Assessment of cardiorespiratory fitness
Baseline measurements included digital questionnaires, anthropometrical measurements, and objective measurements of cardiorespiratory fitness and heart rate. Cardiorespiratory fitness was assessed using a submaximal cycle ergometer test (Åstrand and Ryhming
1954), on an Ergomedic 874 E cycle ergometer (Monark AB, Varberg, Sweden). This test measures the heart rate required to achieve a given power output (work conducted) to provide an estimate of
VO
2max that takes into account both age and gender. Initial power output estimation was made using the information on age and estimated fitness. Typical output was estimated at between 60 and 90 W at a cadence of 60 revolutions per minute. Tests lasted a maximum of 10 min and were terminated once a steady-state heart rate was registered—defined as a heart rate change of less than 5 beats/min from 5 to 6 min. If a heart rate of less than 110 beats/min was registered after the first minute, power output was increased to achieve a registered heart rate at or above 60% of the estimated maximum or at least 120 beats/min. Heart rate was measured with a handheld pulse oximeter (Nellcor OxiMax N-65, United States), fixed to the fingertip. The combination of power output and heart rate were then used in the estimation of maximal oxygen uptake (Åstrand and Ryhming
1954).
Assessment of aerobic workload
Following baseline measurements, a single Actiheart device (Actiheart, CamNTech Ltd., Cambridge, England) was used to measure 24-h heart rate, over four consecutive days. Actiheart consists of a single electrocardiography node attached at the apex of the sternum and connected to a secondary node, which is attached over any intercostal space on the left side of the ribcage. Workers were also requested to log work, leisure, and sleep periods, and periods without wearing the monitors in a diary provided after device placement. Furthermore, workers were instructed to remove any monitors causing irritation or discomfort.
For heart rate data to be considered a valid representation of aerobic workload, the Actiheart needed to be worn for at least one valid day. A valid day was defined as ≥ 4 h of heart-rate data collection during work hours or ≥ 75% of the individual's average work hours. We also excluded measurements with beat error > 50%.
The calculation of heart rate reserve (%HRR) was as follows (Karvonen
1957):
$$ {\text{\% HRR}}\, = \,\frac{{{\text{HR}}_{{{\text{HRwork}}}} \,{ - }\,{\text{HR}}_{{{\text{HRmin}}}} }}{{{\text{HR}}_{{{\text{HRmax}}}} \,{ - }\,{\text{HR}}_{{{\text{HRmin}}}} }} \times 100{ } $$
(1)
where HR
min was the minimum heart rate over an average of ten beats using a moving window during the course of the whole measurement period and HR
max was estimated according to (Tanaka et al.
2001):
$$ {\text{HR}}_{{\text{max }}} = 208 - 0.7 \times {\text{Age }} $$
(2)
We chose to use HRmin because heart rate, like blood pressure, can be much influenced by mental stress or physical activities if measured in a clinical health setting. Since we measured the heart rate continuously during the 24 h of the day, we could therefore subtract the minimum heart rate throughout the 24 h. Thus, to attain the most precise estimation of HRR we used the lowest heart rate period over the day, which we believe provides the most valid estimate of the true sleeping heart rate of an individual.
Actiheart devices were initialized and downloaded using Actiheart software. All were set to short-term recording mode, allowing inter-beat intervals to be captured for up to 440,000 heartbeats. Once downloaded, the inter-beat intervals recorded on Actiheart were further processed and checked for errors using Acti4 software using methods described in full previously (Kristiansen et al.
2011). In brief, Acti4 software splits 24-h heart rate measurements into periods of work, leisure, and sleep time based on the information contained in participant diaries (Skotte and Kristiansen
2014). Inter-beat intervals were then resampled at a 4 Hz frequency through the implementation of a linear interpolation scheme and the calculations outlined above were completed (Eqs.
1,
2).
Inter-beat intervals corresponding to less than 36 beats/min or greater than 200 beats/min were discarded (Skotte and Kristiansen
2014). Additionally, intervals differing by more than 15% compared to their neighbouring intervals or containing an error rate greater than 50% were discarded (Skotte and Kristiansen
2014).
Assessment of covariates and other variables
Age was determined based on the date of birth of the participant and sex was determined using the question “Are you male or female?” self-rated general health was determined by the question “How will you rate your overall health?” rated on a scale of 1 = very good, 2 = good, 3 = fairly good, 4 = poor, 5 = very poor. Participant’s use of medication was determined using the question “Have you in the last three months been taking prescription medication?” with a dichotomous ‘yes/no’ response category. This question was followed by “If yes, what kind of medication?” also using a dichotomous ‘yes/no’ response category. Occupational sector was determined according to the workplace, and occupation itself was determined by the question “What is your present main occupation?” with the response categories: 1 = blue-collar, 2 = white-collar and 3 = manager. For our analyses, these two variables (sector and occupation) were combined into a new occupation variable in the following way: blue-collar workers from each of the three occupational sectors (cleaning, manufacturing, transportation) were classified according to their sector, while white-collar workers and managers from each of the sectors were classified together as ‘administration workers’. Shift workers were identified with the question “At what time of the day do you usually work in your main occupation?” with category responses “fixed day work” and “night/varying/other”. Participants reported their workability on a 0–10 scale in response to “Please rate your present work ability?”.
Statistical analyses
We chose to express the time spent above and below 30% of HRR as a proportion of an approximately 8-h workday. Therefore, this data was first transformed into isometric log-ratio coordinates in accordance with the principles of compositional data analyses (CoDA). Simply put, this method allows for the time spent above and below 30% HRR to be expressed relative to each other. As such, instead of investigating only the time spent above 30% HRR, we investigate the time spent above 30% HRR relative to the time spent below 30% HRR. In this manner, we can account for individual differences in the daily composition of HRR (i.e. the proportion a 24-h day with a HRR above or below 30% of an individual’s overall HRR).
The association between cardiorespiratory fitness and aerobic workload
The primary analysis involved two regression models for each outcome. The first was an unadjusted model that investigated the association between cardiorespiratory fitness and each HRR outcome. The second was an adjusted model that included age, sex, self-rated health, shift-work, prescription medication and occupation as potential confounders. These potential confounders were retained in the model regardless of significance. Where significant effects of potential confounders were identified (α = 0.05), the interaction was assessed by adding an interaction term to the model. We used scatterplots with locally weighted polynomial regression (LOWESS) lines to visualise the relationship between cardiorespiratory fitness and each outcome. Contour plots were used to visualise the effects of identified interactions when the interacting variable (i.e. age) was continuous. We chose to include LOWESS lines as they are a widely utilized method for displaying relationships between investigated variables that is unconstrained by the assumptions used in a regression model. This same logic is used for the inclusion of the contour plots, which are developed using LOWESS principles. Linear regression models were used as they provided a satisfactory fit for the data.
Stratification by occupation, pace of work, and age
The secondary analysis involved stratification by occupational group and age. We chose to stratify by the occupational group due to differences in activity levels between each of these groups. However, this stratification revealed that there was also a wide variation in physical activity levels within occupational groups. Therefore, we decided post-hoc to further stratify according to quartiles of steps-per-hour at work as a proxy for the pace or intensity of work, thus providing a rough approximation of work pace regardless of the occupational group. We also chose to stratify by age (using quartiles) to understand the identified interaction between cardiorespiratory fitness and age. We chose quartiles to maintain a balanced number of participants between groups and because we had no specific reason to choose any other value(s). In all cases, stratification was based upon the adjusted model (i.e. all potential confounders were included in the model except the variable used to stratify). All analyses were conducted in R (R Core Team
2018)/Studio (RStudio Team
2016) using packages: compositions (van den Boogaart et al.
2018), robcompositions (Matthias et al.
2019), car (Fox
2018), lmtest (Hothorn et al.
2018), and the tidyverse suite of packages (Wickham et al.
2019).