Abstract
Background: Job insecurity has been associated with certain health outcomes. We examined the role of job insecurity as a risk factor for incident diabetes.
Methods: We used individual participant data from 8 cohort studies identified in 2 open-access data archives and 11 cohort studies participating in the Individual-Participant-Data Meta-analysis in Working Populations Consortium. We calculated study-specific estimates of the association between job insecurity reported at baseline and incident diabetes over the follow-up period. We pooled the estimates in a meta-analysis to produce a summary risk estimate.
Results: The 19 studies involved 140 825 participants from Australia, Europe and the United States, with a mean follow-up of 9.4 years and 3954 incident cases of diabetes. In the preliminary analysis adjusted for age and sex, high job insecurity was associated with an increased risk of incident diabetes compared with low job insecurity (adjusted odds ratio [OR] 1.19, 95% confidence interval [CI] 1.09–1.30). In the multivariable-adjusted analysis restricted to 15 studies with baseline data for all covariates (age, sex, socioeconomic status, obesity, physical activity, alcohol and smoking), the association was slightly attenuated (adjusted OR 1.12, 95% CI 1.01–1.24). Heterogeneity between the studies was low to moderate (age- and sex-adjusted model: I2 = 24%, p = 0.2; multivariable-adjusted model: I2 = 27%, p = 0.2). In the multivariable-adjusted analysis restricted to high-quality studies, in which the diabetes diagnosis was ascertained from electronic medical records or clinical examination, the association was similar to that in the main analysis (adjusted OR 1.19, 95% CI 1.04–1.35).
Interpretation: Our findings suggest that self-reported job insecurity is associated with a modest increased risk of incident diabetes. Health care personnel should be aware of this association among workers reporting job insecurity.
The increasing use of temporary contracts, zero-hours contracts and other forms of flexible employment have made job insecurity a feature of much previously secure employment in high-income countries.1 In addition to impacts on social circumstances, the health consequences of job insecurity are becoming recognized.2 Most evidence to date has relied on self-reported health outcomes, such as mental and physical health symptoms.3–5 In addition, an association has been reported between job insecurity and cardiovascular risk factors, such as dyslipidemia and weight gain,6 and a recent meta-analysis of individual data for 170 000 workers showed an association between job insecurity and clinically verified incident coronary events.7
The prevalence of diabetes has increased steadily over recent decades, mostly owing to rising rates of overweight and obesity, and aging populations.8,9 There is indirect evidence to suggest an association between job insecurity and incident diabetes because previous work has shown an association between job insecurity and a subsequent increase in body mass index (BMI).6 A high BMI, in turn, is a strong risk factor for diabetes.10,11 However, a comprehensive search of the literature (Appendix 1, available at www.cmaj.ca/lookup/suppl/doi:10.1503/cmaj.150942/-/DC1) revealed no published studies examining the association between job insecurity and diabetes.
To address this gap in the literature, we undertook a meta-analysis of individual participant data from 8 cohort studies identified in 2 open-access data archives and 11 cohort studies from the Individual-Participant-Data Meta-analysis in Working Populations Consortium (IPD-Work Consortium). This approach allowed us to quantify the prospective association between job insecurity and subsequent incident diabetes in a large data set that included a wide variety of workers and countries.
Methods
Study population
We used individual-level data on job insecurity and incident diabetes for participants in 19 prospective cohort studies. Eight studies had open-access data and were identified from collections at the Interuniversity Consortium for Political and Social Research (www.icpsr.umich.edu/icpsrweb/ICPSR) and the UK Data Service (http://ukdataservice.ac.uk/).12–19 Six of these studies involved general population samples.12–17 The other 2 included random samples of graduates from Wisconsin high schools and their siblings.18,19
The other 11 were European cohort studies20–30 participating in the IPD-Work Consortium.31 Four of the 11 studies included general population samples, 20–22,24 and the rest involved either workers in the public sector or employees in private companies.23,25–30 Further details about the studies are available in Appendix 2 (www.cmaj.ca/lookup/suppl/doi:10.1503/cmaj.150942/-/DC1).
For our meta-analysis, we included all women and men from the cohort studies who were in employment and free of diabetes at baseline and for whom complete data on job insecurity were available.
Measurement of job insecurity
Job insecurity was measured once at baseline in all 19 studies (Table 1). In the 8 studies from the open-access data sets, a question was asked about the level of insecurity in the person’s current job or about satisfaction with job security. In the other 11 studies, a question was asked about the level of insecurity in the person’s current job or about fear of layoff or unemployment. In all of the studies, the exposure was dichotomized into high or low job insecurity, as described previously.7
Measurement and prevalence of self-reported job insecurity in the included cohort studies
Outcome measure
The primary outcome was incident diabetes. The 8 studies from the open-access data sets defined incident diabetes over the follow-up period as the first self-report of diabetes. Of the 11 studies from the IPD-Work Consortium, the Whitehall II study32 used the gold-standard World Health Organization criteria (a 75-g oral glucose tolerance test, with diabetes defined as a fasting glucose level of at least 7.0 mmol/L, or a 2-hour post-load glucose level of at least 11.1 mmol/L, except for patients who had physician-diagnosed diabetes or who were using diabetes medication). The other studies from IPD-Work Consortium defined incident diabetes as the first record of diabetes, diagnosed according to the International Statistical Classification of Diseases and Related Health Problems 10th Revision code E11. This information was collected from hospital admission, hospital discharge and mortality registers that had a mention of diabetes in any of the diagnostic codes. In addition, in the Finnish studies,23,24,27 participants were defined as having incident diabetes the first time they were eligible for diabetes medication in the national drug reimbursement register. The date of diabetes diagnosis was defined as the date of the first record in any of the above-mentioned sources over the study follow-up period.
Participants with evidence of prevalent diabetes at baseline were excluded. Prevalent diabetes was defined on the basis of information from any of the following: hospital records, baseline oral glucose tolerance test results, self-report from the baseline questionnaire or drug reimbursement register (Finnish studies only).
Assessment of covariates
Confounders of the association between job insecurity and incident diabetes include age, sex, socioeconomic position, obesity, and reporting or common-method bias for studies in which both exposure and outcome are self-reported.
We were able to obtain the following data from almost all of the studies: participants’ age, sex, socioeconomic status (based on participants’ highest occupational grade or educational qualification and classified as low, intermediate or high) and obesity (defined as a BMI above 30). Other risk factors for diabetes, which may be associated with job insecurity and so act as potential confounders of the association, were physical activity (low, intermediate or high), smoking (current, former or never) and alcohol use (none, moderate, intermediate or heavy); these risk factors were similarly predefined and harmonized across the studies.
Data were not available on obesity from 2 studies;14,27 on alcohol use from 1 study;26 and on obesity, physical activity and alcohol use from another study.20 These 4 studies were excluded from the multivariable-adjusted models.
Statistical analysis
Our analyses included 19 prospective cohort studies in which job insecurity was measured once at baseline and subsequent incident diabetes was measured over the follow-up period. Because not all of the studies included an exact date of diabetes diagnosis, we used logistic regression in all studies to calculate study-specific odds ratios (ORs) and 95% confidence intervals (CIs) as the measure of association between job insecurity and subsequent incident diabetes.33
Meta-analysis was used to produce a common risk estimate.34 Because there was no significant heterogeneity between the study-specific estimates, we conducted the meta-analyses using fixed-effect models. Heterogeneity of the study-specific estimates was examined using the I2 statistic (higher values denote greater heterogeneity).35
In the preliminary analysis, we calculated age-and sex-adjusted study-specific effect estimates of the association between job insecurity and incident diabetes. In the main analysis, we used multivariable models that were further adjusted for socioeconomic status, obesity, physical activity, alcohol use and smoking. To examine whether the association between job insecurity and incident diabetes differed between subgroups of studies and participants, we stratified the analyses by method of diabetes diagnosis (self-reported, electronic medical records or clinical examination), study quality (assessed as low or high using the Cochrane risk-of-bias tool for cohort studies,36 see Appendix 3, available at www.cmaj.ca/lookup/suppl/doi:10.1503/cmaj.150942/-/DC1), age (< 50 yr or ≥ 50 yr), sex, socioeconomic status (low, intermediate or high) and study location (Europe or United States).
We used Stata/MP version 13.1 (StataCorp) to analyze data from the open-access data archives and to compute the results of all the meta-analyses. We used SAS version 9.2 (SAS Institute Inc.) to analyze study-specific data from the IPD-Work studies.
Results
Sample characteristics
The 8 studies from the open-access data sets included a total of 44 770 working women and men with data on age, sex, socioeconomic status, job insecurity and diabetes. The 11 studies from the IPD-Work Consortium included a further 96 055 working women and men with suitable data, bringing the total study population to 140 825 (mean age 42.2 yr; 81 816 [58.1%] women) (Table 2). Overall, 3954 incident cases of diabetes occurred over a mean follow-up of 9.4 (range 4.0–21.1) years. Although 2 studies were started in 1986,12,27 baseline assessment for the remaining studies was between 1991 and 2009. Studies were from Australia, Denmark, Finland, Sweden, the United Kingdom and the US (Table 1, Appendix 2).
Characteristics of participants and assessment of incident diabetes in the included cohort studies
Association between job insecurity and incident diabetes
The prevalence of high job insecurity ranged from 6.3% to 40.3% (Table 1). The mean incidence of diabetes per 10 000 person-years ranged from 9.0 to 85.2 (Table 2).
Age- and sex-adjusted estimates of the association between job insecurity and incident diabetes for the 19 studies are presented in Figure 1A. The multivariable-adjusted analyses, additionally adjusted for socioeconomic status, obesity, physical activity, alcohol use and smoking, are presented in Figure 1B for the 15 studies with data on all covariates (n = 108 525; 2850 incident diabetes cases).
Study-specific analyses of association between job insecurity and incident diabetes (A) after adjustment for age and sex and (B) after adjustment for age, sex, socioeconomic status, obesity, physical activity, alcohol use and smoking. Values greater than 1.0 indicate an increased risk of incident diabetes. CI = confidence interval, OR = odds ratio. See Table 1 for full study names.
High job insecurity at baseline was associated with an increased risk of diabetes in the age- and sex-adjusted analysis compared with low job insecurity (pooled OR 1.19, 95% CI 1.09–1.30). The effect was attenuated in the multivariable-adjusted analysis but remained statistically significant (pooled OR 1.12, 95% CI 1.01–1.24). Heterogeneity between the study-specific estimates was low to moderate (age- and sex-adjusted analysis: I2 = 24%, p = 0.2; multivariable-adjusted analysis: I2 = 27%, p = 0.2). Sequential adjustment of the association between job insecurity and incident diabetes for socioeconomic status and the lifestyle covariates are presented in Appendix 4 (available at www.cmaj.ca/lookup/suppl/doi:10.1503/cmaj.150942/-/DC1).
The results of the subgroup analyses are shown in Figure 2. We found no statistically significant differences in the association between job insecurity and incident diabetes in the multivariable-adjusted analyses when stratified by method of diabetes diagnosis, study quality, age, sex, socioeconomic status or study location (p value > 0.1 for all subgroup differences). Odds ratios for the subgroups divided by diagnosis method and study quality were identical because the diagnosis of diabetes is a key feature of high-quality (electronic medical records or clinical examination [oral glucose tolerance test]) and low-quality (self-report) studies. Although the correlation between diabetes identified by self-report and medical records is relatively high37 and the difference between the high- and low-quality studies was not statistically significant, these analyses provide stronger evidence in support of an association between job insecurity and incident diabetes in the high-quality studies (pooled OR 1.19, 95% CI 1.04–1.35).
Subgroup analyses of association between job insecurity and incident diabetes after adjustment for age, sex, socioeconomic status, obesity, physical activity, alcohol use and smoking (15 cohorts, n = 108 525; 2850 incident cases of diabetes). Values greater than 1.0 indicate an increased risk of incident diabetes. CE = clinical examination (oral glucose tolerance test), CI = confidence interval, EMR = electronic medical record, OR = odds ratio.
Loss to follow-up ranged from less than 5% to 34%, and length of follow-up from 4 to 21 years (Appendix 2), but neither factor had an effect on the association between job insecurity and incident diabetes (Appendix 4). The rate of unemployment at baseline varied from 4.6% to 11.3% (Appendix 2), but there was no evidence that the association between job insecurity and incident diabetes differed between the cohorts (Appendix 4).
Interpretation
In our meta-analysis of individual-level data from 19 prospective cohort studies involving more than 140 000 participants and close to 4000 incident cases of diabetes, we observed a 19% increase in the age- and sex-adjusted odds of incident diabetes among workers who reported high levels of job insecurity. In the 15 studies with complete covariate data, the multivariable-adjusted association was attenuated to 12%, but it remained statistically significant. Most of this attenuation resulted from adjustment for the lower socioeconomic status among the workers who reported job insecurity.
Because we were unable to find previous studies, either cross-sectional or longitudinal, that examined the association between job insecurity and incident diabetes, our study appears to be the first to report on this association. Our findings are congruent with those from studies showing an association between job insecurity and weight gain,6 a risk factor for diabetes, and between job insecurity and incident coronary artery disease,7 a complication of diabetes. In the latter meta-analysis of cohort studies from the IPD-Work Consortium,7 employees who reported job insecurity had a 19% increase in the multivariable-adjusted odds of incident myocardial infarction or coronary death. The strength of the association was the same as for incident diabetes in high-quality studies in the current analysis.
Limitations
Our study needs to be considered in view of several limitations. Although we were able to adjust our analyses for age, sex, socioeconomic status and obesity at baseline, data on other potential confounders and mediators, such as anxiety and weight gain over the follow-up period, were not available in most of the data sets.
We cannot claim that our analysis included all possible data. However, we were able to include a large, diverse sample of workers from 19 well-characterized prospective cohort studies that together cover the US, Australia and several European countries. Therefore, our findings are likely to apply more widely to workers in other high-income countries.
Job insecurity was measured with the use of single questions that were not uniform across the studies. In common parlance, job insecurity is understood to refer to employed workers who feel threatened by unemployment, a broad concept around which the single-item measures in our meta-analyses appear to coalesce.38,39 Low to moderate heterogeneity, as indicated by the I2 statistics suggests effects that differ little between the studies. However, the use of single, rather than multi-item questionnaires at one point in time only to measure job insecurity may result in an underestimation of the association between job insecurity and health-related outcomes,40 a limitation which may also apply to our study. Previous work has also shown that chronic or repeated exposure to job insecurity is more harmful to health than exposure to job insecurity at one point in time.41
Ascertainment of diabetes varied between the studies. Only the Whitehall II study administered a repeated oral glucose tolerance test, the gold standard. This enabled the study to detect both diagnosed and undiagnosed diabetes. The remaining studies, based on health records or self-reports, will have missed undiagnosed diabetes cases. In Whitehall II, the age and sex-adjusted odds ratio for the association between job insecurity and diabetes was 1.19; the same as the overall estimate for all the studies (1.19).
Conclusion
Our findings suggest that self-reported job insecurity is associated with a modest increased risk of incident diabetes. These findings are most appropriately interpreted in a public health context in which small long-term effects on common disease outcomes can have high relevance. Ideally in such situations, policy responses should take a population-level approach to reducing exposure to job insecurity. Also, health care personnel should be aware of that workers reporting job insecurity may be at modest increased risk of diabetes.
Footnotes
Competing interests: None declared.
This article has been peer reviewed.
Contributors: All of the authors contributed to the study concept and design and to the analysis and interpretation of data. Jane Ferrie and Marianna Virtanen undertook the literature search, and Markus Jokela searched the relevant open-access data sets. Markus Jokela and Ida Madsen performed the statistical analysis. Mika Kivimäki, TÖres Theorell, Reiner Rugulies and Nico Dragano obtained funding for the IPD-Work Consortium. Jane Ferrie, Marianna Virtanen and Mika Kivimäki drafted the manuscript. All of the authors revised the manuscript critically for important intellectual content, approved the final version to be published and agreed to be guarantors of the work.
Funding: The IPD-Work Consortium is supported by Nord-Forsk (grant no. 75021), the Nordic Programme on Health and Welfare; the EU New OSH ERA Research Programme (funded by the Finnish Work Environment Fund; the Swedish Research Council for Health, Working Life and Welfare; the German Social Accident Insurance; and the Danish National Research Centre for the Working Environment); the Academy of Finland (grant nos. 132944 and 258598); and the Bupa UK Foundation (grant no. 22094477). Mika Kivimäki is supported by the Medical Research Council (grant no. K013351) and the Economic and Social Research Council, UK. Funding bodies for the participating cohort studies are listed on their websites. The study sponsors had no role in the design of the study, the collection, analysis or interpretation of data, the writing of the report or the decision to submit the article for publication.
- Accepted August 2, 2016.
References
In this issue
Article tools
