Physical activity, body mass index and heart rate variability-based stress and recovery in 16 275 Finnish employees : a cross-sectional study

This is an electronic reprint of the original article. This reprint may differ from the original in pagination and typographic detail. Author(s): Title: Year: Version: Please cite the original version: All material supplied via JYX is protected by copyright and other intellectual property rights, and duplication or sale of all or part of any of the repository collections is not permitted, except that material may be duplicated by you for your research use or educational purposes in electronic or print form. You must obtain permission for any other use. Electronic or print copies may not be offered, whether for sale or otherwise to anyone who is not an authorised user. Physical activity, body mass index and heart rate variability-based stress and recovery in 16 275 Finnish employees : a cross-sectional study Föhr, Tiina; Pietilä, Julia; Helander, Elina; Myllymäki, Tero; Lindholm, Harri; Rusko, Heikki; Kujala, Urho Föhr, T., Pietilä, J., Helander, E., Myllymäki, T., Lindholm, H., Rusko, H., & Kujala, U. (2016). Physical activity, body mass index and heart rate variability-based stress and recovery in 16 275 Finnish employees : a cross-sectional study. BMC Public Health, 16, Article 701. https://doi.org/10.1186/s12889-016-3391-4 2016 RESEARCH ARTICLE Open Access Physical activity, body mass index and heart rate variability-based stress and recovery in 16 275 Finnish employees: a cross-sectional study Tiina Föhr 1* , Julia Pietilä 2 , Elina Helander 2 , Tero Myllymäki 3 , Harri Lindholm 4 , Heikki Rusko 5 and Urho M. Kujala 1 Abstract Background: Physical inactivity, overweight, and work-related stress are major concerns today. Psychological stress causes physiological responses such as reduced heart rate variability (HRV), owing to attenuated parasympathetic and/or increased sympathetic activity in cardiac autonomic control. This study’s purpose was to investigate the relationships between physical activity (PA), body mass index (BMI), and HRV-based stress and recovery on workdays, among Finnish employees. Methods: The participants in this cross-sectional study were 16 275 individuals (6863 men and 9412 women; age 18–65 years; BMI 18.5–40.0 kg/m 2 ). Assessments of stress, recovery and PA were based on HRV data from beat-to-beat R-R interval recording (mainly over 3 days). The validated HRV-derived variables took into account the dynamics and individuality of HRV. Stress percentage (the proportion of stress reactions, workday and working hours), and stress balance (ratio between recovery and stress reactions, sleep) describe the amount of physiological stress and recovery, respectively. Variables describing the intensity (i.e. magnitude of recognized reactions) of physiological stress and recovery were stress index (workday) and recovery index (sleep), respectively. Moderate to vigorous PA was measured and participants divided into the following groups, based on calculated weekly PA: inactive (0 min), low (0 < 150 min), medium (150–300 min), and high (>300 min). BMI was calculated from self-reported weight and height. Linear models were employed in the main analyses. Results: High PA was associated with lower stress percentages (during workdays and working hours) and stress balance. Higher BMI was associated with higher stress index, and lower stress balance and recovery index. These results were similar for men and women (P< 0.001 for all). Conclusion: Independent of age and sex, high PA was associated with a lower amount of stress on workdays. Additionally, lower BMI was associated with better recovery during sleep, expressed by a greater amount and magnitude of recovery reactions, which suggests that PA in the long term resulting in improved fitness has a positive effect on recovery, even though high PA may disturb recovery during the following night. Obviously, several factors outside of the study could also affect HRV-based stress. Keywords: Body mass index, Heart rate variability, Physical activity, Physiological stress, Stress, Stress assessment * Correspondence: [email protected] 1 Department of Health Sciences, University of Jyväskylä, P.O. Box 35FIN-40014 Jyväskylä, Finland Full list of author information is available at the end of the article © 2016 The Author(s). Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated. Föhr et al. BMC Public Health (2016) 16:701 DOI 10.1186/s12889-016-3391-4 Background Physical activity (PA) is known to have positive effects on health [1, 2]. Routine PA reduces stress and enhances psychological wellbeing, which is particularly important for the prevention and management of cardiovascular disease, among other chronic diseases [3]. Regular PA is known to reduce the risk of many adverse health outcomes. Some PA is better than none; however, for most health outcomes, additional benefits are achieved if the amount of PA increases through higher intensity, greater frequency, and/or longer duration. According to the 2008 Physical Activity Guidelines for Americans, most health benefits occur with at least 150 total minutes of moderate intensity or at least 75 min of vigorous intensity aerobic PA per week. However, additional benefits occur with more PA [4, 5]. In addition to the beneficial effects of PA on physical health, these guidelines are also relevant for mental health [6]. Although leisure-time PA has increased among Finnish adults [7], physical inactivity is a major problem and risk for health, in all countries. Furthermore, physical inactivity is associated with being overweight [8] and the current rate of overweight adults worldwide has been described as an epidemic or even a pandemic. This situation is a major public health risk because being overweight is associated with diseases including coronary heart disease, stroke, diabetes and cancer [9]. Together with physical inactivity and overweight, stress at work is a major public health risk. It may even lead to cardiovascular disease [10] without complete recovery [11]. Stress has been shown to reduce participation in leisure-time PA [12, 13]. Furthermore, workplace stress may predict a future increased risk of insufficient PA [14]. Normal weight is associated with good self-reported subjective health [15], including low stress levels [16, 17]. Evidence suggests that psychosocial stress is associated with the development of adiposity [18]. However, according to previous studies, the association between subjective stress and body composition is inconsistent, with evidence both supporting [16, 19] and refuting [20, 21] the idea that stress is associated with adiposity. A recent systematic review reported that the associations of psychosocial factors at work with weight-related outcomes were weak and somewhat inconsistent [22]. Psychological stress causes sympathetic responses in the autonomic nervous system (ANS), such as reduced heart rate variability (HRV) [23]. HRV refers to the variation in intervals between heartbeats and reflects cardiac autonomic modulation. Physiological stress can be defined as an increased body activation level, when sympathetic activity dominates the ANS and parasympathetic activation is low. Stress is associated with reduced HRV, owing to attenuated parasympathetic and/or increased sympathetic activity in cardiac autonomic control. Recovery refers to a reduced body activation level, when parasympathetic activation dominates the ANS over sympathetic activity [24–26]. HRV analysis can be used as a complementary tool to assess general health [27]. HRV analysis during sleep has the potential to explore the sleeping brain, with possible implications for mental health [28]. Previous HRV-studies have mainly used traditional time-domain and frequency-domain measures of HRV, such as root mean square of successive R-R intervals (RMSSD) and the ratio of low frequency power to high frequency power (LF/ HF ratio). The traditional measures of HRV represent the average level of the autonomic activity over a period of the time. Cardiac autonomic activity is very dynamic and varies during the day depending on stress, recovery and PA. Therefore, the usability of the traditional measures of HRV is limited in real-life conditions. Additionally, these measures are very individual which further limits their usability in stress assessment and clinical work. However, it is also possible to provide applied heart rate (HR) and HRVderived stress and recovery variables that take into account the dynamic changes in autonomic activity and individuality of HRV including information that is difficult to obtain from traditional measures of HRV. The majority of previous studies on the association of PA with stress have used subjective assessment methods or traditional measures of HRV in the assessment of stress. The previous studies support the association of PA with increased HRV [29, 30]. However, accurate and objective methods are needed to reliably assess PA, as well as to assess HRV-based stress and recovery in reallife. By utilizing a method that acknowledges the dynamics and individuality in HRV in real-life, the aim of this study was to investigate the extent to which PA and BMI are associated with HRV-based indicators of stress and recovery on workdays. The study was conducted among 16 275 Finnish employees who had participated in beat-to-beat R-R interval recording as a part of lifestyle counseling between 2007 and 2015. More specifically, accounting for age and sex, we investigated the prevalence of stress and recovery according to the participants’objectively measured PA level and self-reported body mass index (BMI). Uniqueness of the present study is in the individual and dynamic method used in the assessment of physiological stress and recovery. Methods Study design and participants This cross-sectional study investigated the amount and intensity of objective HRV-based stress and recovery on workdays in a real-life sample of 16 275 Finnish employees (6863 men and 9412 women; age 18–65 years; BMI 18.5–40.0 kg/m 2 ). The participants nonselectively represent a cross-section of typical Finnish employees including both manual and non-manual labour employees. Föhr et al. BMC Public Health (2016) 16:701 Page 2 of 13 The majority of the participants were apparently healthy without chronic diseases. The exclusion criteria for participation in the R-R interval recordings included severe cardiac disease, very high blood pressure (≥180/ 100 mmHg), type 1 or 2 diabetes with autonomic neuropathy, severe neurological disease, fever or other acute disease, and BMI >40 kg/m 2 . These exclusion criteria represented by the analysis software manufacturer are presented in detail previously [31]. The characteristics of the participants are presented in Table 1. Data collection The novel methodology used to determine the participants’stress, recovery and level of weekly PA, was based on HRV data from beat-to-beat R-R interval recordings. These recordings were voluntarily performed on employees as a part of the preventive occupational health care programs provided by their employers between 2007 and 2015. The clinical purpose of these measurements is presented comprehensively in a previous paper by Mutikainen et al. [31]. The data recordings used in the previous study were gathered between 2007 and 2013, with a study population size of 9554. These data were used in the present study, supplemented with recordings from 2014 to 2015. The study had a further inclusion criterion of a minimum of 4.5 h beat-to-beat R-R interval recording during sleep after a workday. Another inclusion criterion was the availability of R-R interval data, including at least one workday (≥4 h of work) and one day off, with a measurement period of 16–30 h/day (from wake-up to wake-up). Participants who had consumed alcohol on the monitoring days were excluded. Information about workdays, working hours, days off and sleep periods was obtained from diaries that the participants were requested to keep during the measurement period. The analyzed data consisted of successfully recorded (measurement error <15 % and <30-min recording break) days. The flow of the participants included in the analysis is presented in Fig. 1. HRV-based assessment of PA, stress and recovery Ambulatory beat-to-beat R-R interval data were used to determine the amount and intensity of PA, stress and recovery. Using the Firstbeat Bodyguard device (Firstbeat Technologies Ltd., Jyväskylä, Finland), real-life R-R interval data were recorded, usually over 3 days (typically two workdays and one day off) and analyzed using Firstbeat Analysis Server software (version 6.3, Firstbeat Technologies Ltd.), which included a powerful artifact detection and correction feature for irregular ectopic beats and signal noise. The software calculates HRV indices second-by-second using the short-time Fourier transform method, and calculates HRand HRV-derived variables of respiration rate, oxygen consumption, on-off kinetics (increasing or decreasing HR), and parameters describing excess post-exercise oxygen consumption using neural networks. Thereafter, the software divides the measurement data into coherent data segments and categorizes these segments into different physiological states, such as PA of different intensities, stress and recovery [32–34], by taking into account individual characteristics (e.g. individual levels and scales of HR and HRV, and the individual relationships between HRV and autonomic control) [35]. The categorization of the data is described in Additional file 1: Table S1. More information about this analysis method is available in a paper by Firstbeat Technologies Ltd. [36]. Detection of stress and recovery variables After data categorization, the HRV-based variables describing the amount and intensity of stress and recovery on workdays were detected. Stress percentages (i.e. proportions of stress reactions, during the day and during working hours) and stress balance (ratios between recovery and stress reactions during sleep) describe the amount of stress and recovery, respectively. The variables describing the intensity (i.e. magnitude of recognized reactions) of stress and recovery were stress index (during the day) and recovery index (during sleep), respectively. These variables and their calculations are presented in Additional file 1: Table S1. The correlation coefficient between two consecutive workdays varied from 0.74 to 0.88 for the traditional HRV variables, from 0.64 to 0.93 for HRV-derived variables of stress, and from 0.42 to 0.49 for HRV-derived variables of recovery during sleep. Calculation of weekly PA Background information about age, sex, self-reported height and weight, and self-reported PA class [37] modified from Ross and Jackson [38], was collected in Table 1 Characteristics of the participants Variable All (n= 16275) Men (n= 6863) Women (n= 9412) Mean ± SD Min Max Mean ± SD Min Max Mean ± SD Min Max Age 44.8 ± 9.9 18.0 65.0 44.5 ± 9.9 18.0 65.0 45.0 ± 9.9 18.0 65.0 Body mass index (kg/m 2 ) 26.0 ± 4.1 18.5 40.0 26.6 ± 3.5 18.6 40.0 25.5 ± 4.4 18.5 40.0 Selfreported activity class 0–10 4.8 ± 1.8 0.0 10.0 4.9 ± 1.9 0.0 10.0 4.8 ± 1.8 0.0 10.0 Physical activity (mins/week) 186 ± 227 0 2629 246 ± 258 0 2629 142 ± 189 0 1865 Föhr et al. BMC Public Health (2016) 16:701 Page 3 of 13 conjunction with R-R interval recordings using questionnaires. Background information was used to estimate maximal HR [39] and maximal VO 2 [40] which were then used in the estimation of VO 2 . The maximal HR used for further calculations was corrected accordingly if a period with HR higher than the estimated maximal was found from the recording. From the second-bysecond VO 2 estimations, each participant’s mean VO 2 for each minute of the measurement day was calculated. The minute-by-minute VO 2 estimations were then converted to multiples of the resting metabolic rate (MET) by dividing the VO 2 values by 3.5. The total number of 1-min segments within the following thresholds: moderate PA 3 to <6 METs and vigorous PA ≥6 METs, during each measurement day (including days off), were calculated. Continuous bouts of PA lasting for ≥10 min were included in the estimation of weekly PA. These continuous bouts of PA were calculated separately for workdays and days off, and, if the measurement period included two or more workdays or days off, an average was calculated. The activity minutes score for each day (moderate PA minutes + vigorous PA minutes × 2) was calculated. Thereafter, the amount of PA was extrapolated using the following formula: PA minutes per week = (5 × mean workday activity score) + (2 × mean day-off activity score). These calculations have been previously described in more detail [31]. Based on the weekly PA minutes, the participants were divided into the following PA groups: inactive (0 min), low (0 < 150 min), medium (150–300 min) and high (>300 min). Assessment of body composition BMI was calculated from the self-reported weight and height (kg/m 2 ). The participants were then divided into the following groups: normal weight (18.5 to <25 kg/m 2 ), overweight (25 to <30 kg/m 2 ) and obese (30–40 kg/m 2 ). Analysis Data processing and statistical analysis were performed using R 3.2.2 version (R Foundation for Statistical Computing). P-values were two-sided and a p-value of <0.05 was considered statistically significant. Because of the size of the data, 99 % confidence intervals (CIs) were determined (Fig. 2) instead of conventional 95 % CIs. The main outcome variables of the study were stress percentage and stress index, calculated for the whole day, stress percentage calculated for working hours, and stress balance and recovery index calculated for sleep. These variables were derived from the beat-to-beat R-R interval recordings on workdays. For a more detailed description of the variables see Additional file 1: Table S1. In addition, HR and traditional HRV parameters, including RMSSD and the LF/HF ratio, were calculated from the beat-to-beat R-R interval recordings on workdays. These variables were calculated separately for waking hours and sleep, and RMSSD was calculated using a 5min window. If the measurement period of a subject included two or more workdays, an average was calculated and the mean values of the outcome variables were used in the analysis. Fig. 1 Flow of the participants and measurement days included in the analysis (BMI, body mass index) Föhr et al. BMC Public Health (2016) 16:701 Page 4 of 13 For the descriptive statistics, the means and standard deviations of the outcome variables were calculated separately for men and women, and stratified based on PA, BMI and age. Differences in the outcome variables between PA, BMI and age groups were tested using the Kruskal-Wallis test. The results are shown in Additional file 2: Tables S2-S4. To show the effects of BMI and PA group on the HRV-based stress and recovery variables, the age-controlled mean values and 99 % CIs for the HRV-based stress and recovery variables, by BMI and PA group, are presented in Fig. 2. Linear models were employed to study the effects of PA group, BMI and age on HRV-based stress and recovery variables. In the models, age and BMI were incorporated as continuous predictor variables and objectively measured PA group was incorporated as a categorical predictor variable. The models were generated separately for men and women. The reference value for age was set to 18 years and for BMI to 18.5 kg/m 2 . A simple linear least squares regression model (procedure lm in R) was applied to predict the stress percentage during the day. As confirmed by visual inspection, the assumption of Fig. 2 Stress and recovery by physical activity and body mass index groups with age-controlled mean values and 99 % CIs for output variables. Physical activity groups: inactive (0 min/week), red ●; low (0 < 150 min/week), black ▲; medium (150–300 min/week), blue ■; high (>300 min/ week), green ♦ Föhr et al. BMC Public Health (2016) 16:701 Page 5 of 13 linear regression considering the normal distribution of the residuals was not fulfilled for stress percentage during working hours, stress index and recovery index. Thus, a Box-Cox transformation was applied on these dependent variables [41]. The Box-Cox coefficient was determined by maximizing the log-likelihood function and was rounded to two decimal places before transformation. Tobit regression model (procedure vglm using iteratively reweighted least squares in R) was applied for modeling stress balance with a fixed lower and upper limit of −1 and 1, respectively. The interactions of the predictors were not included in the final regression models because the coefficient of determination for the interaction models was only a few percentage points greater than for the simple models. Results The total number of workdays included in the analysis was 28 314, with measurements obtained from 16 275 participants (men 6863; women 9412). The participants’ characteristics are shown in Table 1. The participants’ mean age was 44.8 years (44.5 years for men; 45.0 years for women) and the mean BMI was 26.0 kg/m 2 (26.6 kg/ m 2 for men; 25.5 kg/m 2 for women). The participants’ mean self-reported activity class was 4.8 (4.9 for men; 4.8 for women) indicating that, on average, the participants were involved in PA 2–3 times per week and their total weekly PA was about 2 h. The mean weekly minutes of objective monitoring-based PA was 186 (246 for men; 142 for women). The mean weekly minutes of PA in the group of low PA was 78 for men and 74 for women, in the group of medium PA 222 for men and 215 for women, and in the group of high PA 545 for men and 496 for women. The number of participants in the PA, BMI, and age groups is presented in Additional file 3: Table S5. Differences in outcome variables between PA groups (Additional file 2: Table S2) were statistically significant except for LF/HF ratio during waking hours and sleep, stress balance in men, and HR and stress balance in women. For both men and women, the high PA group had the highest RMSSD (during waking hours and during sleep) and recovery index, and the lowest stress percentage (during the day and during working hours) and stress index. Differences in outcome variables between BMI groups were statistically significant for both men and women (Additional file 2: Table S2). Normal-weight individuals had the highest RMSSD (both during waking hours and during sleep), stress balance and recovery index, the lowest stress percentage during the day and the lowest stress index. Stress percentage during working hours was lowest in obese individuals. In both sexes, differences in outcome variables were statistically significant between age groups, except for HR during sleep in women (Additional file 2: Table S3). The youngest age group (18–30 years) had the highest RMSSD (both during waking hours and during sleep), stress balance and recovery index, and the lowest stress index. Stress percentages during the day were lowest in the youngest age group in women, and in the oldest age group (51–65 years) in men. Stress percentages during working hours were lowest in the oldest age group in both sexes. Figure 2 shows the effect of PA and BMI group on the stress and recovery variables with the effect of age controlled. The high PA group had the lowest mean stress percentage during the day and during working hours in all three BMI groups, after adjustment for age (Fig. 2). Mean stress index values increased as the BMI group changed from normal weight to overweight and overweight to obese, regardless of the PA group. In addition, regardless of the PA group, obese individuals had the lowest stress balance and recovery index. The linear model results are shown in Table 2. Medium (P<0.05)andhigh(P< 0.001) PA groups, lower BMI (P< 0.001), and older age (P< 0.001) were associated with lower stress percentages during the day. Medium (P< 0.05) and high (P< 0.001) PA level, higher BMI (P< 0.001), and older age (P< 0.001) were associated with lower stress percentages during working hours. Stress percentage results during the day and during working hours were similar for men and women. Higher BMI (P<0.001) and older age (P< 0.001) were associated with higher stress index, both in men and in women. In addition, medium (P<0.01) and high(P< 0.01) PA were associated with lower stress index in women. Medium (P< 0.01) and high (P< 0.001) PA, and higher BMI (P< 0.001) were associated with lower stress balance, both in men and in women. Moreover, older age was associated with lower stress balance in men (P< 0.001). Higher BMI and older age were associated with lower recovery index, in men and in women (P< 0.001). BMI explained the highest proportion of variance in stress balance (2.2 % for men and 3.1 % for women) compared with PA and age. Discussion The purpose of this study was to investigate the amount and intensity of objective HRV-based stress and recovery on workdays. The sample group comprised 16 275 Finnish employees, who had participated in beat-to-beat R-R interval recording as a part of lifestyle counseling in the course of their everyday lives between 2007 and 2015. More specifically, the relationships between PA, BMI, and HRV-based stress and recovery were investigated. For both sexes, a high level of PA and lower BMI Föhr et al. BMC Public Health (2016) 16:701 Page 6 of 13 Table 2 Results of the linear models Men Women Parameter Estimate 95 % Cl Lower Upper Pvalue Variance explained (%) Parameter Estimate 95 % Cl Lower Upper Pvalue Variance explained (%) Stress (%), 24 h a 2.356 h 1.476 h Intercept 52.3153 50.7730 53.8575 <0.001 50.0658 48.9790 51.1527 <0.001 Age (18 years = 0) −0.1263 −0.1600 −0.0926 <0.001 0.780 i −0.0869 −0.1151 −0.0587 <0.001 0.387 i BMI (18.5 kg/m 2 = 0) 0.1541 0.0580 0.2502 0.002 0.144 i 0.0909 0.0264 0.1553 0.006 0.081 i Physical activity level (inactive = 0) 1.586 i 1.050 i Low physical activity class 0.3235 −0.6903 1.3372 0.53 −0.8892 −1.5471 −0.2313 0.008 Medium physical activity class −1.1405 −2.2144 −0.0667 0.04 −1.9539 −2.7608 −1.1470 <0.001 High physical activity class −3.9695 −5.0062 −2.9328 <0.001 −4.3772 −5.2620 −3.4923 <0.001 Stress (%), working hours b 4.152 h 1.721 h Intercept 807.3480 777.0476 837.6483 <0.001 677.8091 656.0738 699.5443 <0.001 Age (18 years = 0) −4.8729 −5.5355 −4.2103 <0.001 2.942 i −2.66482 −3.22838 −2.10127 <0.001 0.905 i BMI (18.5 kg/m 2 =0) −4.6640 −6.5515 −2.7766 <0.001 0.341 i −4.54067 −5.82984 −3.25149 <0.001 0.504 i Physical activity level (inactive = 0) 1.590 i 0.584 i Low physical activity class −5.5094 −25.4261 14.4073 0.59 −3.40975 −16.5664 9.746938 0.61 Medium physical activity class −24.7847 −45.8819 −3.6874 0.021 −17.9339 −34.0695 −1.79829 0.030 High physical activity class −85.2896 −105.6576 −64.9215 <0.001 −61.7878 −79.4833 −44.0923 <0.001 Stress index, 24 h c 27.113 h 27.448 h Intercept 1.2406 1.2403 1.2409 <0.001 1.2414 1.2412 1.2417 <0.001 Age (18 years = 0) 0.0001 0.0001 0.0001 <0.001 21.992 i 0.0001 0.0001 0.0001 <0.001 19.655 i BMI (18.5 kg/m 2 = 0) 0.0002 0.0001 0.0002 <0.001 3.819 i 0.0001 0.0001 0.0001 <0.001 3.031 i Physical activity level (inactive = 0) 0.115 i 0.145 i Low physical activity class 0.0002 0.0000 0.0004 0.024 −0.0001 −0.0003 0.0000 0.031 Medium physical activity class 0.0000 −0.0002 0.0002 0.91 −0.0002 −0.0004 −0.0001 0.002 High physical activity class 0.0001 −0.0001 0.0003 0.47 −0.0003 −0.0004 −0.0001 0.002 Stress balance, sleep e 3.669 h 3.244 h Intercept1 0.9975 0.9287 1.0663 <0.001 0.6166 0.5672 0.6661 <0.001 Intercept2 −0.5292 −0.5503 −0.5082 <0.001 −0.5537 −0.5705 −0.5368 <0.001 Age (18 years = 0) −0.0066 −0.0081 −0.0052 <0.001 1.153 i −0.0007 −0.0019 0.0006 0.31 0.006 i BMI (18.5 kg/m 2 =0) −0.0273 −0.0315 −0.0231 <0.001 2.179 i −0.0253 −0.0282 −0.0224 <0.001 3.080 i Physical activity level (inactive = 0) 0.445 i 0.233 i Low physical activity class −0.0715 −0.1162 −0.0269 0.002 −0.0126 −0.0423 0.0172 0.41 Föhr et al. BMC Public Health (2016) 16:701 Page 7 of 13 Table 2 Results of the linear models (Continued) Medium physical activity class −0.0799 −0.1272 −0.0326 <0.001 −0.0572 −0.0937 −0.0207 0.002 High physical activity class −0.1327 −0.1784 −0.0870 <0.001 −0.0822 −0.1222 −0.0421 <0.001 Recovery index, sleep f 9.685 h 3.297 h Intercept 592375.4298 573358.1497 611392.7100 <0.001 457913.8301 444648.7930 471178.8673 <0.001 Age (18 years = 0) −5092.7330 −5508.5947 −4676.8712 <0.001 7.753 i −2100.7061 −2444.6430 −1756.7692 <0.001 1.501 i BMI (18.5 kg/m 2 =0) −4825.0722 −6009.6783 −3640.4662 <0.001 0.921 i −4038.5841 −4825.3668 −3251.8014 <0.001 1.065 i Physical activity level (inactive = 0) 0.154 i 0.026 i Low physical activity class −19214.5686 −31714.7934 −6714.3437 0.003 1051.0555 −6978.4656 9080.5766 0.80 Medium physical activity class −6153.1124 −19394.2693 7088.0445 0.36 −4452.2347 −14299.7786 5395.3093 0.38 High physical activity class −9653.5430 −22437.0465 3129.9606 0.14 −5732.9247 −16532.4796 5066.6302 0.30 BMI body mass index a Linear regression b Box-Cox linear regression using transformation coefficient of 1.62 c Box-Cox linear regression using transformation coefficient of −0.79 e Tobit regression f Box-Cox linear regression using transformation coefficient of 3.18 h The proportion of variance explained by the whole model i The proportion of variance explained by the predictor variable. Calculated as the difference between the proportion of variance explained by the whole model and the proportion of variance explained by a model including all the predictor variables, except for the predictor in question Föhr et al. BMC Public Health (2016) 16:701 Page 8 of 13