Body Mass Index: Accounting for Full Time Sedentary Occupation and 24-Hr Self-Reported Time Use

Objectives We used linked existing data from the 2006–2008 American Time Use Survey (ATUS), the Current Population Survey (CPS, a federal survey that provides on-going U.S. vital statistics, including employment rates) and self-reported body mass index (BMI) to answer: How does BMI vary across full time occupations dichotomized as sedentary/non-sedentary, accounting for time spent in sleep, other sedentary behaviors, and light, moderate, and vigorous intensity activities? Methods We classified time spent engaged at a primary job (sedentary or non-sedentary), sleep, and other non-work, non-sleep intensity-defined behaviors, specifically, sedentary behavior, light, moderate, and vigorous intensity activities. Age groups were defined by 20–29, 30–39, 40–49, and 50–64 years. BMI groups were defined by 18.5–24.9, 25.0–27.4, 27.5–29.9, 30.0–34.9, and ≥35.0 kg/m2. Logistic and linear regression were used to examine the association between BMI and employment in a sedentary occupation, considering time spent in sleep, other non-work time spent in sedentary behaviors, and light, moderate, and vigorous intensity activities, sex, age race/ethnicity, and household income. Results The analysis data set comprised 4,092 non-pregnant, non-underweight individuals 20–64 years of age who also reported working more than 7 hours at their primary jobs on their designated time use reporting day. Logistic and linear regression analyses failed to reveal any associations between BMI and the sedentary/non-sedentary occupation dichotomy considering time spent in sleep, other non-work time spent in sedentary behaviors, and light, moderate, and vigorous intensity activities, sex, age, race/ethnicity, and household income. Conclusions We found no evidence of a relationship between self-reported full time sedentary occupation classification and BMI after accounting for sex, age, race/ethnicity, and household income and 24-hours of time use including non-work related physical activity and sedentary behaviors. The various sources of error associated with self-report methods and assignment of generalized activity and occupational intensity categories could compound to obscure any real relationships.


Introduction
Questionnaires typically capture domain-specific activity (e.g., leisure time physical activity, occupational activity) which presumes that these are the only activities worth tracking. Objective monitoring protocols typically direct respondents to take body worn sensors off at night [1] and there is evidence that they are frequently taken off early and/or put on late [2], resulting in lengthy segments of unaccounted time. Until now, an opportunity to account for activities during the entire 24-hour day has been missing from large-scale surveys.
The American Time Use Survey (ATUS) queries primary activities performed over the course of a 24-hour day (1,440 minutes) and allows for a full accounting of time spent in work, sleep, and other non-work, non-sleep behaviors. We have produced a crosswalk to assign metabolic equivalent (MET; 1 MET is the metabolic cost of quietly resting or >3.5 mL of oxygen uptake per kg body weight per minute) values to these categorized behaviors [3], specifically identifying each ATUS primary activity as a sedentary behavior, or a light, moderate, or vigorous intensity activity. This process included linking summary MET values to generalized occupational categories in an attempt to better characterize the intensity of time originally captured simplistically as ''at work'' in the ATUS. Of particular interest is the fact that sedentary occupations (i.e., those characterized by much sitting) have a major bearing on objectively monitored physical activity levels [4]. We have previously shown that workers employed in full time sedentary occupations were actually sedentary for approximately 11 hours per day, leaving little time to achieve recommended levels of physical activity for overall health [5]. Since the prevalence of sedentary occupations (and associated weight gain) has increased in recent decades [6], there is a growing interest in examining the impact of sedentary occupations on body habitus, taking into account behaviors enacted outside of working hours, especially for full time workers with limited personal time [7,8,9].
There are only a finite number of minutes in a day and the different obligations and choices about how to spend each minute likely have a complex and profound impact on BMI. The final piece of the ''BMI-time use'' puzzle was potentially provided when, from 2006 to 2008, the ATUS rolled out an ''Eating & Health'' (EH) module which included a query of self-reported height and weight, necessary for computing BMI. Therefore, this analysis combined the 2006-2008 ATUS with corresponding occupational codes from the Current Population Survey (CPS, a federal survey that provides on-going U.S. vital statistics, including employment rates) and self-reported BMI from the EH module to answer the following question: How does BMI vary across full time occupations dichotomized as sedentary/non-sedentary, with and without accounting for time spent in sleep, other sedentary behaviors, and light, moderate, and vigorous intensity activities?

Source data sets
The publically available 2006-2008 ATUS data along with the corresponding CPS and EH module data were downloaded, extracted, and linked (using data source specified identifiers). Details about the CPS are located at and those related to the ATUS and EH are at. A brief description of each source data set follows. A subsample of individuals (age 15 and older) drawn from households that completed the CPS in the preceding 2-5 months formed the ATUS response sample. Individuals selected from the CPS for the ATUS are routinely interviewed using a computerassisted telephone-interviewing (CATI) system on a designated day about their use of time over the previous 24-hour day. Subsequently, each reported activity is assigned a 6-digit code guided by software that organizes and broadly classified activities into more specific categories using a hierarchical 2-digit tier system. As indicated above, we have exploited this tier-based system to assign MET values to each category [3] and have published the operative SAS program at http://riskfactor.cancer. gov/tools/atus-met/. ATUS respondents routinely report time worked at their ''primary job,'' however, the linked ''primary occupation'' that they are employed in must be retrieved using links to their corresponding and previously collected CPS data. To be clear, ATUS captures time spent at primary jobs and these can be linked to CPS records of primary occupations. We used the summary MET values for reported occupations [3] and also the more detailed values [10] available at http://riskfactor.cancer. gov/tools/ocs-met/. We have previously reported that application of the more detailed values reduces the prevalence of reported sedentary occupations and increases those with more vigorous intensity occupations. The summary values are considered a more conservative approach to estimating overall energy expenditure of occupational categories [10]. The corresponding EH module (details available at http://ers.usda.gov/data-products/eatingand-health-module-(atus) and http://www.bls.gov/tus/) was only deployed from 2006-2008, thus defining the survey window (and eligible sample size) for the analysis of these combined data sets.

Sample description
There were 20,637 respondents to the 2006-2008 EH module of the ATUS. The analysis sample included individuals who were 20-64 years of age (inclusively), had a CPS occupation code assigned to their primary job, reported working more than 7 hours (a proxy indicator of full time work [10]) at their primary jobs on their designated time use reporting day, and had replete height and weight fields needed to calculate BMI values. Data were excluded from respondents who reported that they were pregnant, had a BMI,18.5 kg/m 2 or incomplete household income information, or for whom the original interviewer questioned the reliability of the interview immediately following its completion. Figure 1 is a flow chart summarizing the number of survey respondents excluded with implementation of each decision rule culminating in the final analytical data set of 4,092 individuals.
Reported activities were classified by time spent engaged in work at their primary job (identified by ATUS tiered code 050101), sleep (identified by ATUS tiered code 010101, 010102 and 010199), and other non-work, non-sleep intensity-defined behaviors, specifically, sedentary behavior (identified by 1.0 to 1.5 METs) [10,11,12], light-(1.6 to 2.9 METs) [5], moderate-(3.0 to 5.9 METs) [10,13], and vigorous-intensity ($6.0 METs activities) [10,13]. Undefined time was otherwise aggregated as unaccounted time. Together, all time-based variables added up to account for the full 24-hour (1,440 minutes) period. However, since only subsamples of the population are known to engage in any amount of moderate and vigorous intensity activities [1,5], and resulting estimates are thus distorted by an overwhelming number of respondents who do none of these types of activity [5], we also consolidated all intensities (light through vigorous) of non-work physical activities.
CPS-reported occupations and linked MET values [5] were used to categorize respondents and further classify time spent working at primary jobs as sedentary occupations or nonsedentary occupations. A ''sedentary occupation'' (e.g., one typically characterized by the universe of office-type administrative work) is not precisely the same as a ''sedentary behavior'' (stoically sitting, e.g., seated reading), and likely includes at least some light activity (e.g., walking to department meetings, filing, standing to greet office visitors, etc.). Since the occupational MET crosswalk [10] considered stereotypical workday body position (sit, stand, walk, heavy labor) and intensity (light, moderate, vigorous) in the process of allocating summary MET values, the rare floor value possible was 1.5 METs. Less restrictively, we have previously identified ''sedentary occupations'' as those assigned values of

Analysis
Constructed sampling weights were used by following the online documentation (http://www.bls.gov/tus/ehdatafiles.htm) to ensure representativeness of the U.S. non-institutionalized civilian population. Descriptive analysis included means (6 standard error, SE) and frequencies (weighted %), as appropriate, for descriptive characteristics (sex, age, race/ethnicity, BMI and household income), employment in sedentary occupation (yes or  no), and reported time spent in work, sleep, sedentary behaviors, and light, moderate, and vigorous intensity activities (these last three separately and combined as a ''total non-work physical activity'' category). We tested for differences in all time-based variables by sex and dichotomy of employment in sedentary/nonsedentary occupations. Finally, we employed logistic and linear regression to examine the association between BMI and employment in a sedentary occupation, accounting for time spent in sleep, other non-work time spent in sedentary behaviors, and light, moderate, and vigorous intensity activities, sex, age, race/ethnicity and household income. All statistical analyses were conducted using SAS 9.3 (SAS institute, N.C.).

Results
Since none of the findings reported below differed between occupation intensity categories classified using either summary or detailed approaches to linking MET values [10], we present the results using only the summary values. Descriptive characteristics of the analysis sample (total sample and stratified by sex) are presented in Table 1 and Table 2. Despite apparent mean differences (e.g., men working longer than women, women engaged in more light intensity physical activity than men), there were no statistically significant differences between men and women (i.e., no p,0.05) in any of the time-based variables.
The . There were no statistically significant differences in age or BMI by sedentary/non-sedentary occupation dichotomy for either men or women. Likewise, there were no statistically significant differences in any of the time-based variables by sedentary/non-sedentary occupation dichotomy for men or women ( Table 3). The weighted percent of the population and their mean BMI are presented by sedentary/non-sedentary occupation dichotomy and age groups for men ( Figure 2) and women ( Figure 3). Time spent in sleep, in other non-work sedentary behaviors, and in all non-work physical activities (light-, moderate-, and vigorous-intensity combined) are also presented by sedentary/non-sedentary occupation dichotomy for men ( Figure 4) and women ( Figure 5).
Logistic and linear regression analyses failed to reveal any associations between BMI and the sedentary/non-sedentary occupation dichotomy considering time spent in sleep, other non-work time spent in sedentary behaviors, and light, moderate, vigorous and undefined intensity activities, sex, age, race/ethnicity and household income (Tables 4 and 5).

Discussion
Previous surveys of Polish [14] (n = 508) and Spanish [15] (n = 12,044) workers did not find any significant associations between occupational energy expenditure and BMI after accounting for time spent in leisure time physical activities. Limiting survey design and subsequent analysis only to self-reported sedentary work and leisure-time physical activity implies that these two behaviors are the only variables worth tracking in relation to BMI. It stands to reason, however, that a complete inventory of obligatory and discretionary behaviors accumulated throughout the day interact to contribute to energy balance. In this analysis of the merged ATUS, CPS, and EH data sets, however, we found no evidence of a relationship between selfreported full time sedentary occupation classification and BMI after accounting for sex, age, race/ethnicity, household income, and 24-hours of time use including non-work related physical activity and sedentary behaviors.
Zick et al. [16] previously used the 2006-2007 ATUS and associated EH modules to conduct a multivariable analysis of the time use relationship with BMI. Although it is possible to account for all behaviors in the full day using these ATUS data as we have done, these authors applied our own MET classification [3] but focused only on time (specifically bouts of $10 minutes) accumulated over 3.3 METs as an indicator of activity, and only television/video viewing and time spent sleeping as indicators of inactivity. Further, although ATUS captures respondents' report- Table 4. Linear regression analysis investigating the association between BMI and occupation intensity (sedentary/non-sedentary) of primary job (n = 4,092). ed time spent at work, these authors' analytical treatment of this behavior was limited to the application of a dummy variable only in the case of males employed in occupations that were considered representative of work intensities greater than 3.3 METs. The authors did not seem to have considered the amount of time spent at work (i.e., full time workers) or the reduced energy expenditure associated with sedentary occupations, although these are much more common than moderate-to-vigorous intensity occupations [10]. Overall, their data reduction and analytical approach implies preconceived valuation of factors related to BMI, as opposed to a more inductive approach to identifying emergent factors from amongst the data. Our own approach was more exploratory, querying how BMI varies across full time occupations categorized as sedentary/non-sedentary, with and without accounting for a full day's contingent of time spent in sleep, other sedentary behaviors, and light, moderate, and vigorous intensity activities. The fact that we did not see a clear relationship between BMI and sedentary/ non-sedentary occupations suggests that occupation alone is not a defining factor in shaping this health indicator. Our study supports a review that concluded that at this time there is limited evidence to substantiate a positive relationship between occupational sitting and health risks [17].
There are alternative explanations, however, for why we did not find such evidence. The ATUS, CPS, and EH are self-reported surveys, and although time use surveys have been validated against accelerometers [18], the assignment of MET codes representing intensity of reported behaviors and simplified occupational classification is admittedly inexact [3,10]. Further, there is a tendency for adults to under-report weight and over-report height (both necessary to compute BMI) [19], although the vagaries in self-reported height and weight are likely more complicated than this generalized statement of reporting bias implies [20,21]. That said, as an illustrative example of the potential for self-report bias to affect BMI measures, the prevalence of obesity (derived from self-reported height and weight) among women depicted herein (22.6%) is dramatically lower than nationally representative estimates of objectively measured obesity among adults ($20 years) during the 2005-2006 and 2007-2008 National Health and Nutrition Examination Survey (35.3% and 35.5%, respectively) [22]. Such a large discrepancy between U.S. nationally representative estimates of self-reported and objectively measured obesity highlights the incongruity between these measures and the potential for self-report bias to introduce substantial amounts of measurement error. The EH module also queries time spent eating as a primary and secondary behavior, but this is not the same as querying what was consumed in terms of energy intake, so it is possible also that energy intake systematically differs in some way between occupational categories and confounds relationships with BMI. All told, it is plausible that these various sources of error could compound to obscure any real relationships.
In summary, we found no evident relationship between selfreported full time sedentary occupation classification and BMI after accounting for time spent in non-work related physical activity and sedentary behaviors, sex, age, race/ethnicity, and household income. Future analyses should focus on 24-hour objectively monitored behavior and directly assessed BMI.