Workplace psychosocial resources and risk of cardiovascular disease among employees: a multi-cohort study of 135 669 participants

Objective In terms of prevention, it is important to determine effects on cardiovascular disease (CVD) when some workplace psychosocial resources are high while others are low. The aim of the study was to assess the prospective relationship between clustering of workplace psychosocial resources and risk of CVD among employees. Methods We pooled data from three cohort studies of 135 669 employees (65% women, age 18–65 years and free of CVD) from Denmark, Finland and Sweden. Baseline horizontal resources (culture of collaboration and support from colleagues) and vertical resources (leadership quality and procedural justice) were measured using standard questionnaire items. Incident CVD, including coronary heart and cerebrovascular disease, was ascertained using linked electronic health records. We used latent class analysis to assess clustering (latent classes) of workplace psychosocial resources. Cox proportional hazard models were used to examine the association between these clusters and risk of CVD, adjusting for demographic and employment-related factors and pre-existing physical and mental disorders. Results We identified five clusters of workplace psychosocial resources from low on both vertical and horizontal resources (13%) to generally high resources (28%). High horizontal resources were combined with either intermediate [hazard ratio (HR) 0.84, 95% confidence interval (CI) 0.74–0.95] or high (HR 0.88, 95% CI 0.78–1.00) vertical resources were associated with lower risks of CVD compared to those with generally low resources. The association was most prominent for cerebrovascular disease (eg, general high resources: HR 0.80, 95% CI 0.67–0.96). Conclusions Individuals with high levels of workplace psychosocial resources across horizontal and vertical dimensions have a lower risk of CVD, particularly cerebrovascular disease.

Supplementary Text S1.The choice of measurements, dimensions and cut-offs of workplace psychosocial resources.
Similar item for 'support from colleagues' (e.g."my colleagues are there for me") has been used in multicohort analyses on type 2 diabetes, including cohort studies from Sweden, United Kingdom and France (1).We applied the same cut-off for high versus low support from colleagues as in these studies, classifying e.g.response options 'completely agree' and 'somewhat agree' as high support.The measure of 'collaboration' is part of the measurement of workplace social capital (2), in which self-reported workplace social capital was associated with a lower risk of chronic hypertension among men and mortality among men and women in previous studies (3,4).We used median cut-offs for high versus low levels of collaboration as in studies on workplace social capital.In previous studies, the measure of 'procedural justice' (Moorman scale) has been divided into four (5) or three (6) categories while analyses of leadership quality have typically been based on quartiles (5,7,8).We used quartiles for both constructs.The measurement of 'leadership' included four dimensions in SLOSH and FPS in agreement with a larger IPD-Work multicohort research program, in which the current study is participating, and a previous large-scale multicohort study (7).In WEHD, the dimensions of leadership quality were slightly different due to data constrains, but this operationalization has also been used previously (8).

Supplementary Text S3. Measurement of covariates
Information on key covariates were extracted from the national register in Denmark, Finland and Sweden, except that marital status in Finland and employment contract in Sweden were measured by self-report.
Pre-existing comorbidities according to Charlson Comorbidity Index (table below) and mental disorders were detected using ICD codes from national patient register.ICD10 for diagnosed mental disorders: F01-F99.
ICD codes for Charlson Comorbidity Index, calculated using both the primary and secondary diagnosis where applicable.174,175,176,179,190,191,192,193,194,1950,1951,1952,1953,1954,1955,1958,200 BIC decreased from 2-class model to 5-class model and when adding a sixth class, models in WEHD and FPS did not converge and BIC for 6 classes in SLOSH was larger than that for 5 classes.Considering the comparability across cohorts and the interpretability of the classes, we only compared between the four-and five-class solutions.Supplementary Figure S3.Additional adjustment for lifestyle factors and prior mental disorders (N=100,517, Number of CVD events=1098).

Four
class solution: WEHD and SLOSH already showed similar patterns in the four-class solution.The only difference across the three cohorts in the four-class solution concerns the latent class 2. In FPS, we identified "intermediate vertical+low horizontal" class, and in SLOSH and WEHD, we identified the "low vertical+high horizontal" class.

Note 1 :
Main adjustment = age, sex, educational level, country of birth, marital status, pre-existing comorbidity score, pre-existing mental disorders and types of employment contract.Note 2: Information on lifestyle factors and self-reported mental health was not available in FPS 2000, 2006 and 2010.The absence of FPS wave 2000, 2006 and 2010 in this analysis for adjustments also lead to a fewer number of CVD events than that in the main analysis (CVD events=2191).

Note 3 :
All analyses were performed based on the same sample size to allow a better comparison across adjustments.

-
Supplementary FigureS1.Comparison of latent class models.BIC= Bayesian Information Criterion, the lower the better.
Risky lcohol Consumption: Women/men drinking exceeding 14/21 alcohol units (12 g of alcohol per unit) per week meant excessive alcohol consumption.Physical inactivity: Physical inactivity was observed if participants reported if having <2.0 metabolic equivalent task hours per day (corresponding to approximately 30 minutes of walking) in WEHD and FPS or reported physical inactivitiy in SLOSH.