleep duration acc_SleepDurationInSpt Total sleep duration based on the nocturnal sustained inactivity bouts within the SPT-window. SIBD duration acc_dur_sibd Duration of accumulated
sustained inactivity bouts during the day.
Sleep efficiency acc_eff Sleep efficiency within the SPT-window calculated as the ratio between sleep duration and SPT duration.
6. R scripts 6.1 GGIR analysis
#Install the GGIR package.
install.packages("GGIR", dependencies = TRUE) library(GGIR)
#Run GGIR.
g.shell.GGIR(mode=c(1,2,3,4,5),datadir="~/Documents/Documents/UvA/Ba chelorstage/Statistiek/AX3Data/Subject1/",outputdir="~/Documents/Doc uments/UvA/Bachelorstage/Statistiek/AX3Data/",do.report=c(2,4,5), strategy = 1, hrs.del.start = 0,hrs.del.end = 0, maxdur = 16, overwrite = TRUE, desiredtz = "Europe/Amsterdam", includedaycrit = 16, qwindow=c(0,24), mvpathreshold =c(100), bout.metric = 4,
excludefirstlast = FALSE, includenightcrit = 16,def.noc.sleep = 1, nnights = 16,outliers.only = FALSE, criterror = 4,do.visual = TRUE, threshold.lig = c(30),threshold.mod = c(100),threshold.vig =
c(400),boutcriter = 0.8,boutcriter.in = 0.9,boutcriter.lig = 0.8, boutcriter.mvpa = 0.8, boutdur.in = c(1,10,30), boutdur.lig = c(1,10),boutdur.mvpa = c(1), timewindow = c("WW"),
visualreport=TRUE, print.filename=TRUE, IVIS_windowsize_minutes=60, IVIS_epochsize_seconds=30)
6.2 Repeated measures correlation
#Install the RMCORR package. install.packages("rmcorr") library(rmcorr)
#Import a .txt table as dataset. Make sure you use dots as decimal separator in excel.
pilotdata = read.table(file.choose(), header=T)
#Check if the assumptions for repeated measures correlation are met. First, check whether predictors are a lineair function of the
dependent measure.
glm = lm(pilotdata$PSS~pilotdata$L5onset) par(mfrow=c(2,2))
plot(glm)
#Look at the QQ plot to assess linearity. Look at the top left plot to check for distribution of residuals.
#Check for equal variance, are the errors identically distributed? install.packages("car")
library("car")
#Check whether the errors/residuals are normally distributed. shapiro.test(resid(glm))
#Run rmcorr. Make sure all measures are numeric!
#rmcorr(participant = ..., measure1 = ..., measure2 = ..., dataset) rmcorr(Participant, L5onset, PSS, pilotdata)
cor = rmcorr(Participant, L5onset, PSS, pilotdata) #Visualize rmcorr results.
pilotdata %>% ggplot(aes(L5onset, PSS)) + geom_jitter(aes(color = as.factor(Participant)), alpha=0.7) + geom_line(aes(color =
as.factor(Participant)), stat="smooth",method = "lm") +
geom_line(stat = "smooth", method = "lm", se = FALSE, color = "black", linetype = "longdash", size = 0.8) + labs(x = "L5 onset (hours within the time interval)", y = "PSS score") +
guides(color=guide_legend(title="Participant")) + theme_linedraw() #Use the BestNormalize package to normalize each seperate GGIR value if needed.
TransL5onset = bestNormalize(pilotdata$L5onset) View(TransL5onset[["x.t"]])
#L5onset is used as example here. Repeat previous steps for all GGIR outcome values to generate separate correlations.
6.3 Repeated measures ANOVA
#Install and load packages (if needed). install.packages("tidyverse") library(tidyverse) install.packages("ggpubr") library(ggpubr) install.packages("rstatix") library(rstatix)
# Import a .txt table as dataset. Make sure you use dots as decimal separator in excel.
anovaL5onset = read.table(file.choose(), header=T)
#Plot the data to see whether there are any significant outliers. bxp = ggboxplot(anovaL5onset, x = "Time", y = "L5onset", add = "point")
bxp
anovaL5onset %>% group_by(Time) %>% identify_outliers(L5onset) #Check if the assumptions for a one-way repeated measures ANOVA are met. First check whether each group is normally distributed.
anovaL5onset %>% group_by(Time) %>% shapiro_test(L5onset) #Sphericity is automatically checked and corrected with the following ANOVA function. Run the ANOVA.
aov <- anova_test(data = anovaL5onset, dv = L5onset, wid = Participant, within = Time)
get_anova_table(aov)
#Run a post-hoc pairwise T-test to see where the significant differences are.
pairwisecomp <- anovaL5onset %>% pairwise_t_test(L5onset~Time, paired = TRUE, p.adjust.method = "bonferroni")
pairwisecomp
#Visualize the RM ANOVA results.
anovaL5onset %>% ggplot(aes(Time, L5onset, fill = Time)) + geom_boxplot(outlier.shape = NA, alpha = 0.7) + geom_point() + labs(y = "Onset of L5 (hours within the time interval)", x = "") + theme_classic() + scale_fill_locuszoom() + theme(legend.title = element_blank())
#L5onset is used as example here. Repeat previous steps for all GGIR outcome values to generate separate correlations.
7. Supplementary figures
7.1 Supplementary figures of repeated measures correlation
Figure 5. Repeated measures correlation between daily perceived stress score and L5/M5 GGIR outcome values. (A)
Levene’s test of equal variance was not violated for the onset of L5 (F22, 103 = 0.82, P = 0.7). Shapiro-Wilk Normality test showed that the residuals were not normally distributed (W = 0.9, P = 0.0001). Repeated measures correlation showed no significant correlation between PSS score and onset of L5 (r116 = 0.05, P = 0.592). (B) Assumptions of equal variance were not violated for L5 acceleration (F22, 103 = 1.2, P = 0.27). According to Shapiro-Wilk Normality test, residuals were not normally distributed (W = 1, P = 0.0004). There was no significant repeated measures correlation between L5 acceleration and PSS score (r116 = -0.118, P = 0.205). (C) Levene’s test showed that assumptions of equal variance were not violated for the onset of M5 (F22, 103 = 0.93, P = 0.56). Shapiro-Wilk Normality test showed that the residuals were not normally distributed (W = 1, P = 0.001). Repeated measures correlation showed no significant correlation between PSS score and onset of M5 (r116 = -0.07, P = 0.455). (D) Assumptions of equal variance were not violated for M5 acceleration (F22, 103 = 0.66, P = 0.87). Residuals were not normally distributed according to Shapiro-Wilk Normality test (W = 1, P = 0.001). There was no significant repeated measures correlation between PSS score and M5 acceleration (r116 = -0.017, P = 0.857).
A B
Figure 6. Repeated measures correlation between daily perceived stress score and mean acceleration values. (A) Levene’s
test of equal variance was not violated for 1AM to 6AM acceleration (F22, 103 = 1.08, P = 0.38). Shapiro-Wilk Normality test showed that the residuals were not normally distributed (W = 1, P = 0.0006). Repeated measures correlation showed no significant correlation between PSS score and mean 1AM to 6AM acceleration (r116 = 0.05, P = 0.592). (B) Assumptions of equal variance were not violated for total mean acceleration (F22, 103 = 0.86, P = 0.64). According to Shapiro-Wilk Normality test, residuals were not normally distributed (W = 1, P = 0.001). There was no significant repeated measures correlation between total mean acceleration and PSS score (r116 = -0.035, P = 0.696).
A B
A B
Figure 7. Repeated measures correlation between daily perceived stress score and unbouted time spend in activity. (A)
Levene’s test of equal variance was not violated for unbouted time spend in inactivity (F22, 103 = 0.79, P = 0.73). Shapiro-Wilk Normality test showed that the residuals were not normally distributed (W = 1, P = 0.0007). Repeated measures correlation showed no significant correlation between PSS score and unbouted time spend in inactivity (r116 = -0.09, P = 0.333). (B) Assumptions of equal variance were not violated for unbouted time spend in light activity (F22, 103 = 0.61, P = 0.91). According to Shapiro-Wilk Normality test, residuals were not normally distributed (W = 1, P = 0.0007). There was no significant repeated measures correlation between time spend in unbouted light activity and PSS score (r116 = -0.114, P = 0.221). (C) Levene’s test showed that assumptions of equal variance were not violated for unbouted time spend in moderate activity (F22, 103 = 0.66, P = 0.87). Shapiro-Wilk Normality test showed that the residuals were not normally distributed (W = 1, P = 0.0007). Repeated measures correlation showed no significant correlation between PSS score and unbouted time spend in moderate activity (r116 = -0.077, P = 0.408). (D) Assumptions of equal variance were not violated for unbouted time spend in vigorous activity (F22, 103 = 0.89, P = 0.61). Residuals were not normally distributed according to Shapiro-Wilk Normality test (W = 1, P = 0.0007). There was no significant repeated measures correlation between PSS score and unbouted time spend in vigorous activity (r116 = -0.119, P = 0.201).
Figure 8. Repeated measures correlation between daily perceived stress score and total time spend in activity. (A)
Assumptions of equal variance were not violated for total time spend in inactivity (F22, 103 = 1.39, P = 0.14). Shapiro-Wilk Normality test showed that the residuals were not normally distributed (W = 1, P = 0.0009). There was no significant repeated measures correlation between PSS score and total time spend in inactivity (r116 = 0.175, P = 0.058). (B) Levene’s test showed that assumptions of equal variance were not violated for total time spend in light activity (F22, 103 = 0.84, P = 0.67). Assumptions of normal distribution were violated (W = 1, P = 0.0009). Repeated measures correlation showed no significant correlation between total time spend in light activity and PSS score (r116 = -0.144, P = 0.119). (C) According to Levene’s test, assumptions of equal variance were not violated for total time spend in vigorous activity (F22, 103 = 1.04, P = 0.43). Shapiro-Wilk Normality test
C
showed that the residuals were not normally distributed (W = 1, P = 0.0006). There was no significant repeated measures correlation between PSS score and total time spend in vigorous activity (r116 = 0.073, P = 0.43).
Figure 9. Repeated measures correlation between daily perceived stress score and sleep parameters. (A) The Levene’s test
showed that assumptions of equal variance were met for onset of sleep (F22, 103 = 0.87, P = 0.63). Assumptions of normal distribution were violated according to the Shapiro-Wilk Normality test (W = 1, P = 0.001). There was no significant repeated measures correlation between onset of sleep and PSS score (r116 = -0.132, P = 0.155). (B) Levene’s test of equal variance showed that assumptions were not violated for duration of sleep (F22, 103 = 1.54, P = 0.08). Shapiro-Wilk Normality test showed that the
A B
C D
residuals of sleep duration was not normally distributed (W = 1, P = 0.0002). Repeated measures correlation showed no significant correlation between PSS score and sleep duration (r116 = -0.1, P = 0.28). (C) Assumptions of equal variance were not violated for duration of the SPT-window (F22, 103 = 1.36, P = 0.15). Shapiro-Wilk Normality test showed that the residuals were not normally distributed (W = 1, P = 0.0004). There was no significant repeated measures correlation between PSS score and duration of the SPT-window (r116 = -0.151, P = 0.102). (D) Levene’s test showed that assumptions of equal variance were not violated for DURSIBD (F22, 103 = 1.25, P = 0.23). Assumptions of normal distribution were violated (W = 1, P = 0.002). Repeated measures correlation showed no significant correlation between PSS score and DURSIBD (r116 = -0.179, P = 0.053). (E) According to Levene’s test, assumptions of equal variance were not violated for sleep efficiency (F22, 103 = 1.17, P = 0.29). Shapiro-Wilk Normality test showed that the residuals were not normally distributed (W = 1, P = 0.0007). Repeated measures correlation showed no significant correlation between PSS score and sleep efficiency (r116 = 0.142, P = 0.125).
7.2 Supplementary figures of repeated measures ANOVA
Figure 10. One-way repeated measures ANOVA between different measurement times for L5/M5 GGIR outcome values. (A) After using the arcsinh transformation, the Shapiro-Wilk Normality test showed that the assumption of normal distribution
was not violated for L5 onset measured over weeks (W = 0.918, P = 0.373), weekdays (W = 0.924, P = 0.424) or weekends (W = 0.926, P = 0.44). The repeated measures ANOVA showed no significant differences between weeks, weekdays and weekends of measuring L5 onset (F1, 8 = 3.361, P = 0.104). (B) According to Shapiro-Wilk Normality test, assumptions of normal distribution were not violated for weeks (W = 0.982, P = 0.976), weekdays (W = 0.971, P = 0.907) or weekends (W = 0.975, P = 0.932) of
A B
measuring L5 acceleration. One-way repeated measures ANOVA showed that there were no significant differences between measurement times of L5 acceleration (F2, 16 = 0.005, P = 0.995). (C) The Shapiro-Wilk Normality test showed that assumptions of normal distribution were met for measuring weeks (W = 0.972, P = 0.909), weekdays (W = 0.932, P = 0.496) and weekends (W = 0.84, P = 0.058) of M5 onset. Based on repeated measures ANOVA, there were no significant differences between measurement times of M5 onset (F2, 16 = 0.817, P = 0.459). (D) After performing an arcsinh transformation, Shapiro-Wilk Normality test showed that assumptions of normal distribution were not violated for weeks (W = 0.966, P = 0.856), weekdays (W = 0.944, 0.624) or weekends (W = 0.876, P = 0.144) of measuring M5 acceleration. One-way repeated measures ANOVA showed that there were no significant differences between measurement times of M5 acceleration (F1, 8 = 0.928, P = 0.364).
Figure 11. One-way repeated measures ANOVA between different measurement times for mean acceleration values. (A)
Shapiro-Wilk normality test showed that data for weekdays of 1 AM to 6AM acceleration followed a normal distribution (W = 0.92, P = 0.389). However, assumptions of normality were violated for measuring weeks (W = 0.668, P = 0.0006) and weekends (W = 0.533, P = 0.00002) of 1AM to 6AM acceleration. Friedman’s non-parametric test showed that there were no significant differences between measurement times of 1AM to 6AM acceleration (F2 = 0.222, P = 0.895). (B) According to Shapiro-Wilk Normality test, assumptions of normal distribution were not violated for total mean acceleration measured over weeks (W = 0.95, P = 0.687), over weekdays (W = 0.98, P = 0.963) or over weekends (W = 0.891, P = 0.202). One-way repeated measures ANOVA showed that there were no significant differences between measurement times of total mean acceleration (F2, 16 = 2.186, P = 0.145).
A B
Figure 12. One-way repeated measures ANOVA between different measurement times for sleep parameters. (A) According
to the Shapiro-Wilk Normality test, assumptions of normal distribution were met for measuring duration of sleep over weeks (W = 0.947, P = 0.663), over weekdays (W = 0.959, P = 0.789) and over weekends (W = 0.974, P = 0928). One-way repeated measures ANOVA showed that there were no significant differences between measurement times of sleep duration (F2, 16 = 3.314, P = 0.062). (B) Shapiro-Wilk Normality test showed that the data of duration of the SPT-window was normally distributed for measuring weeks (W = 0.944, P = 0.621), weekdays (W = 0.953, P = 0.724) and weekends (W = 0.927, P = 0.449). Based on one-way repeated measures ANOVA, there were no significant differences between measurement times of duration of the SPT-window (F2, 16 = 1.705, P = 0.213). (C) Shapiro-Wilk normality test showed that assumptions of normal distribution were met for measuring weeks (W = 0.851, P = 0.077) and weekdays (W = 0.891, P = 0.203) of DURSIBD. However, assumptions of normality were violated for measuring weekends (W = 0.712, P = 0.002) of DURSIBD. Friedman’s non-parametric test showed that there were no significant differences between different measurement times of DURSIBD (F2 = 0.222, P = 0.895). (D) Shapiro-Wilk Normality test showed that measurements of sleep efficiency followed a normal distribution during weeks (W = 0.935, P = 0.534), weekdays (W = 0.933, P = 0.513) and weekends (W = 0.920, P = 0.393). One-way repeated measures ANOVA showed that there were no significant differences between different measurement times of sleep efficiency (F2, 16 = 0.983, P = 0.396).
