Article Text
Abstract
Objective To determine the number of latent body mass index (BMI) trajectories from 1994 to 2010 among working Canadians and their association with concurrent trajectories in work environment exposures.
Methods Data of employed individuals from the longitudinal Canadian National Population Health Survey were used. Group-based trajectory modelling was used to determine the number of latent BMI trajectories and concurrent psychosocial work environment trajectories. A multinomial logistic regression of BMI trajectory membership on trajectories in work environment dimensions (skill discretion, decision latitude, psychological demands, job insecurity, social support, physical exertion) was then explored.
Results Four latent BMI trajectories corresponding to normal, overweight, obese and very obese BMI values were found. Each trajectory saw an increase in BMI (~2–4 kg/m2) over the 17-year period. A higher decision authority trajectory was associated with lower odds of belonging to the overweight and obese trajectories when compared with the normal weight trajectory. A decreasing physical exertion trajectory was associated with higher odds of belonging to the very obese trajectory when compared with the normal weight trajectory.
Conclusions Four BMI trajectories are present in the Canadian workforce; all trajectories saw increased body weight over time. Declining physical exertion and lower decision authority in the work environment over time is associated with increased likelihood of being in overweight and obese trajectories.
- epidemiology
- public health
- longitudinal studies
Statistics from Altmetric.com
Key messages
What is already known about this subject?
Previous literature is conflicted about if, and what physical exertion and psychosocial–related work environment factors influence body mass index (BMI) and unhealthy weight gain.
The inconclusion in prior evidence may be due to methodological limitations, such as only exploring one measurement of the work environment and using traditional BMI cut points over only one to two follow-up periods.
What are the new findings?
Using nine assessments of BMI over 17 years, this is the first study to determine that four, distinct latent BMI trajectories—corresponding to normal weight, overweight, obese and very obese—exist among a representative sample of a North American workforce.
All BMI trajectories had a BMI increase of at least 1.62 kg/m2 over the follow-up, with the greatest increases in BMI seen in the obese groups.
Lower decision authority and decreasing physical exertion were associated with belonging to overweight and obese trajectories.
Ensuring healthy body weight among the working-aged population is a critical public health concern. Heavier body weight and poor weight maintenance are negatively associated with work productivity, earnings and employment.1 2 Employees with an obese body mass index (BMI, >30 kg/m2)3 are more likely to have sickness absence, require work disability and have higher healthcare expenditures, resulting in more expensive health benefit plans for employers.4 At the macro level, obesity is associated with a myriad of physical and mental chronic conditions, resulting in a workforce with suboptimal health and a loss in economic human capital.4 5
Much research has explored if the work environment is associated with unhealthy weight gain.1 A negative work environment may increase stress, activate the hypothalamic–pituitary–adrenal axis, and result in increased cortisol and metabolic changes leading to increased fat storage.6 7 A negative work environment may also be associated with reduced physical activity, less sleep, and consuming a higher calorie and less nutrient dense diet, which may potentially lead to weight gain.7
The psychosocial and physical demand aspects of the work environment have been thought to be associated with BMI. Two systematic reviews have explored the association between the psychosocial work environment dimensions and weight gain.7 8 The first review found weak associations between specific aspects of the work environment (high job demands, job control, job strain, psychological demands) and obesity,7 while the second review found job strain was not associated with obesity or weight gain.8 In addition, literature has suggested that certain aspects of the physical work environment (eg, hazardous exposures and sedentary work) may be associated with weight gain and overall physical functioning.9–11
However, previous literature has important methodological limitations. Longitudinal studies typically explore the association between dimensions of the work environment and change in body weight defined by traditional BMI cut points3 over two time points.12–15 Despite some studies boasting follow-up periods of 16–31 years,16 17 measuring body weight or the work environment at two time points limits researchers from determining nuances of how BMI changes over time, or how changes in the work environment over the same time period may be associated with changes in BMI. This limitation can be overcome by collecting information on work stress and BMI over multiple time points and using person-centred analytical methods such as latent trajectory modelling to determine how individuals in the workforce maintain their weight over time and how trajectories of working conditions are associated with weight trajectories; this evidence can then be used to inform workplace weight management policies and interventions.
Therefore, the purpose of this study was to determine the number of latent BMI trajectories of working men and women, measured at multiple time points over a 17-year period, and then determine how psychosocial and physical work environment trajectories were associated with latent BMI trajectories.
Methods
The longitudinal component of the National Population Health Survey (NPHS) was used for this study. The NPHS was a 17-year national longitudinal survey aimed at determining the health and health behaviours of Canadians. The NPHS aimed to be a nationally representative survey by employing a complex population-based survey design; further details can be found elsewhere.18 Cycles occurred every 2 years (1994–2010). The first cycle of the NPHS used a two-stage sampling design of clusters and dwellings to survey at least 1200 homes in each province, resulting in a baseline sample of 17 276.18 Over time, the sample decreased slightly due to participation refusal, non-response and participant death.18
This study restricted the NPHS sample to those who were at least 18 years of age, employed, working at least 15 hours per week and were not self-employed in 1994 (n=6407). The analytical sample included participants with at least two responses on BMI between 1994 and 2010 and who had complete information on work environment factors and covariates in 1994 (n=5455). Compared with participants included in the study sample (n=5455), participants not in the analytical sample (n=952) were statistically more likely to be born in Canada, male, working full time, living in Quebec, have older age and have lower levels of education in 1994.
Body mass index
Participants were asked their height and weight in each of the nine cycles which was used to compute BMI (weight (kg)/height (m2)). Given that men in Canadian population surveys are known to overestimate their height and women underestimate their weight, we applied a correction factor to BMI estimates.19 For men, BMI was calculated as −1.08+1.08×self-reported BMI; for women, BMI was calculated as −0.12+1.05×self-reported BMI.19 BMI was truncated at the 99th percentile across cycles to aid model convergence (~40–49 kg/m2).
Work environment dimensions
An abbreviated measure of Karasek and Theorell’s Job Content Questionnaire was included in the 1994 NPHS cycle, and in cycles between 2000 and 2010. Participants were asked 12 questions on their perception of five psychosocial and one physical demand work environment factors: skill discretion (your job requires that you: learn new things; have high level of skill; do things over and over), decision authority (your job allows you: freedom to decide how you do your job; to have a lot to say about what happens in your job), psychological demands (your job is very hectic; you are free from conflicting demands others make), job insecurity (your job security is good), physical exertion (your job requires a lot of physical effort) and workplace social support (you are exposed to hostility or conflict from the people you work with; your supervisor is helpful in getting the job done; the people you work with are helpful in getting the job done). Questions were answered on a 5-point Likert scale (ranging from “strongly agree” to “strongly disagree”) and combined to derive the psychosocial working factors. Given that abbreviated scales were used to capture broad concepts, the internal consistency of these scales was relatively low (α=0.34–0.61).20
Covariates
Sex, age, race, country of birth, province of residence (Atlantic provinces, Quebec, Ontario, Pacific provinces), rural/urban living location, marital status, dependent status and highest level of education were included as demographic covariates in analyses. Presence of a health condition that restricted the ability to go to work, school or be involved in the community was also included. Employment covariates included normal shift schedule (regular, rotating, irregular/on call/other) and full-time versus part-time employment. Covariate response patterns were relatively stable over time; therefore, all covariates were treated as time invariant and measured in the 1994 cycle.
Table 1 highlights the sample’s descriptive characteristics. In 1994, the average age was approximately 37 years. Roughly 53% of participants were male, 41% were living in Ontario, 46% had a high-school education and most were working full time (85%).
Although some covariates have been suggested to be effect modifiers (eg, sex, age, education),7 21 statistical interaction effects were not seen to be present for the association between psychosocial work factors and BMI trajectory groups in the data.
Analysis
Group-based trajectory modelling (GBTM) was used to determine the number of latent BMI and work environment factor trajectories.22 Model building was iterative where we explored one to five latent classes with an intercept, slope and quadratic slope with fixed variances. A full-information maximum-likelihood approach was used in models, which incorporated participants into analysis so long as they had at least two BMI assessments.
Models were assessed with statistical criteria using log-likelihood, Akaike information criterion (AIC), Bayesian information criterion (BIC), entropy values and the Lo-Mendell-Rubin (LMR) likelihood-ratio test. The smallest class size and alignment of classes to previous literature23–26 were also considered. In instances where statistical criteria alone could not suggest a clear model, study authors discussed and decided on the most appropriate model to present. Once completed, we quantified what latent BMI trajectory each participant belonged to using participants’ highest probability of belonging to a specific latent BMI trajectory.
Latent trajectories for work environment factors in this population have been described in a previous study.27 The trajectory models for each dimension included the following: a three-class model (low, medium, high) for skill discretion; a two-class model (low, high) for decision authority; a two-class model for psychological demands (low, high); a four-class model for job insecurity (high job insecurity, decreasing job insecurity, low job insecurity, increasing job insecurity); a four-class model for physical exertion (high physical exertion, increasing physical exertion, low physical exertion, decreasing physical exertion); and a four-class model for social support (low, mid-level, high, increasing social support).27 In building these trajectories, a two-class model was also supported for social support; however, we found the four-class model had predictive utility and was also supported by model fit indices. The four-class model was also supported for decision authority; however, as results did not differ in this model, we selected to present the most parsimonious two-trajectory model. Visuals of these trajectories are included in online supplementary figures S2–S7.
Supplemental material
We regressed participants’ most likely BMI trajectory class on participants’ most likely work environment trajectories, along with all covariates using a multinomial logistic regression. Work trajectories were conceptualised as categorical variables. It should be noted that each work environment dimension has a complex relationship with other dimensions; therefore, including only one dimension in the model may overestimate its effect, while including all dimensions in the model may underestimate its effect. In this manuscript, we have presented models where only one work environment dimension is included (along with covariates). Comparable results were observed in the model including all work environment dimensions (online supplementary table S5).
For each analysis, a sampling weight was used to account for initial probability of selection and non-response to the 1994 cycle of the NPHS. A bootstrap procedure using 500 replications for each model was used to account for the clustered survey design of the NPHS. Data preparation and multinomial logistic regressions were conducted in SAS V.9.4 (SAS Institute, Cary, NC, USA), while trajectory building was completed in Mplus V.7 (Muthén & Muthén, Los Angeles, CA, USA).
Results
Latent BMI trajectories
The four-class BMI model was determined to be the most appropriate model since it had a significant LMR test statistic, high entropy, and the second-lowest AIC and BIC values (online supplementary table S4). Although the five-class model had a slightly lower AIC and BIC value, the LMR test was not statistically significant, suggesting it did not provide more information about the sample when compared with the four-class model.
The first class (n=1719) was composed of individuals with a BMI of 22 kg/m2 in 1994 which increased to roughly 24 kg/m2 in 2010 (figure 1). The second and largest class (n=2243) included individuals with a BMI of 26 kg/m2 in 1994 which increased to 28 kg/m2 in 2010. The third class (n=1216) was composed of individuals with a BMI of 30 kg/m2 in 1994 which increased to 33 kg/m2 in 2010. The final and smallest class (n=277) included individuals who had a BMI of 36 kg/m2 in 1994 which increased to roughly 40 kg/m2 in 2010. These groups will be referred to as normal weight, overweight, obese and very obese, respectively. A gradient trend was seen in how BMI increased over the 17-year period (figure 1): the normal BMI group had an increase in BMI of ~1.62 kg/m2 over the 17 years, the overweight group had an increase in BMI ~2.13 kg/m2, the obese group increased by ~3.06 kg/m2 and the very obese group increased by ~4.29 kg/m2. Demographic information of each latent BMI class may be seen in table 1.
Association between latent BMI and work environment dimensions
Using χ2 tests of association, statistically significant differences in participant proportions between latent work environment and BMI trajectories were seen for decision authority and physical exertion (table 2). When compared with the other BMI trajectories, the overweight trajectory had slightly more participants belonging to the low decision authority trajectory (77%). The very obese trajectory group had a lower proportion of participants in the high physical exertion trajectory (eg, 26% vs 29% in normal BMI group) and higher proportion of participations in the decreasing physical exertion trajectory (eg, 22% vs 16% in the normal BMI group) when compared with the other BMI trajectory groups.
Table 3 presents the results of the multinomial logistic regression of BMI trajectory group on work environment trajectories and covariates using the lowest BMI group as the reference. When compared with the lowest BMI trajectory, it was seen that a higher decision authority trajectory was associated with lower odds of belonging to the overweight (OR 0.82, 95% CI 0.67 to 1.00) and obese trajectories (OR 0.68, 95% CI 0.54 to 0.86). A similar but non-statistically significant effect size was seen for the very obese group. When compared with low physical exertion at work, decreasing physical exertion at work was associated with a statistically higher odds of being in the very obese group (OR 1.74, 95% CI 1.13 to 2.68) and a marginally statistical higher odds of being in the obese group (OR 1.34, 95% CI 0.98 to 1.82). Lastly, mid-level social support compared with high social support was associated with higher odds of belonging to the overweight trajectory (OR 1.26, 95% CI 1.02 to 1.56). Similar results were seen when all work stressors were included in the same model (online supplementary table S5).
When using the overweight trajectory as the reference (online supplementary table S6), decreasing (compared with low) physical exertion was associated with higher odds of being in the very obese trajectory (OR 1.58, 95% CI 1.02 to 2.44). No other statistically significant associations were seen when using the obese trajectory as the reference (online supplementary table S7).
Discussion
The purpose of this study was to determine the number of latent BMI trajectories among participants in the Canadian workforce, and if latent trajectories in the psychosocial and physical demand work environments were associated with specific BMI trajectories over 17 years. Our study produced three main findings: identifying four latent BMI trajectories among the Canadian workforce, differences between trajectories and how they all increase over time, and how certain aspects of the work environment, specifically decision authority and physical demands, are associated with unfavourable BMI trajectories. These findings have important implications given the high population health costs of obesity; the indirect costs of overweight and obesity in part due to lost productivity have been suggested to be higher than direct healthcare costs (~$C16.4 billion vs $C6.9 billion).28 29
The four latent BMI trajectories identified began with a BMI in 1994 of roughly 22, 26, 30 and 36 kg/m2, aligning to categories of normal weight, overweight, obese and very obese.3 Only 32% of the sample was in the normal weight BMI group, suggesting that a large portion of the workforce is overweight. Although there may be misclassification in this estimate—given that participants were placed into BMI trajectories based on their highest probability of belonging to a trajectory—this still leaves roughly 27% of the sample belonging to a trajectory corresponding with an obese BMI; this is slightly higher than previous Canadian workforce prevalence estimates (~21%),23 but is in line with previous estimates from the US workforce (~26%).24
It is thought that as we age, BMI increases.25 All latent classes gained weight over time at a rate of ~2–4 kg/m2 across the 17-year follow-up. It has been reported that the average BMI in high-income countries has increased by 1.00 kg/m2 per decade.30 At a crude level, our findings suggest that the average increase in BMI among the Canadian workforce is slightly higher, at approximately ~1.34 kg/m2 per decade. Given that it is important for labour forces to be composed of healthy workers and that a gradient trend has been seen between increasing body weight and reduced productivity,28 this finding is important to reinforce focusing on weight management in working populations.
Lower education levels, occupational prestige and income are associated with greater obesity risk31; although effect modification of these factors was not a central focus of our study, it was seen that the very obese group had the highest proportion of participants with less than a high school education (22%) and working an irregular schedule. Approximately 21% of respondents in the normal weight BMI group were employed part time (working less than 30 hours per week); this perhaps provided these participants more opportunity to be physically active outside of their working hours. In regression results, the inclusion of these factors did not impede associations between the work environment and BMI trajectories.
Our study suggests that lower decision authority and decreasing physical exertion at work were associated with belonging to overweight and obese BMI trajectories. Previous research has suggested that individuals in jobs with high decision authority makes workers more active at work, or that these jobs are held by individuals with high socioeconomic status, which is associated with lower BMI.32 A previous study by Choi and colleagues (2010) found that low physical activity at work was a significant predictor of obesity, even while adjusting for low physical activity and overeating outside of work.33
We observed that the middle social support trajectory was associated with being in the overweight BMI trajectory group (compared with the normal BMI trajectory), with a similar sized estimate also observed for the increasing social support trajectory. However, association patterns between social support trajectory levels and BMI trajectories were less consistent than for decision authority and physical exertion. In addition, the two-class model of social support did not show statistically significant associations with BMI trajectories (online supplementary table S8). While some studies suggest that workplace social support may improve weight management behaviours (eg, physical activity, fruit and vegetable consumption),34 the statistically significant observation in this study may be spurious.
There may be a few reasons as to why the other work environment dimensions were not predictive of BMI trajectory membership. Individuals with very high BMI as well as those who are employed in very stressful work environments may be pushed out of the workforce, leading to a healthy worker selection bias.7 This selection may result in less variability in BMI and work stressors leading to insignificant findings. In addition, it has been suggested that other workplace characteristics associated with eating and physical activity behaviours perhaps are more strongly associated with weight management among the workforce compared with the work factors in this study, such as workplace food options (eg, proximity to fast food restaurants, number of colleagues who eat unhealthy food when at work), workplace distance from a fitness facility or the ability to walk to the workplace.21 35
The only covariate showing a statistically significant association with BMI trajectory membership was sex in multinomial regression models. Being female was associated with membership in the heavier BMI trajectories. Despite the fact that women tend to have higher body fat percentages (resulting in higher BMI levels), it is also known that women partake in more hours of unpaid work,36 potentially influencing their physical activity and dietary choices. However, effect modification by sex was not present in our findings.
Limitations
All measures were self-reported, including BMI. Although we attempted to correct BMI estimates,19 there is a risk of measurement bias in our results which may underestimate the BMI groups. However, the extent to which this measurement bias changes over time within individuals—thereby impacting our trajectory modelling—is unknown. Second, the abbreviated nature of the psychosocial work environment scales may also introduce a non-differential measurement bias into associations, potentially underestimating findings. The job insecurity and social support work constructs were slightly skewed (online supplementary table S9), potentially overestimating the number of latent classes found. Some work environment factors (eg, physical exertion) consisted of one question, limiting our understanding of these work environment constructs. Certain factors (eg, job insecurity, physical exertion) were based off of one Likert scale question which may not have met the normality assumption of the GBTM. Participants in poor physical health may have interpreted work environment questions differently, potentially introducing misclassification into results.
Third, in attempt to triangulate study findings to take into account the complex relationship between psychosocial work factors and their influence on BMI, we explored multinomial models including only one work exposure, models including all work exposures and models where the obese trajectory group was used as the reference. This may increase the risk of type I error. Lastly, the cohort ended in 2010; since this time, changes in the workplace and factors related to body weight at the population level have occurred. Therefore, we recommend that this study be replicated in more recent national and occupational cohorts.
Conclusions
This is the first study to determine the number of latent BMI trajectories among a population-based sample of the Canadian workforce, providing critical longitudinal information for workplace weight management policies. Our findings suggest that four distinct BMI trajectories—normal weight, overweight, obese and very obese—exist among workforce participants. All groups had a BMI increase of at least 1.62 kg/m2 over the follow-up. Lower decision authority and decreasing physical exertion were associated with belonging to overweight and obese trajectories.
As all BMI trajectories saw increasing BMI, our findings reinforce that weight management is important to everyone in the workforce; given that decision authority and physical exertion at work were associated with overweight BMI trajectories, these may be factors that workplaces want to incorporate into interventions aimed at improving physical health. Future research should attempt to replicate our findings in more recent and diverse occupational samples, use more advanced trajectory modelling methods—such as joint or multi-trajectory modelling—to determine profiles of workers based on their physical health and work environments, explore how the dietary and physical activity workplace intervention environments influence weight management, and identify whether these interventions interact with the psychosocial and physical work environments to influence weight management.
References
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Footnotes
Contributors KD, MG-O, PMS and CM conceived the research question, study design, and revised and approved this manuscript. KD analysed the data and drafted the initial manuscript. All authors participated in approving the final version to be published and agreeing to be accountable for all aspects of the work by ensuring that questions related to the accuracy or integrity of any part of the work were appropriately investigated and resolved.
Funding This work was supported by a project grant from the Canadian Institutes for Health Research (CIHR) (grant no. 310898). PMS is supported through a Research Chair in Gender, Work and Health from CIHR. KD is supported through a doctoral scholarship through CIHR. MG-O is supported through a CIHR postdoctoral fellowship. Access to the data for this paper was enabled through Statistics Canada’s Research Data Centre at the University of Toronto.
Competing interests None declared.
Patient consent for publication Not required.
Ethics approval This study was approved by the University of Toronto.
Provenance and peer review Not commissioned; externally peer reviewed.
Data availability statement The data for this study were accessed through the Canadian Research Data Centre Network.