Classification of occupational activity categories using accelerometry: NHANES 2003–2004

The benefits of regular physical activity (PA), including lowered mortality rates, improved function and enhanced quality of life, are widely recognized [1–4]. An individual can accumulate daily PA through occupational demands, transportation, household tasks, or engagement in leisure time PA (LTPA) [5, 6]. Higher levels of LTPA have been promoted as health enhancing, [2, 3] while the effects of occupational activity (OA) on health remain inconclusive [7–9]. Many studies have shown high levels of OA to be associated with improved health, [7, 10–13] and the detrimental effects of large amounts of occupational sitting [14, 15]. In contrast, other studies have found high levels of OA to have deleterious health effects [9, 16–19].

Due to the substantial amount of time relegated to paid work in today’s society [20–22], an individual’s occupation likely has a strong influence on daily PA and sedentary behaviors (SB) (e.g., does their job require mostly sitting, standing, walking, engaging in repetitive tasks, or heavy labor) [23]. Also OA and LTPA patterns are profoundly different in varying occupations. Blue collar workers may be highly physically active during work, and highly sedentary during leisure, while white collar workers may engage in greater LTPA, after a sedentary day at the office [24, 25]. While extensive research on the comparison of physical demands between different occupations has been performed for decades [26–28], the incorporation of OA into the measurement of daily PA is still evolving [29, 30]. Including OA in addition to LTPA will allow a better understanding of the prevalence of daily PA, and inform interventions aimed to improve health, increase productivity, and reduce work-related injury of the employed population.

The 2003–2004 NHANES contained the same 40 occupational categories, and included objectively measured PA and SB collected with accelerometry. This serendipitous coupling presented an opportunity to categorize individuals into new OA groupings using objective data. Therefore, the purpose of this analysis of the 2003–2004 NHANES occupation and accelerometry data was to classify the 40 occupational categories into high OA, intermediate OA, and low OA groupings based on objectively measured PA and SB.

As a continuous surveillance program conducted by the National Center for Health Statistics (NCHS), the NHANES assesses the health and nutritional status of non-institutionalized U.S. civilians [34]. Complex, multi-stage sampling was used to obtain a nationally representative sample. This analysis focused on a subgroup of the total population; specifically, employed individuals aged 20–60 years. The NCHS Research Ethics Review Board approved all protocols and each participant provided informed consent [34]. Data from the occupational questionnaire, interview, examination, and accelerometry components of NHANES 2003–2004 were used in this analysis. Subsequent NHANES cycles did not use similar occupational categories so the analysis herein is necessarily limited to the 2003–2004 cycle.

This study is the first to report observed accelerometer-derived PA and SB variables for a wide variety of occupational categories in U.S. adults. Accelerometry provides an objective measure of movement in everyday life and thus should yield a more valid and reproducible ranking of OA than that achieved earlier by researchers’ best estimates [32]. The information in the paper provides future researchers with detailed measures of the volume and intensity of PA and SB variables that can be compared across the 40 occupational categories based on the 1980 U.S. Census Bureau Classification Coding System. These data also highlight the importance of occupation as a determinant of daily PA and SB.

Previously, occupational categories with indeterminate levels of PA and/or those known to have high variability in PA requirements were considered to be unclassifiable in terms of OA [32]. In fact, more than half (23 of 40) of the occupational categories within NHANES 2003–2004 were considered too ambiguous to classify based on the occupation descriptions of the U.S. Department of Labor [32]. For example firefighters, a specific occupation within the ‘protective services’ category, may be very active when fighting a fire, but the majority of their time may be spent sedentary or in light intensity PA while waiting for a call. Despite the high levels of PA required to be an athlete, they were included in the occupational category ‘writers, artists, entertainers, and athletes’ which was previously categorized as “unclassifiable” due to the assumed lower levels of PA of the other occupations in the category. In our new grouping ‘writers, artists, entertainers, and athletes’ belongs to the low OA grouping, which would suggest that overall number of professional athletes sampled was low. These examples highlight some limitations of relying on job classification as an index of PA [43, 44]. Due to the uncertainty of PA levels of many occupational categories, researchers have predominantly limited their analyses to the occupational categories assumed to have more consistently very high or low levels of OA [12, 21, 45]. The ability to quantify the PA and SB of individuals working in a diverse spectrum of occupational categories with accelerometer data allows for more data-driven grouping into low, intermediate, and high OA. While our tertile approach is arbitrary, it logically splits the occupational categories into three equivalent OA groupings, and avoids having the majority of occupations being considered unclassifiable. Accounting for the variation in PA within each occupational category is still a challenge to overcome, however these data-driven OA groupings may enhance researchers ability to rely on occupational categories as an index of PA, and improve our capability to identify relationships between OA, daily PA and health [46].

In the past relying on occupational categories as a proxy for OA has be useful when gathering information on daily PA of various occupations that involve large amounts of sedentary time or greater amounts of physical labor [32, 46]. The use of objectively measured PA and SB provides data to support the use of occupational categories as a proxy for OA differences. Our study corroborates the utility of the original high and low OA groupings established by King et al. [32]. Only three of the 17 occupational categories previously classified as having high, or low OA were differently categorized using objectively measured PA and SB. ‘Laborers, except construction’ went from high to intermediate OA, ‘material recording, scheduling, and distributing clerks’ went from low to intermediate OA, and ‘motor vehicle operators’ went from low to high OA.

Due to the variety of occupations within some occupational categories PA levels may be quite heterogeneous within an occupational category. Also there is no information regarding the breakdown of all the occupations within each of the 40 occupational categories (i.e., there may have been only one athlete in the ‘writers, artists, entertainers, and athletes’ category). Even within the same occupation variability of PA can be large. For example, cross-country truck drivers spend considerably greater amounts of time sitting compared to local delivery truck drivers who spend a significant amount of time unloading [43]. Additionally, this analysis was not restricted to actual working hours due to the inability to isolate work from non-work time using the data as it was originally acquired. Information regarding regular working hours, shift work or alternative work schedules was not available. Because we were unable to account for variability in occupations within occupational categories, and unable to stratify PA and SB taking place during work or leisure time these limitations could result in non-differential exposure misclassification [47], and limits the ability to accurately interpret these data and ultimately make recommendations about changing behaviors within specific domains.

This study provides benchmark values for accelerometer-derived PA and SB variables, organized by occupation category, which will facilitate comparisons between and across studies using the same classification scheme. Creating tertiles of the summary score from all six accelerometer-derived variables to objectively classify the 40 occupational categories into low OA, intermediate OA, and high OA groupings does have limitations. In some cases, solely focusing on the summary score obscured interesting differences only apparent when closely considering the concurrent rankings of all accelerometer-derived variables between all occupational categories. For example, and consistent with previous research, some low OA occupational categories (e.g., ‘engineers, architects, and scientists’, ‘technicians and related support occupations’, ‘management related occupations’, ‘writers, artists, entertainers, and athletes’, and ‘executives, administrators, and managers’) associated with higher education (>80 % had more than a high school education) and higher income (>70 % earned more than 45K/ year) accumulated relatively higher amounts of MVPA, likely due to increased engagement in leisure time exercise [24, 25, 48]. The higher amount of MVPA relative to the lower summary score in the ‘protective service occupations’ (e.g., firefighters, police and sheriff’s patrol officers, fish and game wardens) was likely due to a combination of increased time spent in vigorous leisure time PA and OA [49]. Because we were unable to separate LTPA from OA, such implications are purely speculative. Although the tertile approach is data-driven, it was possible for two occupations with very similar summary scores to end up in separate OA groupings, for example, ‘motor vehicle operators’ with a summary score of 87 in the high OA group, and ‘supervisors and proprietors, sales occupations’ with a summary score of 89 in the intermediate OA group. Because there was no clear break in the categories, we made an arbitrary split between the 14th and 15th summary scores, despite the small difference in summary score.

In line with previous research, we showed that men were more active (inferred from higher values for TAC, activity counts/minute, and MVPA) than women [37, 40, 50, 51] within the same OA groupings after adjusting for characteristics known to be related to PA. In addition our data highlighted differences in PA of men and women within the same occupational category, and showed obvious differences in the proportion of men and women employed in specific occupational categories. For example, in our sample, only men were employed in many high OA occupational categories, while several low OA occupational categories were 100 % women. Due to gender differences in PA, the proportion of women in an occupational category may affect the results. The proportion of men and women employed as ‘engineers, architects, and scientists’, ‘technicians and related support occupations’, ‘management related occupations’, ‘executives, administrators, and managers’, protective services’, and ‘writers, artists, entertainers, and athletes’ may be another explanation for why these low OA occupational categories had higher than expected MVPA relative to their summary score. These six low OA occupational categories contained 60 % men, while the remaining eight low OA occupational categories comprised only 25 % men.

A number of additional analytical limitations to this analysis must be acknowledged. Operating under the assumption that OA is a major determinant of PA for many adults, the sample was restricted to adults working full-time, therefore the results may not be generalizable to populations employed part-time. Because of numerous exclusion criteria a large proportion of the adult population was excluded from the analysis, therefore the study population may not be entirely representative of the U.S. adult population. Of the 40 occupational categories, 16 were represented by less than 20 individuals. Therefore, the expected accelerometer-derived variables for some occupational categories should be interpreted with caution as they may not reflect national estimates of workers in these categories. We considered only reporting the accelerometer-derived variables for the 24 categories with a sample size greater than 20, as recommended by the NCHS [52]. However, despite these relatively small sample sizes, the majority of accelerometer-derived variables displayed a relative standard error (the standard error as a percent of the measure itself, much like the coefficient of variation) within the range deemed acceptable (<30 %) by the NCHS [42]. Thus, we opted for a more comprehensive presentation of all 40 occupational categories. We clearly indicated when violations to the relative standard error rule occurred and presented the sample size of each occupational category within the tables. Because of sample size limitations we chose to present a more conservative descriptive epidemiology of accelerometer-derived variables rather than conducting statistical comparisons between occupational categories, and men and women. However, showing the number of men and women in each category of the current sample was informative, especially in cases where the whole category was men, or exclusively women. As the number of women entering the work force has grown in the last half century, it would be useful to develop sex-specific estimates for OA in different occupational categories. Unfortunately, the occupational categories for NHANES 2005–2006 were different, so enlargement of the sample by incorporation of multiple waves of NHANES was not possible.

A major strength of this study was the use of accelerometers to objectively measure PA. At the time of data collection, uniaxial accelerometers were the method of choice, and weaknesses related to uniaxial, count-based PA measurement must be acknowledged. Accelerometers do not capture all types of PA, nor do they provide information on the type of PA performed, and their ability to accurately identify time spent in specific intensity categories has been questioned [53–55]. When worn at the waist accelerometers are most sensitive to ambulatory PA, and therefore the additional intensity of carrying loads, or other upper body movements is not captured [37]. This instrument has been validated against measured activity energy expenditure [56, 57], however it has not been validated for capturing the PA patterns characterizing OA in different occupations. The PA patterns characterizing OA (e.g., mostly sitting, and standing, with little walking, engaging in repetitive tasks) differ significantly from LTPA patterns which usually involve dynamic movements that engage large muscle groups resulting in increased whole-body metabolism and cardiac output, and are often emphasized for their the health-promoting capacity [16]. Considering the diverse movement requirements of different occupations (e.g., degree of static work, upper body work, standing, moving, lifting and loading occupations) [16], the observed differences in PA between the occupational categories may be more a reflection of the ambulatory movement captured by the uniaxial waist worn accelerometer than differences in other forms of OA. If carrying loads and upper body movements (e.g., food trays, pitchers/pots for ‘waiters and waitresses’ who had the largest proportion of light PA) were considered, the PA intensity may have been considerably higher.

The movement captured by the accelerometer reflects the accumulation of PA at home, in transit, during leisure time, in addition to time at work [37]. The differences in accelerometer-derived PA and SB variables between OA groupings occurred during traditional work hours (9 am–5 pm). Outside of traditional work hours OA groupings had similar levels of PA and SB. This supports our analytical assumption that the majority of workers were working a traditional day shift, and that the differences in accelerometer-derived PA and SB variables were likely due to OA. However, we must acknowledge our inability to separate PA and SB taking place during work and leisure time, and previous research which indicates that OA and LTPA is profoundly different in varying occupations. Blue collar workers with physically strenuous occupations and frequent overtime work are significantly less likely to engage in LTPA, while white collar workers may engage in greater LTPA, after a sedentary day at work [24, 25, 48].

An additional concern regarding waist worn, uniaxial accelerometers is their inability to accurately differentiate between sitting and standing postures [58–60], which can result in the misclassification of standing-light work, a light intensity PA, as sedentary [61–63]. Some occupations traditionally spend much of the work day in an upright posture, standing and/or moving around in light intensity behaviors (e.g., teachers, cooks, retail, waiters and waitresses). The increased energy expenditure and postural demands of standing compared with sitting may be an important distinction to consider when evaluating health outcomes [60, 64]. For assessing and differentiating between sitting and standing postures between different occupations, a thigh-mounted accelerometer like the ActivPAL monitor [65] or Actigraph [66], or triaxial accelerometer (ActiGraph GT3X+) at the thigh or hip may provide greater accuracy [67].

Objectively measured PA allowed for a more precise estimate of the amount of PA and SB associated with different occupational categories, and made it possible to systematically classify 40 different occupational categories into three distinct OA groupings. An individual’s occupational category appeared to have a great influence on daily PA and SB. This information provides new opportunities to explore the relationship between OA and health outcomes. It is also important for the design and implementation of programs and policies to improve health, productivity, and reduce work related injury of the employed population. Future occupational epidemiological research is needed to understand how variations in OA, LTPA, transportation, and household PA interact to influence engagement in beneficial amounts of daily PA, and influence health [17]. In addition, future research should continue to refine recommendations of maximal levels of OA, because strenuous OA can have health-detrimental impacts such as musculoskeletal disorders, decreased work ability and absence due to work related sickness or injury [68]. The results reported here highlight the scarcity of the data available for certain occupational categories to conduct and inform such studies. A larger sampling of underrepresented occupational categories, and a wide range of unique occupational categories would benefit future research concerned with the impact of occupation on PA and SB. The observed values presented in this paper are an important resource that should be expanded and refined as future changes occur in OA and as occupational categorization systems evolve.