Sunday, March 20, 2005

Job Strain and Heart Attack: Not as null as presented.

BrooklynDodger here presents a null study for the association of job strain with adverse health outcomes. The study is not as null as the abstract says. The Dodger again exhorts reviewers and editors to pay more attention to abstracts.

The Belgian study reports 87 cardiac events over 3 years in 14,377 middle aged men. [These are a lot of people, but very few events, especially distributed over 3 or 4 levels of fuzzy variables.] 17% of the workers experienced high strain. The investigators concluded job demands and decision latitude were not significantly related to the development of coronary heart disease. The investigators noted that a 38% risk excess among subjects in the high-strain category did not reach statistical significance [but neglected to include the confidence interval in the abstract. The Dodger would have to go into the full text paper to find the number of cases and the confidence interval.] The investigators noted that coronary heart disease incidence was substantially associated with the social support scale [The Dodger assumes they meant risk goes up as social support goes down. Doesn’t anyone edit these abstracts? They could have said “negatively correlated,” which would be poor diction but at least accurate] independently of other risk factors.

The Dodger feels the conclusion “No convincing evidence for an association of … job strain with the short-term incidence of coronary heart disease was found.” to be unfounded. The upper CI of the high job strain was 2.38, based on 18 cases. A 38% increased risk of heart attack, which could have been a more than doubling a risk is nothing to blow off, when the study clearly had little power. It would have been more accurate to say that the study, which may have been inadequate, provided “equivocal” or “some” evidence.

A perception of low social support was indeed associated with higher risk. The Dodger has glanced at a version of this standard job content questionnaire. There are limited objective elements in the job demands questions [for example “I have to work very fast”]. The social support questions pretty much elicit subjects’ subjective response to the environment. Well, the paper does say “perceived” job stress.


Am J Epidemiol. 2005 Mar 1;161(5):434-41.

Perceived Job Stress and Incidence of Coronary Events: 3-Year Follow-up of the Belgian Job Stress Project Cohort.

De Bacquer D, Pelfrene E, Clays E, Mak R, Moreau M, de Smet P, Kornitzer M, De Backer G.Department of Public Health, Ghent University, Ghent, Belgium.

1 comment:

gribley said...

Good point. "No convincing evidence for an association" is a common, but somewhat misleading, summary of a dataset in which the outcome doesn't rise to statistical "significance," probably from a fairly low number of cases. In fact, job demands and job strain both show reasonable dose-response relationships consistent with a causal connection.

At least they didn't say that they proved no association -- which many have claimed with similar, non-"significant" data.

(On writing style, note that there's an editorial in the same issue of AJE.)