Article Text

Download PDFPDF

Original article
Association between dimensions of the psychosocial and physical work environment and latent smoking trajectories: a 16-year cohort study of the Canadian workforce
  1. Kathleen G Dobson1,2,
  2. Mahée Gilbert-Ouimet1,3,
  3. Cameron A Mustard1,2,
  4. Peter M Smith1,2,4
  1. 1 Institute for Work & Health, Toronto, Ontario, Canada
  2. 2 Dalla Lana School of Public Health, University of Toronto, Toronto, Ontario, Canada
  3. 3 Centre de Recherche du CHU de Québec, Quebec, Canada
  4. 4 Department of Epidemiology and Preventive Medicine, Monash University, Melbourne, Victoria, Australia
  1. Correspondence to Ms Kathleen G Dobson, Institute for Work & Health, Toronto, ON M5G 2E9, Canada; kathleen.dobson{at}


Background This study aimed to determine the number of latent smoking trajectories among Canadians employed in the workforce over a 16-year period, and if latent trajectories in dimensions of the physical and psychosocial work environment were associated with specific smoking trajectories.

Methods We studied 5461 employed adults from the longitudinal Canadian National Population Health Survey. Daily cigarette consumption was measured biannually from 1994 to 2010. Work environment factors (skill discretion, decision authority, psychological demands, job insecurity, physical exertion and workplace social support) were measured in 1994 and then from 2000 to 2010 using an abbreviated form of the Job Content Questionnaire. Smoking and work environment trajectories were derived using group-based trajectory modelling. Associations between work environment trajectory classes and smoking trajectory classes were estimated using multinomial logistic regression.

Results Four latent smoking trajectories were seen: non-smokers; ceasing smokers (consuming ~14 cigarettes/day in 1994 and 0 in 2008–2010); smokers (consuming ~7 cigarettes/day between 1994 and 2010); and heavy smokers (consuming ~22 cigarettes/day in 1994 and ~14 in 2010). Lower skill discretion, high psychological demands, high physical exertion and low social support trajectories were associated with membership in the heavy smoking trajectory compared with the non-smoking trajectory. Low decision authority, high psychological demands and high physical exertion trajectories were associated with membership in the ceasing compared with the non-smoking trajectory.

Conclusions Certain physical and psychosocial work environment trajectories were associated with heavy and ceasing smoking behaviours over a 16-year period. The role of the work environment should be further considered in smoking cessation programmes.

  • psychology
  • public health
  • smoking
  • stress
  • epidemiology

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Key messages

What is already known about this subject?

  • Longitudinal studies of how dimensions of the physical and psychosocial work environment are associated with cigarette use have been mixed.

  • This is potentially due to many studies focusing on only one metric of smoking behaviour; exploring follow-up lengths that are shorter than the time required for an individual to successfully quit smoking; and different approaches to measuring dimensions of the work environment over time.

What are the new findings?

  • Four distinct latent cigarette smoking trajectories were seen over the 16-year period: non-smokers, ceasing smokers, smokers and heavy smokers.

  • Between two and four latent trajectories of each work environment dimension were found.

  • Lower workplace social support and skill discretion were associated with heavy smoking; low decision authority was associated with smoking cessation; and high psychological demands and physical exertion were associated with both smoking cessation and continuing to smoke over the 16-year period.

  • These associations were similar in both men and women.

How might this impact on policy or clinical practice in the foreseeable future?

  • Study findings have identified work environment trajectories that are associated with heavy and ceasing smoking behaviours.

  • This evidence may be used in the development of policy and interventions for workplace smoking cessation.

  • Additionally, the latent physical and psychosocial work environment trajectories developed in this study may be used to better conceptualise work environments in future research.


Dimensions of the physical and psychosocial work environment may be associated with tobacco use.1 2 In 2006, Albertsen and colleagues3 completed a systematic review of 22 studies (5 longitudinal, 12 interventions, 3 time trend and 2 laboratory experiments) that explored the association between the work environment and smoking cessation, relapse and amount smoked. Their review suggested that high job demands, low job control and low social support were associated with higher smoking prevalence. However, more recent findings have been mixed.4–6

These inconsistent findings may be due to certain methodological factors. First, work environment exposures have not usually been modelled in the same way; some studies model them continuously,7–9 categorically8 or within quartiles.5 6 10 This potentially alters the variation within the work environment variable, impacting the association found with tobacco use. Second, there is variation in the working populations studied: some populations comprise only public-sector employees,4 5 10 while others focus on public-sector and private-sector employees.6 11 Further, some studies have small sample sizes7 8 and low response rates (<70% of the baseline cohort at follow-up),12 13 which may potentially introduce selection bias into results.

Third, the follow-up of smoking use seen in longitudinal studies ranges from 6 months to 15 years; the majority of studies explore a follow-up of 2–7 years4–6 and explore associations between baseline smoking measurements and only one4 10 or two11 follow-up time points. Given that quitting smoking may take years,14 exploring a longer follow-up period may provide more accurate information on how dimensions of the work environment truly influence smoking behaviours and may guide workplace health policies for reducing tobacco use.

Most importantly, most previous studies were performed between 1985 and 1994, when smoking rates were much higher than they are today.3 6 8 11 15 Given the declining smoking prevalence in North America,16–18 the association between the work environment and tobacco use requires replication in more contemporary samples. Additionally, given changes and heterogeneity of the labour force in high-income countries over the past several decades, studying this association with statistical methods designed to explore heterogeneity—such as latent growth trajectory modelling—is an alternative way to study this association.


The purpose of this study was to determine the association between dimensions of the physical and psychosocial work environment and cigarette use behaviours over a 16-year period in a general population sample of employed Canadians. Specifically, our research objectives were to (1) determine and describe the number of latent cigarette smoking trajectories among the employed Canadian population from 1994 to 2010; and (2) determine if latent work environment trajectories over the same time period were associated with belonging to a particular smoking trajectory group derived from our first objective.



Data for this study came from the longitudinal component of the Canadian National Population Health Survey (NPHS). The NPHS is a national, representative survey that occurred every 2 years from 1994 to 2010.19 The initial sample recruited totalled 17 276.19 Additional details about the NPHS and its sampling frame can be found elsewhere.19

The sample for this study was restricted to participants who were employed in the labour force in 1994, working at least 15 hours a week, and not self-employed (n=6407). The analytic sample included participants with at least two valid responses on cigarette use over the nine cycles and complete information on covariates (n=5461). Participants excluded from the analytic sample were more likely to be male, older, working full time, living in Quebec and married in 1994; there were no differences regarding work environment conditions.

Outcome: daily cigarette consumption

We defined participant smoking use as the daily number of cigarettes consumed at each cycle. Participants were asked about their current smoking habits, specifically if they smoked cigarettes daily, occasionally or not at all. Those who reported smoking cigarettes occasionally or daily were then asked, ‘how many cigarettes do you smoke each day now?’ Those who reported that they did not smoke at all were recoded to have a daily cigarette value of zero.

Cigarette use was examined at each cycle, resulting in 9 measurements over 16 years. To improve the convergence of the latent trajectory models, daily cigarette use was truncated to the 99th percentile; therefore, daily cigarette consumption ranged from 0 to 35–40 cigarettes per day over the 9 cycles.

Exposure: work environment dimensions

An abbreviated measure of the Karasek and Theorell’s20 Job Content Questionnaire was included in the 1994 and 2000–2010 NPHS cycles. The measure contained 12 questions and assessed dimensions of skill discretion, decision authority, psychological demands, job insecurity, physical exertion and workplace social support (table 1). Respondents answered questions on a 5-point Likert scale. Questions were then combined to make a scale for each psychosocial work factor. The internal consistency of psychosocial work factors ranged from 0.34 to 0.61,21 attributable to the abbreviated nature of each scale.

Table 1

Description of psychosocial work factors


All covariates were measured at baseline in 1994. Demographic covariates included sex, age, country of birth, race, marital status, dependent status, highest level of educational attainment, province of residence, and residence in a rural or urban location (defined as a city with a population concentration of 1000 or more and a population density of 400 or more per square kilometre based on the previous census). Employment covariates included the shift schedule of participants’ main employment (regular, rotating or irregular/on call schedules), and full versus part-time employment status. Having a health restriction was also included in the model and was defined as a condition impacting the individual’s ability to go to work, school or be involved in the community. Correlations between all covariates assessed over the nine cycles were examined; study covariates did not change consequentially over time, and thereby were treated as time invariant in analyses.

Statistical analysis

To complete the study objectives, a series of four steps were followed. First, we derived latent smoking trajectories using a group-based trajectory modelling (GBTM) approach with a Poisson distribution. GBTM is a subset of finite mixture modelling with the purpose of modelling latent trajectories.22 Additional details of this technique may be seen elsewhere.22 23 An iterative process of exploring different models was completed to determine and then describe the number of smoking trajectories. In this process, we specified different numbers of latent smoking classes and trajectory growth factors (ie, intercepts and slopes), and constrained and varied the variance of the growth factors to determine the most appropriate model. The most appropriate number of latent classes was selected by examining model fit statistics (Bayesian information criterion (BIC), Akaike information criterion (AIC), Lo-Mendell-Rubin likelihood ratio test and entropy values). External generalisability factors (eg, smallest class size and alignment of classes to previous literature) were also considered. A full information maximum likelihood approach was used to account for missing data.24

Second, we derived latent trajectories of each work environment dimension using the same GBTM approach with a continuous distribution. It was found that specifying an intercept, linear and quadratic slope for growth factors and constraining their variance to zero provided the best model fit for both the smoking and work environment trajectories.

Once the best trajectory model for smoking and each dimension of the work environment was complete, in the third step we quantified what latent smoking, physical and psychosocial work environment trajectory each participant belonged to. This was based on a participant’s highest probability of belonging to a specific latent class for smoking and each work environment dimension.

In the final step, we regressed participants’ most likely smoking trajectory class on participants’ most likely physical and psychosocial work environment trajectory class using a multinomial logistic regression adjusted for all covariates. It should be noted that there is a complex relationship between dimensions of the work environment, which creates challenges in isolating the effects associated with a single environment dimension. Because of this, we ran models including only a single dimension of the work environment (adjusted for covariates) and a model including all dimensions of the work environment. It is likely that when only including one work environment dimension in a model, we may overestimate the effect of each dimension. However, when adjusting for all dimensions in a model, it is similarly likely that we may underestimate the effects of specific dimensions. We have presented models where smoking class was regressed on each work environment predictor in separate models, quantifying the association between psychosocial work factors and smoking consumption trajectory group. A fully adjusted model is included in the online supplementary file.

Supplemental material

We examined if the relationships between work environment dimensions and smoking trajectories differed for men and women using stratified models.25 As they did not, we combined male and female samples and included sex as a covariate in our analyses. Additionally, given their mediating nature, other health behaviours closely related to smoking (eg, body mass index, alcohol use, physical activity) were not included as covariates in presented models; however, they were explored in sensitivity analyses. As the results did not differ substantially between these two analyses, estimates are presented without this further adjustment.

For analyses, a sampling weight was used to account for initial probability of selection and non-response to the original NPHS survey. To account for the clustered survey design of the NPHS, variances for each point estimate were estimated using a bootstrap procedure involving 500 replications for each model. Trajectory building was completed in Mplus V.7 (Muthén & Muthén, Los Angeles, California). Data preparation and multinomial logistic regressions were explored in SAS V.9.4.


Online supplementary table 1 shows the demographic, employment and health characteristics of the study sample. The average age of the sample in 1994 was approximately 37 years (SD=11.25 years). The sample was 48% female and 90% of participants were Caucasian. Most participants had at least a high school education, lived in Ontario (41%), had full-time jobs (85%) and worked regular shift schedules (80%).

Latent smoking trajectories

After computing latent trajectory models with one to five latent smoking classes, the four-class model was determined to be the most appropriate given its low AIC and BIC values, and high entropy (table 2). Although the Lo-Mendell-Rubin likelihood ratio test was not statistically significant in comparison with the three-class model, each of the four latent smoking trajectories seen (figure 1) was greater than 5% of the overall sample and was supported by previous literature.26–28

Figure 1

Latent smoking trajectories.

Table 2

Model fit information for latent smoking models

The first and largest class (n=3714) comprised individuals who reported never smoking. Roughly half of participants in this class were female and ~45% had at least some post-secondary education (online supplementary table 1). The second class (n=547) reported smoking approximately 14 cigarettes per day in 1994, which decreased to 0 cigarettes per day in 2008–2010. The third class (n=474) consisted of individuals smoking roughly 6–8 cigarettes per day for the entire follow-up. The average age of this class was approximately 34 years old, and roughly 55% of this class were female, Caucasian (91%) and had a high school diploma (58%). The final class (n=727) reported smoking roughly 22 cigarettes per day in 1994, which decreased to approximately 14 cigarettes per day in 2010. Roughly 63% of this class were male (online supplementary table 1), primarily Caucasian (~97%) and ~25% had less than a high school education. These four classes are referred to as non-smokers, ceasing/former smokers, smokers and heavy smokers, respectively.

Latent work environment trajectories

After completing latent trajectory analysis for each work environment dimension (table 3), the best models included a three-class model for skill discretion (low, medium and high skill discretion); a two-class model for decision authority (low and high); a two-class model for psychological demands (low and high); a four-class model for job insecurity (high job insecurity, decreasing job insecurity, low job insecurity and increasing job insecurity); a four-class model for physical exertion (high physical exertion, increasing physical exertion, low physical exertion and decreasing physical exertion); and a two-class model for social support (low and high). Model fit information and graphs of latent trajectories are included in the online supplementary file. Final trajectories were based on model fit indices and practical and theoretical considerations.29 For two dimensions, decision authority and social support, the optimal number of classes was not clear from the data, with four-class models also supported. As relationships with smoking and dimensions were similar across models, we opted to present the most parsimonious class models.

Table 3

Contingency table of work environment trajectories and smoking trajectory classes

Association between latent work environment and smoking trajectories

Odds ratios where participants’ most probable smoking trajectory class was regressed on each work environment dimension’s most probable trajectory class and covariates are presented in table 4. Lower skill discretion, higher psychological demands, higher or decreasing physical exertion, and low social support trajectories were associated with being in the heavy smoking trajectory when compared with the non-smoking trajectory. Lower decision authority, higher psychological demands, and higher physical exertion trajectories were associated with being in the former smoking trajectory compared with the non-smoking trajectory. When using the former smoking trajectory group as the reference, high and increasing physical exertion trajectories were associated with membership in the heavy smoking trajectory group (online supplementary file). No psychosocial work factor was associated with membership in the lighter smoking trajectory when compared with the non-smokers or former smokers.

Table 4

Associations between work environment trajectories and latent smoking behaviour trajectory membership, adjusted for covariates


The purpose of this study was to determine the number of latent cigarette smoking trajectories over a 16-year period in a representative sample of employed Canadians and to explore if latent work environment trajectories were associated with specific smoking behaviour trajectories. Our first objective suggested that four smoking behaviour trajectories exist among the Canadian workforce: non-smokers, baseline smokers who gradually stopped smoking over 14 years, respondents smoking roughly seven cigarettes per day for the entire follow-up and heavy smokers who slightly reduced their smoking intake over the 16-year period.

Roughly 32% of the sample reported smoking in 1994, and ~30% of these individuals quit smoking over the 14-year follow-up. The heaviest smoking group reduced their cigarette consumption by approximately eight cigarettes per day over the 16-year period. The other constant smoking group remained relatively stable in the daily number of cigarettes they reported. These trajectory groups generally align to those seen in other studies of adult populations,27 28 30 but build on previous work by exploring a representative sample of the Canadian working population that includes middle age.

To complete our second study objective, we developed latent trajectories across six dimensions of the work environment. Generally, all psychosocial work factors had trajectory groups with minimal variation over time, with the exception of job insecurity and physical exertion. Comparing these trajectories with previous studies is hampered since little longitudinal literature has explored latent psychosocial work trajectories.31 32 However, limited variation in skill discretion, decision authority and social support is consistent with a previous Finnish study of kitchen workers.29

Our second study objective illustrated that the work environment trajectories most strongly associated with trajectories in smoking behaviours were high perceived physical exertion, low social support, high psychological demands, and in certain cases low skill discretion and decision authority. High physical exertion was associated with both heavy smoking and smoking cessation when compared with non-smokers, and heavy smoking when compared with ceasing smokers. As such, it appears that consistently high workplace physical demands and increasing physical demands over time are associated with more deleterious smoking trajectories. This finding aligns with Albertsen and colleagues,11 who suggested that high physical exertion jobs may have more break periods, which provide employees more opportunities to smoke.11 It has been suggested that individuals working in manual labour may feel less pressure to quit smoking given the accepting nature of the habit among their workplace and the high prevalence of smoking among their coworkers10; this may be in part due to lack of restrictions related to workplace smoking in occupations involved in outdoor work.10 31 It has also been proposed that the precarious nature of some physically demanding employments may also be associated with continuing negative health behaviours such as smoking.2 33

Over half of the participants in the heavy smoking class were members of either the high or increasing physical exertion trajectories, and roughly 18% of the heavy smoking class had a decreasing smoking trajectory. A statistically significant association was also seen between the decreasing physical exertion trajectory and being a member of the heavy smoking class when compared with non-smokers. The heavy smoking class had the second highest average age (~37 years) of the smoking classes and the majority of participants had a high school education or less in 1994 (~77%); these demographic factors may also contribute to their continuing smoking habits, irrespective of whether their physical activity increased or decreased over time.

Physically demanding careers may also be associated with smoking cessation. Smoking cessation among blue-collar workers has been a priority given that the men and women in these sectors are required to be in optimal health to adequately perform their daily tasks.32 34 In a study exploring the association between workplace policies and smoking behaviours, it was seen that despite receiving fewer workplace interventions to reduce smoking compared with white-collar workers, blue-collar workers had similar intentions to quit smoking as white-collar workers.35 Therefore, we recommend future studies explore other characteristics specific to physically demanding work environments to better determine what characteristics are associated with positive and negative smoking behaviours.

The second psychosocial factor associated with smoking at least 14 cigarettes per day over the 16-year period was low workplace social support. Smoking may be a coping behaviour to deal with an adversarial work environment.8 It has been established that social support acts as a buffer against stressful life events or experiences,36 and is pertinent in individuals quitting smoking. It is therefore not surprising that individuals among the heavy smoking trajectory reported statistically lower perceived social support in the workplace compared with non-smokers. It has been suggested that employees with less social support may be less receptive to health and antismoking promotion activities in the workplace.37 Additionally, greater social support in the workplace may influence an individual’s overall self-efficacy, providing the motivation to stop smoking.10 38

Our study also found that consistently high psychological demands were associated with both heavy and ceasing smoking trajectories. This finding supports previous literature that suggests higher psychological demands up to a certain point may be protective as they keep workers busy, but as they further increase smoking may become a coping mechanism for dealing with these higher demands.9 11 39 Lower skill discretion was associated with being a heavy smoker compared with a non-smoker. It should be noted that although effect estimates were in a similar direction when compared with the former-smoking group, these were not statistically significant. Additionally, lower decision authority was associated with being in the former smoking class when compared with the non-smoking class; this is contrary to previous literature that observed medium levels of decision authority were predictive of smoking cessation.11

Strengths and limitations

This study has numerous strengths, including the use of latent trajectory modelling for both smoking and the work environments in a large, heterogeneous sample of the workforce of Canada. This provided the opportunity to examine the association between the work environment and smoking over a variety of occupations and industries, increasing the generalisability of our results. Second, the use of the full information maximum likelihood estimator in our trajectory analysis allowed us to retain ~85% of the eligible sample. Third, we explored a variety of psychosocial conditions as predictors, encompassing a holistic view of the work environment. In the multinomial logistic regression model where all psychosocial work conditions were included as predictors, the results are slightly attenuated, but align with those highlighted in table 4. Lastly, the 16-year period in which smoking behaviour was explored is one of the longest follow-up periods in the literature, providing a rich opportunity to explore changes in smoking behaviour in an adult population sample.

However, study findings should be considered with the following limitations. First, all variables used in the analyses were self-reported, increasing the risk of measurement bias in our results. Second, by exploring cigarette use as the average number smoked per day, the findings of this study may be limited to those with distinct smoking patters (eg, current vs non-smokers), and its generalisability may be limited to specific smoking profiles, such as social smokers. Third, latent trajectory methodology may misclassify participants into certain trajectory groups depending on how similar their probabilities are between latent groups. The entropy values (ie, how well the latent models can distinguish between latent classes for participants) are adequate for the latent smoking trajectories, but were lower for work environment dimensions, which may be due to the abbreviated nature of the psychosocial work factor scales. Further, for two dimensions—social support and decision authority—it was not immediately clear which number of classes was optimal given the data. Fourth, since changes in smoking and the work environment are explored simultaneously in this study, our findings cannot be interpreted causally. Finally, the workforce and psychosocial environment depicted in our study may differ from the present workforce, which could benefit from replication in more current and diverse working samples.


This study found four latent cigarette smoking trajectories among the Canadian workforce: non-smokers, ceasing smokers, smokers and heavy smokers. Apart from the 30% of smokers who stopped smoking, the remaining 70% of smokers continued to smoke at least 6-7 cigarettes per day over the 16-year period studied. Our results demonstrated that among these smoking groups, high physical exertion, low social support, high psychological demands and low skill discretion trajectories in the workplace were most strongly associated with continued smoking behaviours over time. The findings from this study may be used to more accurately inform smoking profiles for workplace smoking cessation policies and interventions. We recommend that future research use latent growth modelling techniques to explore how other workplace factors that are industry-specific may impact future smoking behaviours among workers.



  • Contributors KGD aided in conceiving the study and its design, analysed and interpreted the data, drafted the initial manuscript, and approved the final manuscript as submitted. MG-O aided in interpreting the data, and in reviewing and revising the manuscript. CAM conceived the study and its design, and aided in interpreting the data, and in reviewing and revising the manuscript. PMS conceived the study and its design, aided in analysing and interpreting the data, and in reviewing and revising the manuscript. All authors participated in approving the final version to be published and agreed 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 through a project grant from the Canadian Institutes of Health Research (CIHR) (grant number 310898). PMS is supported through a Research Chair in Gender, Work and Health from CIHR. KGD is supported through a CIHR doctoral scholarship. 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 Not required.

  • Ethics approval This study was approved by the University of Toronto Health Sciences Research Ethics Board.

  • Provenance and peer review Not commissioned; externally peer reviewed.