An orbiter of the occupation cluster and the psychosocial channel of occupational exposure —
distinct from the cluster’s nucleus The Physical Activity Paradox, which is the physical-demand
channel. Job strain is Karasek’s demand-control construct: «Job strain—the combination of high job
demands and low control at work—is one of the most widely studied defi nitions of psychosocial stress»
(Kivimäki et al., 2012). The finding:
job strain carries a small but consistent excess risk of incident coronary heart disease — a real,
credibly-non-artifactual association that is nonetheless a minor lever against the standard risk
factors. Held single-source (a gold IPD meta-analysis), confidence: low.
(inferred from Kivimäki et al., 2012)
The effect estimate
Kivimaki 2012 is a collaborative individual-participant-data (IPD) meta-analysis (IPD-Work Consortium): 13 European cohorts, 197,473 CHD-free employed adults, 1.49 M person-years, mean follow-up 7.5 y, 2358 incident CHD events (first non-fatal MI or coronary death).
- Exposure / comparator: job strain (high demands AND low control) vs no job strain — a binary contrast on study-median splits, self-reported at baseline.
- Outcome: incident CHD event (non-fatal MI + coronary death) — a harder, narrower endpoint than the all-cause / CVD mortality used on the sibling stress pages (Allostatic Load and Mortality, Stress Management and Cardiometabolic Health).
- Effect (relative): «the hazard ratio for job strain versus no job strain was 1·23 (95% CI 1·10–1·37). This eff ect estimate was higher in published (1·43, 1·15−1·77) than unpublished (1·16, 1·02−1·32) studies.» After adjustment for socioeconomic status it «attenuated to 1·17 (1·05–1·31)»; further adjustment for lifestyle (BMI, PA, smoking, alcohol) or Framingham score did not materially change it. (Kivimäki et al., 2012)
- Population-attributable risk: «The population attributable risk for job strain was 3·4%.» (assuming causality, at ~15% strain prevalence). (Kivimäki et al., 2012)
- Certainty: LOW — observational cohorts only (no RCT possible), self-reported exposure, binary strain definition losing the demand/control gradient. Gold design (IPD-MA correcting publication bias), but the magnitude inherits the observational confounding structure.
- Studied contrast, not a curve: high-vs-no-strain binary; there is no dose-response shape here, so no knee/plateau claim is available.
Absolute framing (illustrative,). The source reports no absolute risk. At the cohort’s baseline CHD incidence of ~15.8 per 10,000 person-years, a 1.23 relative hazard implies roughly 3-4 extra CHD events per 10,000 person-years in strained workers (~19 vs ~15 per 10,000 py) — on the order of 1 additional event per ~300-400 strained workers over the ~7.5-y follow-up. A modest relative effect on a modest baseline is a small absolute effect; the excess scales with the worker’s baseline risk (route-(a), Baseline Risk and the Relative-Absolute Split) — larger for an older/higher-risk worker. (inferred from Kivimäki et al., 2012)
Why the association is credible — three artifact defences the IPD design supplies
The prior job-strain literature was contested (single studies showed «more than a doubling»; a prior meta-analysis ~40%). The IPD design attacks the three standard objections, and the association survives all three:
- Publication bias — corrected downward. Published studies gave 1.43, unpublished 1.16; pooling both (only an IPD collaboration can retrieve the unpublished cohorts) yields 1.23. The published-only literature was inflated — the honest estimate is the lower, pooled one. (Kivimäki et al., 2012)
- Reverse causation — ruled out by lag exclusion. «Exclusion of coronary heart disease cases at the fi rst 3 years and 5 years of follow-up to minimise reverse causality slightly strengthened the association» (3-y 1.31; 5-y 1.30). (Kivimäki et al., 2012) The signature of reverse causation is weakening when early events are dropped; job strain does the opposite, so early (potentially disease-driven) exposure reports are not manufacturing the effect. This is the The U-Shaped Association Artifact check run and passed on the harm arm.
- Confounding — SES carries part, lifestyle none. Full adjustment leaves 1.17, and behavioural risk factors do not mediate it — so the strain->CHD association is not simply strained workers smoking or sitting more. SES is the one material confounder/pathway (crude 1.23 -> SES-adjusted 1.17).
Consistency is high: broadly similar HRs across sex, age, SES, and region, with negligible pooled heterogeneity (I2 <0.1%). (Kivimäki et al., 2012)
Layer-1 ranking — a small lever, and the source says so
The decision-relevant fact is not whether job strain raises CHD risk but how much it is worth acting on relative to the big rocks. Kivimaki states the ranking directly: «prevention of workplace stress might decrease disease incidence; however, this strategy would have a much smaller eff ect than would tackling of standard risk factors, such as smoking.» (Kivimäki et al., 2012)
- A PAF of 3.4% puts job strain far below the dominant modifiable CHD levers (smoking, blood pressure, lipids), whose population-attributable fractions are multiples larger. This is the ceiling is a finding rule made concrete: for a worker who still smokes, has untreated hypertension, or is inactive in leisure, workplace-stress reduction is not the next move — the big rock is.
- The exposure earns attention as a stratum-level refinement, not a headline lever: for someone who has already pulled the large levers (non-smoker, controlled BP, active), a modest consistent occupational-stress risk is among the smaller remaining gaps — precisely the marginal question the ranking says becomes relevant only after the big rocks. Attention is an anti-signal applies: work-stress is heavily discussed relative to its ~3% PAF.
Occupation is at least two exposures — the cross-cluster decomposition
Job strain and the physical-activity paradox are different occupational exposures that reach the same organ system by different routes, so “occupation” is not one lever for a cardiovascular decision. Same-quantity check against the cluster nucleus:
| Parameter | Occupational physical activity (Coenen, nucleus) | Job strain (Kivimaki, this page) | Same quantity? |
|---|---|---|---|
| Exposure | high physical demand at work (>=40 h/wk load) | high psychological demand + low control | NO — physical vs psychosocial |
| Proposed mechanism | sustained HR/BP load, no recovery, sub-fitness stimulus | chronic HPA/sympathetic activation (stress physiology) | NO — different pathway |
| Outcome measured | all-cause mortality | incident CHD event | NO — different endpoint |
| Direction | higher mortality in men only (HR 1.18) | higher CHD risk, both sexes (HR 1.23) | same sign, different exposure |
| Sex pattern | sex-discordant (null/inverse in women) | broadly similar men and women | NO |
Because «same quantity?» is NO on exposure, mechanism, and endpoint, this is a type-B/C decomposition (one occupational context -> at least three distinct causal channels), not a duplicate of the nucleus and not a tension with it — the findings are consistent and additive, each a separate channel of occupational risk. A worker can carry any combination. The third channel is cognitive stimulation (Cognitive Stimulation at Work and Dementia), and it stands in an especially sharp relation to this page — see below. (Coenen et al., 2018; inferred from Kivimäki et al., 2012, 2021)
The demand-control model yields two opposite-signed exposures — job strain is one pole
Job strain is one quadrant of Karasek’s demand-control model, not the whole of it. The model crosses a demand axis with a control (decision-latitude) axis, and the sibling Cognitive Stimulation at Work and Dementia holds the opposite control-axis pole:
| Karasek quadrant | Demand | Control | This corpus | Direction |
|---|---|---|---|---|
| High-strain (this page) | high | low | job strain -> CHD | harmful (HR 1.23) |
| Active (sibling page) | high | high | cognitive stimulation -> dementia | protective (HR 0.77) |
So high work demand is not one exposure — decision latitude (control) is the axis that flips it from a CHD risk (low control) to a dementia-protective factor (high control), on different endpoints by different biology (HPA/sympathetic here vs cognitive-reserve/neurodegeneration there). Reading demanding job = bad collapses two opposite exposures; the control axis is doing the work. This is a distinction, not a tension — the two answer different questions (which quadrant, which outcome) and are consistent (not-joined check (ii): different unit/endpoint). (inferred from Kivimäki et al., 2021)
The mechanism — a candidate mediator on the HPA/allostatic-load spine
Job strain is a chronic psychosocial stressor, and Allostatic Load and Mortality (the psychosocial
cluster nucleus) already names job strain among the chronic-stress exposures that act through cumulative
multi-system dysregulation to reach a physical outcome. The plausible chain is job strain -> sustained
HPA/sympathetic activation -> allostatic load -> CHD, the same spine on which social isolation and low
purpose already sit as candidate mediators. This makes job strain a third distinct exposure routing
through the AL spine — and the AL page’s marker-not-lever caution transfers: the mediation is an
unmeasured bridge, not a demonstrated one (Kivimaki measures no AL index; the lifestyle-adjustment
result shows the pathway is not behavioural, consistent with but not proof of a direct-physiology
route).
(inferred from Kivimäki et al., 2012)
Where it changes a decision — and where it does not
- In scope (appraise): for a worker who has already handled the big rocks, high job strain is a small, credible, consistent CHD risk factor — a stratum-level refinement worth naming, not a headline.
- Realistic intervention is workplace/organizational (raising control / lowering demands), which the wiki does not evidence for a hard CHD outcome — no RCT of workplace-stress reduction reports CHD events. Individual stress-management (e.g. MBSR, Stress Management and Cardiometabolic Health) is a different, low-certainty, surrogate-only lever and is not shown to move CHD incidence.
- Out of scope: diagnosing or treating an established cardiac condition; prescribing.
Confidence: LOW — and the gaps
- Single gold source, observational, self-reported exposure, binary definition.
AWAITSan independent job-strain->CHD SR/MA and any guideline position before a recommendation-level read. G-gaps. (i) No trial shows that reducing job strain reduces CHD — the marker/lever gap, same as the AL nucleus. (ii) Residual SES confounding: SES is partly a pathway and partly a confounder, and the 1.23->1.17 attenuation cannot fully separate them. (iii) The AL-mediation bridge is unmeasured. (iv) The demand/control gradient is collapsed to a binary — the shape of any dose-response is unknown.- Coherence, not validity (R1): the association is what these cohorts report; the open loop (no realized-outcome check of a strain-reduction intervention) stands.
Self-critique [run 2026-08-14, before commit]
- Not laundered from one source restated. The beyond-summary moves are (a) the cross-cluster decomposition — occupation resolves into a physical-demand channel and a psychosocial channel that are non-comparable on exposure/mechanism/endpoint (parameter table is the guard), and (b) the spine-extension placing job strain as a third candidate mediator on the AL/HPA spine. Both are made against held fabric (the PA-paradox nucleus, the AL nucleus), not restated from Kivimaki.
- Independence discipline held. Neither cross-source move is tagged
[E-independent]: the Coenen and AL links are different exposures, not one claim reached by two independent backings — explicitly flagged at each. - Not overclaimed. Confidence LOW; every causal step is hedged; the mediation is an unmeasured bridge; the absolute framing is flagged illustrative-INFERRED; the Layer-1 ranking is Kivimaki’s own words.
- Big-rock ranking is the decision-change, not the bare association — the page’s job is to place the ~3.4% PAF below smoking/BP/lipids and license stop optimizing this until the big rocks are handled.