Job strain and effort–reward imbalance as risk factors for type 2 diabetes mellitus: A systematic review and meta-analysis of prospective studies

Objectives This systematic review and meta-analysis aimed to synthesize the available data on prospective associations between work-related stressors and the risk of type 2 diabetes mellitus (T2DM) among adult workers, according to the demand–control–support (DCS) and the effort–reward imbalance (ERI) models. Method We searched for prospective studies in PubMed, EMBASE, Web of Science, Scopus, CINHAL and PsychInfo. After screening and extraction, quality of evidence was assessed using the ROBINS-I tool adapted for observational studies. The effect estimates extracted for each cohort were synthesized using random effect models. Results We included 18 studies (reporting data on 25 cohorts) in meta-analyses for job strain, job demands, job control, social support at work and ERI. Workers exposed to job strain had a higher risk of developing T2DM when compared to unexposed workers [pooled rate ratio (RR) 1.16, 95% confidence interval (CI) 1.07–1.26]. This association was robust in several supplementary analyses. For exposed women relative to unexposed women, the RR was 1.35 (95% CI 1.12–1.64). The RR of workers exposed to ERI was 1.24 (95% CI 1.08–1.42) compared to unexposed workers. Conclusions This is the first meta-analysis to find an effect of ERI on the onset of T2DM incidence. It also confirms that job strain increases the incidence of T2DM, especially among women.


Supplementary Material 2. Correspondence to: Ana Paula Bruno
Supplementary List the confounding domains relevant to all or most studies Major confounding domains (for which we want the analyzes to be compulsorily adjusted): Socio-economic status (ideally education or income, but we also accept occupation), Age and Sex. Additional confounding domains, but optional (we use the most adjusted model without including intermediate domains): Work Environment Factors, Family Charge, Stressful Events, Out of Work Social Support, Gender Confounding and Intermediate domains (should not be adjusted for): Body mass index (BMI), Lifestyle factors, Comorbidities, hours worked per week, multiple jobs List co-interventions that could be different between intervention groups and that could impact on outcomes

Specify the outcome
Specify which outcome is being assessed for risk of bias (typically from among those earmarked for the Summary of Findings table). Specify whether this is a proposed benefit or harm of intervention.

Specify the numerical result being assessed
In case of multiple alternative analyses being presented, specify the numeric result (e.g. RR = 1.52 (95% CI 0.83 to 2.77) and/or a reference (e.g. to a table, figure or paragraph) that uniquely defines the result being assessed.

Preliminary consideration of confounders
Complete a row for each important confounding domain (i) listed in the review protocol; and (ii) relevant to the setting of this particular study, or which the study authors identified as potentially important.
"Important" confounding domains are those for which, in the context of this study, adjustment is expected to lead to a clinically important change in the estimated effect of the intervention. "Validity" refers to whether the confounding variable or variables fully measure the domain, while "reliability" refers to the precision of the measurement (more measurement error means less reliability).

Number of hours worked
Multiple jobs * In the context of a particular study, variables can be demonstrated not to be confounders and so not included in the analysis: (a) if they are not predictive of the outcome; (b) if they are not predictive of intervention; or (c) because adjustment makes no or minimal difference to the estimated effect of the primary parameter. Note that "no statistically significant association" is not the same as "not predictive".

Risk of bias assessment
Responses underlined in green are potential markers for low risk of bias, and responses in red are potential markers for a risk of bias. Where questions relate only to sign posts to other questions, no formatting is used.

YES:
Always yes in our case, except with a cohort of new workers or have selected a cohort of participants who would all be exposed, or all not exposed at recruitment, and analyze the change in exposure over time. YES: Always yes, because we can expect more exposed people to leave work before the start of the study, or to participate less in the study YES: Always yes, because we can expect more exposed people to leave work before the start of the study, or to participate less in the study.
NA / Y / PY / PN / N / NI 2.4. Do start of follow-up and start of intervention coincide for most participants?
NO: due to our field of study, this answer should always be no, except with a cohort of new workers or having selected a cohort of participants who would all be exposed or all unexposed at recruitment, and analyzed the change in exposure over time. NO: Always no, because we never know the characteristics of the participants before the start of the study.

Note:
In occupational studies start of follow-up and start of exposure rarely coincide. For this reason, we choose to start the risk of bias in selection of participants into the study to moderate levels for this criterion. However, this criterion will not be considered in the other levels in order to keep a gradation in this risk of bias. Low: Never, due to the point (ii) (i) All participants who would have been eligible for the target trial were included in the study; and (ii) For each participant start of follow up and start of intervention coincided. Moderate: Participation rates of ≥80% or ≥ 70% with a comparison showing that refusals are similar to those included for age, sex and socio-economic status, or for exposure and outcome (i) Selection into the study may have been related to intervention and outcome; and The authors used appropriate methods to adjust for the selection bias; or (ii) Start of follow up and start of intervention does not coincide for all participants; and (a) the proportion of participants for which this was the case was too low to induce important bias; (90% de participation) or (b) the authors used appropriate methods to adjust for the selection bias; or (c) the review authors are confident that the rate (hazard) ratio for the effect of intervention remains constant over time. Serious: Participation rates between 80-60% or 60-50% with a comparison showing that refusals are similar to those included for age, sex and socio-economic status, or for exposure and outcome (i) Selection into the study was related (but not very strongly) to age, sex and socio-economic status or the intervention and outcome; and This could not be adjusted for in analyses; or (ii) Start of follow up and start of intervention does not coincide; and A considerable amount of follow-up time is missing from analyses; and The rate ratio is not constant over time.
Low / Moderate / Serious / Critical / NI 13 Critical: Participation rates of less than <60% or <50% with a comparison showing that refusals are similar to those included for age, sex and socio-economic status, or for exposure and outcome (i) Selection into the study was very strongly related to ) to age, sex and socio-economic status or the intervention and outcome; and This could not be adjusted for in analyses; or (ii) A substantial amount of follow-up time is likely to be missing from analyses; and The rate ratio is not constant over time.
Optional: What is the predicted direction of bias due to selection of participants into the study?
Favours experimental / Favours comparator / Towards null /Away from null / Unpredictable Bias in classification of interventions 3.1 Were intervention groups clearly defined?
YES: Exposure must have been measured by a validated tool based on one of two models studied. The validity must have been demonstrated in a study on the psychometric qualities of the instrument (internal consistency, factorial validity, predictive validity and discriminant validity). Note: If the tool used is an original validated tool, but the translation has not been validated, it is considered to be a well-defined intervention, but with a moderate level of risk. NO: Exposure measured with a proxy or translation whose validation has not been demonstrated, or by using different questionnaires from one participant to another. Exposure measured by a matrix based on job titles or based on the response of colleagues in the same work unit, as there is a risk of significant misclassification. Bias due to deviations from intended interventions: NA: Hard to apply in our field of research. Exposure deviations are almost always natural and expected, unless there is an intervention by a researcher that is differential depending on the level of exposure. This criterion will always be at a moderate level of risk. Therefore, it is not systematically evaluated in the included studies. If your aim for this study is to assess the effect of assignment to intervention, answer questions 4.1 and 4.

Risk of bias judgment
Optional: What is the predicted direction of bias due to deviations from the intended interventions?
Bias due to missing data : NOTE: Here, missing participant data is evaluated starting at recruitment and excluding the rate of participation in recruitment that has been taken into account in the selection bias analysis. YES: if a sensitivity analysis was performed to account for missing data (multiple imputation, inverse probability weighting) and the results are similar to the main analysis, or the results are different but the interpretation is done on the sensitivity analysis and not on the main analysis. NO: if no sensitivity analysis is done for missing data

Risk of bias judgment
Low: (i) Data were reasonably complete; (95% or 90% with demonstrations that they are similar or an analysis was done for missing data) or (ii) Proportions of and reasons for missing participants were similar across intervention groups; or (iii) The analysis addressed missing data and is likely to have removed any risk of bias. Moderate (between 94 (or 89) and 80% at follow-up, can go down to 75% if a comparison shows that they are similar): (i) Proportions of and reasons for missing participants differ slightly across intervention groups; and (ii) The analysis is unlikely to have removed the risk of bias arising from the missing data. Serious (between 79% (or 74%) and 50% at follow-up with comparison): (i) Proportions of missing participants differ substantially across interventions; or Reasons for missingness differ substantially across interventions; and (ii) The analysis is unlikely to have removed the risk of bias arising from the missing data; or Missing data were addressed inappropriately in the analysis; or The nature of the missing data means that the risk of bias cannot be removed through appropriate analysis. Critical (<50%): (i) (Unusual) There were critical differences between interventions in participants with missing data; and (ii) Missing data were not, or could not, be addressed through appropriate analysis Bias in measurement of outcomes Critical High fasting glucose was defined by a plasma glucose level of >100 mg/dL (5.6 mmol/L).

Type of bias Classification Reason -explanation
Bias due to confounding Serious Adjusted for age, sex, education level and post-intervention variables that could have been affected by the intervention (chronic medical conditions). Bias in selection of participants into the study

Serious
Participation at baseline was 63% and total non-response was handled by adjusting the weight of households that responded to the survey to compensate for those who did not respond. Bias in classification of interventions

Serious
Short version of the job demands scale; partial validation with low Cronbach's α.
Bias due to missing data Low Complete data for 95% of baseline participants were included in the analyses. Bias in measurement of outcomes

Low
Obtained objectively by register (administrative data and physician diagnoses).

Type of bias Classification Reason -explanation
Bias due to confounding Serious Adjusted for age and sex, no adjustment for socioeconomic factors. Bias in selection of participants into the study

Serious
Participation at baseline was 73% without comparison between participants and nonparticipants Bias in classification of exposure

Moderate
Intervention status is well defined. Shorter version of demand scale was validated with good α.
Bias due to missing data Moderate Complete data for 82% of baseline participants were included in the analyses. Authors provide comparison between included and missing participants showing that those lost to follow-up are rather different in terms of exposure, age and/or sex. No imputation was done. Bias in measurement of outcomes

Moderate
The methods of outcome assessment were comparable across intervention groups, but some of the cases were ascertained objectively by clinical evaluation, while some were ascertained by self-reported questionnaire.
Overall Serious

Type of bias Classification Reason -explanation
Bias due to confounding Serious Stratified by sex, adjusted for age, employment grade and a post-intervention variable that could have been affected by the intervention (diet pattern). Bias in selection of participants into the study

Serious
Participation at baseline was 73% without comparison between participants and nonparticipants Bias in classification of interventions

Moderate
Intervention status is well defined. Shorter version of demand scale was validated with good α.
Bias due to missing data Serious Complete data for 72% of baseline participants were included in the analyses. Authors provide comparison between included and missing participants showing that those lost to follow-up are rather different in terms of exposure, SES, age and sex. No treatment for missing data was done. Bias in measurement of outcomes

Moderate
The methods of outcome assessment were comparable across intervention groups, but some of the cases were ascertained objectively by clinical evaluation, while some were ascertained by self-reported questionnaire.

Overall Serious
Hino 2016

Type of bias Classification Reason -explanation
Bias due to confounding Moderate Restricted by sex (men). Adjusted for age, marital status, job department, employment position and occupation. Bias in selection of participants into the study

Critical
Participation at baseline was 21% without comparison between participants and nonparticipants Bias in classification of interventions

Moderate
Questionnaire validated in Japanese workers for internal consistency.
Bias due to missing data Critical Proportions of missing participants differ substantially across interventions: 43% of the baseline participants included in the analysis, without comparison between included and missing participants. Bias in measurement of outcomes

Critical
Definition very wide, including diabetes defined by HOMA-IR, which is not a method recommended by the ADA.

Overall Critical
Huth 2014

Type of bias Classification Reason -explanation
Bias due to confounding Moderate Adjusted for age, sex, physical intensity at work: low, moderate, high. Education was coded as binary variable. Bias in selection of participants into the study

Serious
Participation at baseline was 75% without comparison between participants and nonparticipants. Bias in classification of interventions

Low
Validated version of questionnaire.
Bias due to missing data Serious Complete data for 73% of baseline participants were included in the analyses, without comparison between included and missing participants. Bias in measurement of outcomes

Moderate
The methods of outcome assessment were comparable across intervention groups. Selfreported T2DM and the date of diagnosis were validated by hospital records or by contacting the participants' treating physicians.

Overall Serious
Kawakami 1999

Type of bias Classification Reason -explanation
Bias due to confounding Serious Restricted by sex (men), adjustment for age, education level, occupation, use of technology, leisure time and physical activity, family history of diabetes and a post-intervention variable that could have been affected by the intervention (BMI). Bias in selection of participants into the study

Moderate
Participation at baseline was 92% without comparison between participants and nonparticipants.

Serious
Very short questionnaire with one item for each dimension, not validated.
Bias due to missing data Serious Complete data for 77% of baseline participants were included in the analyses without comparison between included and missing participants. Bias in measurement of outcomes

Moderate
Obtained objectively by clinical evaluation, low risk of false positive outcomes. Some risk of false negative outcomes due to triage by urine insulin, but this risk is lower because the same test had been conducted annually for 12 years before baseline (exclusion of prevalent cases) and each year during follow-up (incident cases).

Type of bias Classification Reason -explanation
Bias due to confounding Moderate Restricted by sex (women) and profession (nurses), adjusted for age. Bias in selection of participants into the study

Serious
Participation at baseline was 75% without comparison between participants and nonparticipants. Bias in classification of interventions

Low
Job strain was measured by the well-validated 27-item Karasek Job Content Questionnaire.
Bias due to missing data Serious Complete data for 73% of baseline participants were included in the analyses, with a comparison between included and missing participants that shows they are similar for all three important confounders, for exposure and for outcome. Bias in measurement of outcomes

Moderate
Self-reported diabetes with validation (98%) in a sub-sample.

Overall Serious
Kumari 2004

Type of bias Classification Reason -explanation
Bias due to confounding Serious Adjusted for age, length of follow-up, employment grade, ethnic group and a post-intervention variable that could have been affected by the intervention (ECG abnormalities). Bias in selection of participants into the study

Serious
Participation at baseline was 73% without comparison between participants and nonparticipants. Bias in classification of interventions DC (Moderate); ERI: (Serious) DC model: Intervention status is well defined; a slightly shorter version of the demand scale was validated with good α.
ERI model: Unknown number of items. According to Bosma et al 1998: "As there was no original measurement of effort-reward imbalance at phase 1, proxy measures (available from the authors) had to be constructed for the crucial components of the model." Bias due to missing data Moderate Complete data for 82% of baseline participants were included in the analyses, without a comparison between included and missing participants. Bias in measurement of outcomes

Moderate
The methods of outcome assessment were comparable across intervention groups, but some of the cases were ascertained objectively by clinical evaluation, while some were ascertained by self-reported questionnaire. Social support: only two questions without validation.

Overall Serious
Bias due to missing data Critical Complete data for 51% of baseline participants were included in the analyses, without a comparison between included and missing participants. Bias in measurement of outcomes

Serious
The methods of outcome assessment were comparable across intervention groups, but ascertained by self-reported questionnaire. Prevailing cases of T2DM were excluded. DC: Intervention status well defined. Social support: short questionnaire with two items without validation. Bias due to missing data Critical Complete data for 46% of baseline participants were included in the analyses, without a comparison between included and missing participants. Bias in measurement of outcomes

Moderate
Diagnoses were obtained by self-reported questionnaire and supplemented with information on diabetes from hospital admissions.

Type of bias Classification Reason -explanation
Bias due to confounding Moderate Adjusted by age, gender, marital status, occupational grade and follow-up duration. Bias in selection of participants into the study

Serious
Participation at baseline was 73% without comparison between participants and nonparticipants. Social support: short questionnaire with two items, no validation. Bias due to missing data Serious Complete data for 77% of baseline participants were included in the analyses, without a comparison between included and missing participants. Bias in measurement of outcomes

Moderate
The methods of outcome assessment were comparable across intervention groups, but some of the cases were ascertained objectively by clinical evaluation, while some were ascertained by self-reported questionnaire.

Type of bias Classification Reason -explanation
Bias due to confounding Critical Adjusted by education level, race, gender, occupational category, marital status, insurance coverage. No adjustment for age. Adjusted for post-intervention variables that could have been affected by the intervention (BMI, physical activity, alcohol use, hypertension, working hours). Bias in selection of participants into the study

Serious
Participation at baseline was 74% with a comparison between participants and nonparticipants. Selection into the study may have been related to intervention and outcome.

Bias in classification of interventions Serious
Shorter version without information on the validity of the modified JCQ questionnaire, which was a combination of Karasek and Quinn models. Bias due to missing data Critical Complete data of 19% or 50% of baseline participants were included in the analyses, reasons for exclusion unclear. Only 56 participants with missing data were analyzed: "Participants with missing data on the independent variables were excluded from the final multivariate survival analyses. (n = 56, 3.9%). These participants were more likely to be working in high strain jobs at baseline, older, and women." Bias in measurement of outcomes Serious All of the diagnoses were obtained by self-reported questionnaire.

Type of bias Classification Reason -explanation
Bias due to confounding Moderate Adjusted by age, sex, race/ethnicity, education level, and marital status. Bias in selection of participants into the study

Serious
Participation at baseline was 74% without comparison between participants and nonparticipants. Selection into the study may have been related to intervention and outcome.

Serious
Short version of ERI without information on validity.
Bias due to missing data Critical Complete data of between 24%-59% of baseline participants were included in the analyses, reasons for exclusion is unclear; no comparison between included and missing participants. Bias in measurement of outcomes

Serious
All the diagnoses were obtained by self-reported questionnaire. Participation at baseline was 61% and 59% for COPSOQ-I and COPSOQ-II, respectively, without comparison between participants and non-participants.

Serious
Short version of job demands (3 items) with substantial agreement with complete version.
Bias due to missing data COPSOQ-I=Low,

COPSOQ-II=Moderate
Complete data for 95% of participants (COPSOQ-I) and 88% (COPSOQ-II), without a comparison between included and missing participants Bias in measurement of outcomes COPSOQ-I=Low, COPSOQ-II=Moderate COPSOQ I: Obtained objectively from registers (hospitalization registers).
COPSOQ II: The methods of outcome assessment were comparable across intervention groups, but some of the cases were ascertained objectively from registers, while some were ascertained by self-reported questionnaire. Participation at baseline was 75%, without a comparison between participants and nonparticipants Bias in classification of interventions

Serious
Short version (demands 3 items, control 5) with substantial agreement with complete version.
Bias due to missing data Low Complete data for 99% of baseline participants were included in the analyses. Bias in measurement of outcomes

Low
Obtained objectively by register (administrative data mortality and hospitalization registers). The methods of outcome assessment were comparable across intervention groups, but some of the cases were ascertained objectively from registers, while some were ascertained by selfreported questionnaire.

Overall Serious
Nyberg-Gazel Bias due to missing data Serious Complete data for 53% of baseline participants were included in the analyses without a comparison between included and missing participants Bias in measurement of outcomes

Serious
The methods of outcome assessment were comparable across intervention groups, but ascertained by self-reported questionnaire. Participation at baseline was 40%, without a comparison between participants and nonparticipants.

Bias in classification of interventions Low
The original version was complete and validated Bias due to missing data Serious Complete data for 62% of baseline participants were included in the analyses, without a comparison between included and missing participants Bias in measurement of outcomes

Moderate
The methods of outcome assessment were comparable across intervention groups, but some of the cases were ascertained objectively from administrative registers (hospital and reimbursement), while some were ascertained by self-reported questionnaire.

Overall Critical
Nyberg-IPAW Participation at baseline was 76%, without a comparison between participants and nonparticipants Bias in classification of interventions

Serious
Short version (demands 2 items) with substantial agreement with complete version Bias due to missing data Low Complete data for 96% of baseline participants were included in the analyses. Bias in measurement of outcomes

Moderate
The methods of outcome assessment were comparable across intervention groups, but some of the cases were ascertained objectively from administrative registers (hospital), while some were ascertained by self-reported questionnaire.

Overall Serious
Nyberg-PUMA

Type of bias Classification Reason -explanation
Bias due to confounding Moderate Adjusted by age, sex, SES socioeconomic status (occupational title, register based), categorized in low intermediate, high or other) Bias in selection of participants into the study

Moderate
Participation at baseline was 80%, without a comparison between participants and nonparticipants Bias in classification of interventions

Serious
Short version (demands 3 items, control 5 items) with substantial agreement with complete version Bias due to missing data Low Complete data for 96% of baseline participants were included in the analyses. Bias in measurement of outcomes

Low
Obtained objectively from registers (hospitalization)

Overall Serious
Nyberg-SLOSH The methods of outcome assessment were comparable across intervention groups, but some of the cases were ascertained objectively from administrative registers (hospital), while some were ascertained by self-reported questionnaire.

Type of bias Classification Reason -explanation
Bias due to confounding Moderate Adjusted by age, sex, SES socioeconomic status (occupational title, register based), categorized in low intermediate, high or other) Bias in selection of participants into the study

Serious
Participation at baseline was 76%, without a comparison between participants and nonparticipants Bias in classification of interventions

Serious
Short version (demands 2 items, control 5 items) with substantial agreement with complete version. Bias due to missing data Low Complete data for 98% of baseline participants were included in the analyses. Bias in measurement of outcomes

Low
Obtained objectively by register (administrative data reimbursement and hospitalization).

Overall Serious
Nyberg-Whitehall II Complete data for 81% of baseline participants were included in the analyses, without a comparison between included and missing participants. Bias in measurement of outcomes

Moderate
The methods of outcome assessment were comparable across intervention groups, but some of the cases were ascertained objectively by clinical evaluation, while some were ascertained by self-reported questionnaire.

Type of bias Classification Reason -explanation
Bias due to confounding Moderate Adjusted by age, sex, SES socioeconomic status (occupational title, register based, categorized as low, intermediate, high or other) Bias in selection of participants into the study

Moderate
Participation at baseline was 82% together according to Alfredsson et al. (2002). Without comparison between participants and non-participants.

Bias in classification of interventions Low
The original scales of job demand and job control from WOLF N was complete and validated Bias due to missing data Low Complete data for 98% of baseline participants were included in the analyses. Bias in measurement of outcomes

Moderate
The methods of outcome assessment were comparable across intervention groups, but some of the cases were ascertained objectively from administrative registers (hospital), while some were ascertained by self-reported questionnaire.

Moderate for both
Pan 2017

Type of bias Classification Reason -explanation
Bias due to confounding Moderate Adjusted for sex, age, education level, vital status and follow-up Bias in selection of participants into the study

Serious
Participation at baseline was 73% without comparison between participants and nonparticipants.

Serious
Job strain not measured individually but obtained through a job exposure matrix based on job titles.
Bias due to missing data Moderate Complete data for 88% of baseline participants were included in the analyses, without comparison between included and missing participants. Multiple imputation with similar results according to the authors. Bias in measurement of outcomes

Moderate
Some diagnoses were obtained objectively from clinical evaluation and register (administrative data, medical records in Stockholm) and some of the diagnoses were obtained from selfreported questionnaires.

Overall Serious
Smith 2012

Type of bias Classification Reason -explanation
Bias due to confounding Serious Stratified for sex, adjusted for age, education level, marital status, ethnicity, immigration status, urban or rural and also for post-intervention variables that could have been affected by the intervention (chronic diseases, activity limitation at work due to health problems). Bias in selection of participants into the study

Bias in classification of interventions Serious
Shorter versions of job demands (2 items), job control (5 items) and social support (3 items) questionnaires were validated with reasonable α. Bias due to missing data Moderate Complete data for 89.6% of baseline participants were included in the analyses, with a comparison between included and missing participants. All analyses were weighted to account for the probability of selection into the original sample and non-response Bias in measurement of outcomes

Low
Obtained objectively from administrative register: 1 hospitalization or 2 reimbursement requests in 2 years (published validation algorithm).

Overall Serious
Souza Santos 2020 Obtained objectively by clinical evaluation.

Overall Critical
Toker 2012

Type of bias Classification Reason -explanation
Bias due to confounding Serious Adjusted for age, sex, education, follow-up time, family history of type 2 diabetes and a postintervention variable that could have been affected by the intervention (BMI). Bias in selection of participants into the study

Moderate
Participation at baseline was 92% without comparison between participants and nonparticipants Bias in classification of interventions

Moderate
Intervention status was well defined: use of validated questionnaires, but without validation of the translation. Bias due to missing data Serious Complete data for 55% of baseline participants were included in the analyses, with information showing that the included and excluded are different. Bias in measurement of outcomes

Moderate
Some diagnoses were obtained objectively from register (administrative data), and some of the diagnoses were obtained from self-reported questionnaires.

Overall Serious
Yamaguchi 2018

Type of bias Classification Reason -explanation
Bias due to confounding Serious Adjusted for age, sex, site, family structure, marital status, occupational category (blue collar or white collar), work status and post-intervention variables that could have been affected by the intervention (components of metabolic syndrome). Bias in selection of participants into the study

Serious
Participation at baseline was 76% without comparison between participants and nonparticipants Bias in classification of interventions

Moderate
Possible reverse causality: prevalent cases only partly excluded (only if two or more criteria for metabolic syndrome were present). Japanese version of questionnaire with confirmed reliability and validity. Bias due to missing data Serious Complete data for 56% of baseline participants were included in the analyses without information on comparison between included and missing participants. Bias in measurement of outcomes

Critical
Outcome was assessed by a clinical test with a cut-off that is not accepted by the ADA diabetes definition: high fasting blood glucose:100 mg/dl.

Overall Critical
Supplemental Figure Legends Suppl. Figure S1. Flow chart for the selection of the included studies. Suppl. Figure S2. Effect of high demands on type 2 diabetes mellitus. This analysis considers demands, whether defined dichotomously or in tertiles (highest versus lowest). It was not possible to transform OR or HR into RR since the original studies did not give estimates for the incidence of diabetes in men and women separately; the original values were therefore used. Since the estimates by Kumari et al. (2004) and Heraclides et al. (2009) are from the same cohort, but based on different baselines, both are included in the meta-analysis. Due to this overlap, the width of the confidence intervals might be underestimated. SE: standard error. CI: confidence interval at 95%. Suppl. Figure S3. Effect of low job control on type 2 diabetes mellitus. This analysis considers low job control, whether defined dichotomously or in tertiles (highest versus lowest). It was not possible to transform OR or HR into RR since the original studies did not give estimates for the incidence of diabetes in men and women separately; the original values were therefore used. Since the estimates by Kumari et al. (2004) and Heraclides et al. (2009) are from the same cohort, but based on different baselines, both are included in the meta-analysis. Due to this overlap, the width of the confidence intervals might be underestimated. SE: standard error. CI: confidence interval at 95%. Suppl. Figure S4. Effect of low social support at work on type 2 diabetes mellitus. This analysis considers low social support at work, whether defined dichotomously or in tertiles (highest versus lowest). It was not possible to transform OR or HR into RR since the original studies did not give estimates for the incidence of diabetes in men and women separately; the original values were therefore used. Since the estimates by Kumari et al. (2004) and Heraclides et al. (2009) are from the same cohort, but based on different baselines, both are included in the meta-analysis. Due to this overlap, the width of the confidence intervals might be underestimated. SE: standard error. CI: confidence interval at 95%. Suppl. Figure S5. Effect of job strain on type 2 diabetes mellitus irrespective of risk of bias. Job strain is included either defined as a dichotomous variable or as a contrast between high strain and low strain quadrants, or as continuous variable, or from the objective job strain matrix of Pan et al. (2017) Suppl. Figure S7. Funnel plot for the effect of job strain on type 2 diabetes mellitus using pooled estimates as published. For each cohort represented in Suppl. Figure S5, the relative risk is plotted against its standard error. Vertical dashed line: overall relative risk estimate from Suppl Figure S6. Suppl. Figure S8. Effect of effort-reward imbalance (ERI) on type 2 diabetes mellitus using original measures of effect. The values used for each study are the hazard ratios resp. odds ratios as published without transformation. SE: standard error. CI: confidence interval at 95%.