Work stress in the etiology of coronary heart disease—a meta-analysis

If you spend a third of your waking life at work, and if the stress of that work raises your heart attack risk by fifty percent, then your job is not just an economic fact; it is a cardiac one. That is not a hypothesis. It is the result of Kivimäki and colleagues pooling data from eighty-three thousand workers across fourteen prospective cohort studies, published in the Scandinavian Journal of Work, Environment and Health in 2006. The question they set out to answer was straightforward: does work stress actually cause coronary heart disease? The answer turned out to be complicated in exactly the ways that matter. To understand what Kivimäki and colleagues were measuring, you need to know that work stress is not one thing. The research field had developed three distinct theoretical models, each capturing a different way a job can become a threat to health. The first is the job-strain model, built around two dimensions: high demands and low control. If your work asks a lot of you but gives you little say over how you do it, you are in a job-strain situation. An extended version adds social support as a third factor — the worst position being high demands, low control, and social isolation at work, a combination researchers call iso-strain. The second model is effort-reward imbalance, developed by Siegrist. Working hard and receiving little back, whether in pay, esteem, or job security, violates a basic expectation of reciprocity. Overcommitment to work amplifies that imbalance. The third model focuses on organizational injustice: when the procedures governing your workplace are unfair, or when supervisors treat workers without consideration or respect, that sustained sense of being wronged is itself a stressor. These three models are not redundant. They carve up the toxic-job concept from genuinely different angles, and as Kivimäki and colleagues showed, they produce different empirical signals. The meta-analysis pulled together prospective cohort studies published through January 2006. Prospective design matters here — you measure stress exposure first, then watch who develops heart disease over time. That structure avoids the bias of asking sick people to reconstruct their work histories. From an initial screen of fifty-eight articles, fourteen independent cohort studies met the eligibility criteria. The job-strain analysis drew on eighty-three thousand employees, the effort-reward imbalance analysis on eleven thousand five hundred and twenty-eight, and the organizational injustice analysis on seven thousand two hundred and forty-six. Follow-up durations across the cohorts ranged from roughly four to twenty-six years. Results were pooled using a random-effects approach with inverse-variance weighting, and the team reported two levels of adjustment: age-and-gender adjusted estimates and fully multiple-adjusted estimates that added standard cardiovascular risk factors, including blood pressure, cholesterol, smoking, body mass index, physical activity, and socioeconomic position, depending on the study. Now for the numbers. Across all three models, the age-and-gender-adjusted summary risks clustered near a fifty percent excess risk of coronary heart disease, or CHD. For job strain, that figure was a relative risk of 1.43, with a ninety-five percent confidence interval of 1.15 to 1.84. For effort-reward imbalance, it was 1.58. For organizational injustice, it was 1.62. That convergence across three independent models, each measuring something different, is the headline finding. But the story immediately gets more complicated. When Kivimäki and colleagues applied full covariate adjustment to the job-strain estimate, it dropped to 1.16, with a confidence interval of 0.94 to 1.43. That crosses one. It is no longer statistically significant. On the surface, that looks like the job-strain association evaporating under scrutiny. But the authors flag a critical interpretive problem: when you adjust for variables like blood pressure and cholesterol, you may not be removing confounders. You may be subtracting the pathway. If chronic stress raises your blood pressure, and high blood pressure damages your heart, then adjusting for blood pressure removes part of the mechanism — not an unrelated nuisance variable. The paper lists many plausible mediators: sleep disturbance, weight gain, reduced physical activity, smoking intensity, hemostatic changes, and impaired heart-rate variability — and suggests that adjusting them away could explain exactly why the job-strain estimate shrinks. The attenuation is real; what it means is genuinely contested. The effort-reward imbalance results told a different story. That association did not shrink after adjustment. The multiple-adjusted pooled relative risk remained above two in sensitivity analyses — one combination of studies yielded 2.51, with a confidence interval of 1.58 to 3.98. For organizational injustice, the multiple-adjusted relative risk held at 1.47, with a confidence interval of 1.12 to 1.95, remaining statistically significant even after additional controls for job strain and effort-reward imbalance. These two models showed a signal that standard covariate adjustment did not explain away. There is also a statistical wrinkle worth naming. Heterogeneity between studies was real and significant — for job strain, the Q statistic was 17.5 with a p-value of 0.04. The studies did not agree with each other. Some of that disagreement traces back to how exposure was measured. Instruments varied across cohorts; some studies modified standard questionnaire scales, and some cohorts imputed job strain from occupational titles rather than asking workers directly. The Honolulu study, for instance, used occupation-based imputed scores. Nearly all cohorts also measured stress at a single point in time, which is a problem for a disease with a long latency. In the Whitehall II cohort, applying a regression-dilution correction — essentially accounting for the fact that a single measurement underestimates long-term exposure — made the excess risk estimate thirty percent higher than the uncorrected figure. That is a large difference. It suggests the true association may be consistently underestimated across the literature. The gender gap in this evidence base is also worth pausing on. The vast majority of participants and coronary heart disease events in these studies were male. There were too few female-only samples to formally test gender differences for the effort-reward imbalance and injustice models. The findings, as they stand, skew heavily toward men, which is a real constraint on how broadly these conclusions can apply. Kivimäki and colleagues are direct about what their meta-analysis can and cannot establish. It can document associations. It cannot prove that reducing work stress will reduce heart disease rates — that requires intervention studies, and those had not been done. They note that work stress does not appear on the American Heart Association's list of established coronary heart disease risk factors, and they call for large-scale intervention studies with long follow-up. One example they cite: organizational downsizing was associated with increased cardiovascular mortality, with the greatest excess risk in the years immediately following downsizing. That kind of natural experiment points toward causality, but it is not proof. The honest place to leave this is exactly where Kivimäki and colleagues leave it. Fourteen prospective cohort studies, eighty-three thousand workers, three independent theoretical models — all pointing toward roughly a fifty percent excess risk of coronary heart disease among employees with elevated work stress. That is not noise. But the consistency of the association sits alongside genuine uncertainty: heterogeneous measurement, the mediation problem in covariate adjustment, sparse data for women, and the absence of intervention evidence. The work stress and heart disease story is one where the signal is strong enough to take seriously, and the evidence base still too limited to act on with full confidence. That gap between a plausible, consistent association and confirmed causal knowledge is exactly where the next generation of research needs to go. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.