Study design and population
We used data from the JEMPAD (Job Exposure Matrix Analyses of Psychosocial Factors and Healthy Ageing in Denmark) cohort, a Danish population-based cohort with information on employment, psychosocial factors at work, health, and socio-demographics [6, 13]. The study population was drawn from the Integrated Database for Labour Market Research (IDA) at Statistics Denmark  and consisted of all individuals residing in Denmark in the year 2000, aged 30–59, who were employed (excluding the self-employed), and had complete data on age, sex, and migration background yielding 1,680,214 individuals. Using the individuals’ unique Danish civil registration number, we linked these individuals to other population-based registers providing individual-level information on socio-demographics, use of health services, diagnoses for hospital treatment (in- and outpatient), and causes of death.
We included individuals without a history of a hospital-diagnosis of any of eight chronic diseases (diabetes (type 1 or type 2), CHD, stroke, cancer, asthma, COPD, hearth failure, and dementia). We excluded 87,723 (5.5%) individuals with one or more of these diseases [diabetes (n = 17,201), CHD (19,606), stroke (7343), cancer (27,555), asthma (14,812), COPD (6010), heart failure (1864), and dementia (10)] diagnosed from 1977 (when information on diagnosed diseases became available in the registers) to 31 December 2000 (study baseline). The final study population consisted of 1,592,491 individuals (773,354 women and 819,137 men). Figure 1 presents a flowchart for the study population. As data was linked with population-based registers, none of the cohort members were lost to follow-up. Participants who emigrated from Denmark or who died of other causes than from the diseases under study, were censored at the date of emigration and death, respectively.
To estimate the number of years without any of the eight chronic diseases we linked the study population with individual records from the same national registers until the end of follow-up (31 December 2018).
We estimated work stress as the combination of job strain and ERI using JEMs based on information from the Danish Work Environment Cohort study (DWECS) [15, 16].
In DWECS, job strain was measured using three items on psychological demands at work and five items on job control from DWECS. In accordance with previous research [5, 8, 13, 17], we defined job strain as higher than the median on score for psychological demands and lower than the median on the score for job control. In line with previous research [7, 18], we defined ERI in DWECS as the combination of four items on effort and four items on reward and calculated an effort-reward ratio and defined respondents with an effort-reward ratio above one as having ERI. Area under the curve (AUC) for the JEMs was 0.70 and 0.73 for job strain and ERI, respectively. Supplementary material, Appendix 1, including Table A1, provides a detailed description of the measurement of job strain and ERI and the construction of the JEMs. We assigned the predicted probability of job strain and ERI, respectively, to each individual in the JEMPAD cohort by job group, sex, and age in 2000. We categorised each cohort member into high and low prevalence of job strain and ERI based on previous results on the overall prevalence of job strain and ERI from a pooled European study of 90,164 participants conducted between 1985 and 2005 in Denmark, Finland, France, Germany, Sweden, and the United Kingdom (the “IPD-Work consortium”) . The pooled prevalence from the 11 studies were 15.9% and 31.7% for job strain and ERI, respectively . We applied this information on the pooled prevalence to JEMPAD by categorizing the top 15.9% and the top 31.7% of the cohort as high prevalence of job strain and ERI, respectively. We defined work stress as a joint work stress variable of exposure to job strain and ERI simultaneously. We categorised individuals into four groups: (1) no stressors (not exposed to job strain and ERI), (2) job strain only (exposed to job strain but not ERI), (3) ERI only (exposed to ERI but not job strain), and (4) both stressors (exposed to both job strain and ERI). In the groups of individuals categorised as exposed to both stressors, the majority were employed in elementary occupations (37.5%) such as cleaners and helpers, food preparation and manufacturing, and as general office clerks (23.9%).
Chronic disease outcome
We defined chronic diseases based on the World Health Organisation’s priority of non-communicable chronic diseases target for prevention including type 2 diabetes, CHD, stroke, cancer, asthma, and COPD [1, 2] and further added heart failure and dementia as suggested by Nyberg et al.  We ascertained incident chronic disease by diagnoses from the National Patient Register  (including both main and secondary diagnoses) and the Danish Register of Causes of Death  (including both underlying and contributing causes) from 1 January 2001 to 31 December 2018.
We defined the eight chronic diseases by hospital-diagnosis or death during follow-up with ICD-10 codes (see Supplementary material, Appendix 2). We defined prevalent chronic diseases by hospital-diagnosed chronic diseases during or before the baseline year with ICD-8 and ICD-10 codes (ICD-9 was never used in Denmark) from 1977 (outpatient data available from 1995) to 31 December 2000 (see Supplementary material, Appendix 3).
From population-based registers [21,22,23,24] we included sex (women and men), age, migration background (Danish origin (the whole population in Denmark except immigrants and descendants of immigrants), immigrants (born abroad and none of the parents were either Danish citizens or born in Denmark), and descendants of immigrants (born in Denmark and none of the parents were either Danish citizens or born in Denmark)), family type (single without children, single with children below age 8, single with children age 8–17 without children below age 8, married/cohabitant without children, married/cohabitant with children below age 8, or married/cohabitant with children age 8–17 without children below age 8) as covariates. We further included health service use as an indicator for health status (measured as the number of yearly contacts and services within the primary health care system in quartiles) and socioeconomic position (measured by equivalent household disposable income accounting for household size in quartiles) as they might be associated with both work stress and risk of incident chronic disease.
We further included number of risky health behaviours (risk of smoking, high weekly alcohol intake, high BMI, and low leisure time physical activity) estimated by job group aggregated JEMs from the Danish Occupational Cohort (DOC*X) study  as potential confounders or mediators. Intraclass correlation coefficients were 3.52%, 2.12%, 2.81%, and 0.26% for smoking, BMI, alcohol, and leisure time physical activity, respectively. We calculated the number of risky health behaviours separately for women and men to account for overall sex differences in the JEMs. Based on the distributions of the predicted probability of smoking, predicted level of BMI, and the predicted level of weekly alcohol consumptions, we categorised individuals into high risk of smoking, high BMI, and high weekly alcohol consumption with cut-points at the highest tertile, respectively. Based on the distribution of the predicted level of leisure time physical activity, we categorised individuals into low leisure time physical activity with a cut-point at the lowest tertile. These cut-off points correspond at the occupational level to a predicted probability of smoking of 30% or higher, a weekly alcohol intake of more than 7 units/week, a predicted level of BMI higher than 25, and leisure time physical activity of less than 2 h. The number of risky health behaviours was calculated for each individual and as few individuals were assigned four risky health behaviours, we collapsed three and four risky health behaviours.
We measured all covariates in 2000 except the number of health services used, which we measured one year before baseline (1999) to ensure that use of health services took place before the measurement of work stress. See Supplementary material, Appendix 4, for a detailed description of the covariates.
All analyses were conducted in SAS 9.4 separately for women and men to account for overall sex-differences in the average chronic disease-free life expectancy in Denmark  and overall sex-segregation of the Danish labour market . Using Cox proportional hazard model we estimated the hazard ratio (HR) and 95% confidence intervals (CI) for the risk of incident chronic disease using the PHREG procedure. We defined age as the underlying timescale from 1 January 2001 until the first event or censuring due to migration, death (due to other reasons than the eight chronic diseases under study), or end of follow-up, 31 December 2018, whichever came first. We calculated crude associations as cases per 1000 person years, and conducted crude survival analyses with age as the underlying time scale (model 1) and analyses further adjusted for migration background, family type, number of health services used, and household disposable income (model 2), as the main model of the analysis. In addition, we computed a model further adjusted for the number of risky health behaviours (model 3). We considered this model as over-adjusted, as risky health behaviours are likely not only confounders but also potential important intermediate steps in the pathway linking exposure to work stressors with incident chronic disease [28,29,30]. Consequently, we did not consider model 3 as the main model, but we still wanted to conduct this model, as this could provide insight into possible mechanisms between work stress and chronic disease-free life expectancy .
Based on the baseline function from the Cox proportional hazard models we estimated the chronic disease-free life expectancy by calculating the estimated mean survival time from age 30 to age 75 as the area under the estimated survival curve for all possible combinations of work stress and covariates. We then assigned the mean survival time to all individuals based on their individual covariate structure. We estimated 95% confidence intervals for the mean differences in chronic disease-free life expectancy using the 95% upper and lower confidence limit of the estimated survival curves from the baseline function as previously suggested [31, 32]. We defined statistically significant differences in chronic disease-free life years lost due to work stress as non-overlapping confidence intervals.
All supplementary analyses were adjusted for the covariates in model 2. First, we performed an analysis restricted to the six non-communicable chronic diseases priorities by WHO as target for prevention (type 2 diabetes, CHD, stroke, cancer, asthma, COPD) [1, 2]. Second, we conducted an analysis on exposure contrast by using the DWECS 2000 specific prevalence of job strain and ERI instead of the pooled prevalence’s retrieved from the IPD-Work consortium (job strain: 10.7% instead of 15.9% and ERI: 23.8% instead of 31.7%) . Third, we estimated the association between work stress and incident chronic disease and chronic disease-free life expectancy from age 50 to 75 in a subsample of individuals age 50 or above at baseline (n = 461,141). Fourth, we estimated the incidence of chronic diseases in subgroups of household disposable income in quartiles (low, medium–low, medium–high, and high). Fifth, we conducted outcome-specific analyses for the association between work stress and incident risk of the eight included chronic diseases separately. We grouped the eight chronic diseases as described in Supplementary material, Appendix 2 and censored due to hospital-diagnosis or death due to another chronic disease. Sixth, we analysed job strain and ERI as separate exposures. Finally, we estimated age and sex-adjusted associations between the covariates and incident risk of chronic diseases and chronic disease-free life expectancy.