STRESSJEM

Clinical Question

Among French employees followed over 26 years, is occupational exposure to shift work and/or night work associated with increased all-cause and cause-specific mortality (cardiovascular, cancer, preventable, and suicide)?
Does the association vary by gender, type of exposure (shift alone, night alone, combined), and temporal exposure measure (current, cumulative, recency-weighted)?

Major Points

Shift without night work showed the strongest and most consistent associations with mortality among men: HR 1.39 (95% CI 1.32–1.46) for all-cause, HR 1.42 (1.26–1.61) for cardiovascular, HR 1.47 (1.35–1.61) for cancer, HR 1.39 (1.19–1.63) for suicide — all significant with p < 0.001
Night without shift work was independently associated with all-cause mortality among men (HR 1.12, 95% CI 1.06–1.18, p < 0.001) and cardiovascular mortality (HR 1.17, 1.02–1.34, p < 0.05)
Combined night and shift work was associated with all mortality outcomes among men, but contrary to expectations, was not more harmful than either exposure alone
Among women, shift without night work increased all-cause mortality (HR 1.13, 1.03–1.25, p < 0.01), cancer mortality (HR 1.21, 1.05–1.41, p < 0.05), and breast cancer mortality (HR 1.48, 1.12–1.94, p < 0.01)
Cerebrovascular disease mortality was elevated for both genders: HR 1.47 (1.11–1.94) for men with shift work and HR 2.37 (1.03–5.46) for women with night work
When follow-up was extended beyond the last job, associations generally attenuated, suggesting some reversibility of exposure effects, though among women new associations emerged for night work and all-cause and preventable mortality, suggesting delayed effects

Guidelines

IARC (2020) classifies night shift work as Group 2A — probably carcinogenic to humans, based on limited evidence for cancer in humans, sufficient evidence in animals, and strong mechanistic evidence from circadian disruption
EU Working Time Directive (2003/88/EC) limits night worker hours to an average of 8 per 24-hour period and mandates regular health assessments
No specific guideline addresses shift work and all-cause or cardiovascular mortality prevention — this study adds to the evidence base that may inform future occupational health recommendations

Design

Prospective cohort study using the STRESSJEM (STRESs au travail et pathologies cardiovasculaires: le projet JEM) project
Three linked data sources: (1) DADS-INSEE panel — a 1/24th random sample of the French national working population with complete job histories 1976–2002; (2) SUMER survey — occupational exposure data used to construct the job-exposure matrix; (3) INSERM-CépiDc national mortality registry via the COSMOP linkage program
Exposure assessed through a validated JEM constructed using CART segmentation on SUMER survey data (49,984 employees), assigning probability of shift and night work based on occupation, economic activity, and company size
Three time-varying exposure measures: current (probability at any given day), cumulative (mean daily probability over entire follow-up), and recency-weighted cumulative (5-year weighted window favoring recent exposure)
Binary exposure classification: dichotomized at JEM probability thresholds of 0.42 for shift work and 0.255 for night work
Cox proportional hazards models with age as the time scale, adjusted for calendar time and biomechanical, physical, chemical, and biological occupational exposures
Gender-stratified analyses throughout
Sensitivity analyses: continuous JEM exposure, lowest exposure imputation for multiple jobholders, adjustment for occupation

Population

Initial cohort: 1,511,456 employees from the DADS-INSEE panel
Analyzed: 1,496,332 (806,317 men, 704,789 women) after excluding 15,214 with completely missing job histories
Mean age at cohort entry: 28 years (men), 27 years (women)
Mean follow-up duration: 17 years
22,105 deaths during follow-up until end of last job (17,826 men, 4,279 women for all-cause mortality)
118,748 deaths when follow-up extended to end of study period (89,554 men, 29,194 women)
Men were more likely than women to be exposed to shift and night work

Interventions

Exposed group: Employees classified as exposed to shift work (probability ≥ 0.42) and/or night work (probability ≥ 0.255) by the JEM
Shift work: defined as any shift system; night work: any working period between midnight and 5:00 AM more than 52 nights/year (ILO definition)
Three exposure combinations studied: shift without night, night without shift, and combined night and shift
Reference group: Employees with JEM probability below both thresholds — neither shift nor night work exposure
Exposure was time-varying across the follow-up period, allowing for job changes

Outcomes

All-cause mortality (primary): 22,105 deaths until end of last job; 118,748 deaths until end of follow-up
Cardiovascular mortality (ICD-10 I00–I99): 3,092 deaths (men) and 486 deaths (women) — ischemic heart disease and cerebrovascular disease analyzed separately
Cancer mortality (ICD-10 C00–C99): 5,733 deaths (men) and 1,907 deaths (women) — respiratory/intrathoracic malignancies and breast cancer analyzed separately
Suicide: 1,638 deaths (men) and 366 deaths (women)
Preventable mortality (OECD/Eurostat definition): 11,924 deaths (men) and 1,974 deaths (women) — includes smoking-related, alcohol-related, and external causes of death
Key findings among men (current exposure, shift without night work): all-cause HR 1.39 (1.32–1.46), cardiovascular HR 1.42 (1.26–1.61), cancer HR 1.47 (1.35–1.61), suicide HR 1.39 (1.19–1.63), preventable HR 1.39 (1.31–1.48) — all p < 0.001, meaning 39–47% increased hazard of death across all studied outcomes
Among women, notable findings include breast cancer mortality HR 1.48 (1.12–1.94, p < 0.01) and cerebrovascular mortality HR 2.37 (1.03–5.46, p < 0.05) with night work

Criticisms

Exposure misclassification: The JEM assigns group-level exposure probabilities, not individual exposure. This introduces non-differential misclassification and imprecision, known to bias effect estimates toward the null. The JEM for night work had lower validity among women specifically
Residual confounding: Models adjusted for other occupational exposures but not for individual lifestyle factors (smoking, alcohol, BMI, diet, physical activity) or direct socioeconomic status. Shift workers tend to have lower SES and higher smoking prevalence, and the strong association with smoking-related mortality (HR 1.52 in men) raises concern about uncontrolled behavioral confounding
Healthy worker effect: Selection effects to enter and remain in shift/night work (self-selection of healthier workers with greater tolerance) likely underestimate the true association. This is partially mitigated by the cumulative exposure analyses
Exposure definition limitations: No data on permanent vs. rotating shifts, direction of rotation, number of consecutive nights, or start/end times of shifts. Night work was defined broadly (≥52 nights/year) without dose gradation
Limited observation window: Job history data covered only 1976–2002, preventing assessment of lifetime exposure. Mid-censoring was used for rare periods of missing data rather than imputation
Underpowered for women: Many exposure-outcome associations among women had wide confidence intervals crossing the null, particularly for night work and combined exposure categories with small numbers of deaths
No causal inference: As an observational cohort, causal conclusions cannot be drawn despite the large sample size and long follow-up. The time-varying Cox models address temporality but cannot establish causation
Historical cohort: Data are from 1976–2002; working conditions, shift patterns, and occupational health protections have evolved since then, potentially limiting applicability to current workforces

Funding

Supported by the French National Research Program for Environmental and Occupational Health of ANSES (Agence nationale de sécurité sanitaire de l'alimentation, de l'environnement et du travail), Grant Number EST-2016/1/49
Government-funded with no industry involvement
The authors report no conflicts of interest
Ethical approval from Commission Nationale de l'Informatique et des Libertés (no 762430 V1 and no 04-1274) and Conseil National de l'Information Statistique (no 2009X705TV)

Key Results

Shift and/or night work exposure

JEM-assessed probability of exposure dichotomized at 0.42 (shift) and 0.255 (night)

Neither shift nor night work

Reference group: below JEM exposure threshold

Primary Endpoint

All-cause mortality (men, shift without night work, current exposure)

Metric	Value
Outcome type	time-to-event
HR	1.39 (95% CI 1.32–1.46)
P-value	<0.001

Secondary Outcomes

Outcome	Shift and/or night work exposure	Neither shift nor night work	Effect	95% CI	P-value
All-cause mortality — men, night without shift work (current)	NR%	NR%	HR 1.12	1.06–1.18	<0.001
All-cause mortality — men, night and shift work (current)	NR%	NR%	HR 1.24	1.18–1.30	<0.001
All-cause mortality — women, shift without night work (current)	NR%	NR%	HR 1.13	1.03–1.25	<0.01
Cardiovascular mortality — men, shift without night work (current)	NR%	NR%	HR 1.42	1.26–1.61	<0.001
Cardiovascular mortality — men, night and shift work (current)	NR%	NR%	HR 1.29	1.14–1.46	<0.001
Cancer mortality — men, shift without night work (current)	NR%	NR%	HR 1.47	1.35–1.61	<0.001
Cancer mortality — women, shift without night work (current)	NR%	NR%	HR 1.21	1.05–1.41	<0.05
Suicide — men, shift without night work (current)	NR%	NR%	HR 1.39	1.19–1.63	<0.001
Preventable mortality — men, shift without night work (current)	NR%	NR%	HR 1.39	1.31–1.48	<0.001
Breast cancer mortality — women, shift without night work (current)	NR%	NR%	HR 1.48	1.12–1.94	<0.01
Cerebrovascular disease mortality — women, night without shift work (current)	NR%	NR%	HR 2.37	1.03–5.46	<0.05

Visualizations

Subgroup Analysis

Discussion Questions

methodology

1. How does the use of a job-exposure matrix (JEM) to assess shift and night work exposure affect the validity of the findings compared to individual-level exposure assessment?

JEM assigns group-level probabilities of exposure based on job title, introducing non-differential misclassification that typically biases results toward the null. This is a fundamental methodological trade-off in large cohort studies.

Suggested Answer

JEMs assess exposure at the group level (job title), not the individual level, leading to within-group misclassification and reduced exposure variance. This non-differential misclassification is well-documented to bias effect estimates toward the null, meaning the true associations may be stronger than reported. However, JEMs eliminate recall bias and self-report bias, and are the only feasible method for cohorts of this size without prospective individual exposure measurement.

statistics

2. The three exposure metrics (current, cumulative, recency-weighted cumulative) produced similar AIC values and effect estimates. What are the implications for understanding dose-response and latency?

Different temporal exposure metrics should theoretically capture different aspects of the exposure-disease pathway (acute vs. cumulative effects), and their convergence is notable.

Suggested Answer

The similar AIC values suggest that none of the three temporal models fits the data substantially better than the others, which may reflect limited contrast between metrics when exposure is relatively stable over time in occupational cohorts. The convergence of results across metrics provides robustness but limits our ability to distinguish acute from cumulative dose-response relationships. The finding that mortality associations weakened when follow-up extended beyond the last job suggests some reversibility, which would favor current over cumulative exposure models.

clinical applicability

3. Given the associations between shift work and smoking-related mortality and external causes of death, how should we interpret whether shift work is a direct risk factor or a marker for behavioral mediators?

The strong associations with smoking-related causes and external causes (accidents/injuries) suggest that behavioral pathways (smoking, fatigue-related accidents) may mediate the shift work-mortality association, which has different prevention implications than direct biological mechanisms.

Suggested Answer

The associations with smoking-related mortality (HR 1.52 in men) and external causes (HR 1.15 in men) suggest that behavioral pathways are important mediators. Shift workers have higher smoking prevalence, altered eating patterns, and increased accident risk due to fatigue and circadian disruption. These behavioral mediators are modifiable targets for prevention. However, the concurrent associations with cardiovascular and breast cancer mortality suggest direct biological mechanisms (circadian disruption, hormonal changes) also contribute. Prevention should address both pathways: workplace health programs targeting smoking and lifestyle, plus schedule optimization to minimize circadian disruption.

external validity

4. The study found substantially weaker associations among women, with many non-significant results. Does this reflect a true gender difference in susceptibility or a methodological limitation?

Women had fewer deaths, lower exposure prevalence, and the JEM for night work had lower validity among women — all of which reduce statistical power and may mask true effects.

Suggested Answer

The weaker findings among women are most likely due to methodological limitations rather than true gender differences in susceptibility. Women had far fewer deaths (4,279 vs. 17,826 for all-cause), lower exposure prevalence (they were less likely to be exposed to shift and night work), and the JEM for night work had lower validity among women specifically. These factors collectively reduce statistical power. The authors note the study was 'underpowered to detect all exposure-outcome associations, especially among women.' Notably, where power was adequate (breast cancer, cerebrovascular disease), significant associations emerged among women, supporting the view that power — not biology — explains the gender discrepancy.

ethics or controversy

5. The study adjusted for occupational exposures (biomechanical, physical, chemical, biological) but not for lifestyle factors like smoking, alcohol, diet, or socioeconomic status directly. How does this residual confounding affect interpretation?

Occupational cohort studies using administrative data often lack individual-level lifestyle and socioeconomic confounders, raising questions about the attributability of observed associations to shift/night work versus correlated social determinants.

Suggested Answer

The absence of individual-level adjustment for smoking, alcohol use, BMI, diet, and direct socioeconomic status is a significant limitation. Shift workers tend to have lower socioeconomic status, higher smoking rates, and less healthy dietary patterns. The occupational exposures (physical, chemical, biological) serve as partial proxies for job type and social position, but residual confounding by lifestyle factors likely inflates the effect estimates. The strong associations with smoking-related mortality specifically support this concern. While the authors note that JEM-based assessment tends to underestimate true effects, residual confounding works in the opposite direction, and disentangling these biases is not possible with the available data.

Overall Assessment

CONFIRMATORY moderate confidence

Large prospective cohort (1.5 million) with 17-year follow-up confirms associations between shift/night work and multiple mortality outcomes, strengthening existing evidence from smaller studies and meta-analyses, but the observational design and JEM-based exposure assessment preclude causal conclusions.

Clinical Question

Major Points

Guidelines

Design

Population

Interventions

Outcomes

Criticisms

Funding

Further Reading

Key Results

All-cause mortality (men, shift without night work, current exposure)

Secondary Outcomes

Visualizations

Subgroup Analysis

Discussion Questions

Overall Assessment

Further Reading