Is job strain a major source of cardiovascular disease risk?

Empirical studies on job strain and cardiovascular disease (CVD), their internal validity, and the likely direction of biases were examined. The 17 longitudinal studies had the highest validity ratings. In all but two, biases towards the null dominated. Eight, including several of the largest, showed significant positive results; three had positive, nonsignificant findings. Six of nine case–control studies had significant positive findings; recall bias leading to overestimation appears to be fairly minimal. Four of eight cross-sectional studies had significant positive results. Men showed strong, consistent evidence of an association between exposure to job strain and CVD. The data of the women were more sparse and less consistent, but, as for the men, most of the studies probably underestimated existing effects. Other elements of causal inference, particularly biological plausibility, corroborated that job strain is a major CVD risk factor. Additional intervention studies are needed to examine the impact of ameliorating job strain upon CVD-related outcomes.

Clinicians are often called upon to assess the cardiovascular work fitness of patients. As they attempt to make an informed judgment, a fundamental question arises: is the work environment fit, or conducive, to cardiovascular health? With technological advances, jobs characterized purely by heavy physical demands have become progressively less common. New types of workrelated challenges and burdens primarily affecting the higher nervous system of humans (ie, psychosocial stressors) are more and more frequently encountered. Yet most of the clinical guidelines relevant to the interface between the workplace and the patient's cardiovascular system continue to focus upon levels of physical exertion.
Hu & Speizer (1) underscored the importance of identifying job-related and other environmental hazards that contribute to a given disease process. They noted that "physicians commonly treat the sequelae of such disease in the practice of medicine; however, unless the underlying connection with hazardous exposures is identified and mitigated, treatment of the manifestations rather than the cause at best only ameliorates the condition. At worst, the neglect of hazardous exposures may lead to both failure of treatment and failure to recognize a public health problem with wide significance [p 19]." Several decades ago, occupational and environmental health research raised the concern that exposure to psychosocial stressors in the modern work environment may be related to cardiovascular disease (CVD). It was clear, however, that the evidence would be difficult to obtain, that a myriad of thorny methodological problems would arise, and that the critical obstacle would be the theoretical conceptualization, modeling, and measurement of workplace stressors. A major breakthrough came in 1979 with the introduction of the job strain (demand-control) model (2). The model was developed for work environments in which stressors are "chronic, not initially life-threatening and the product of sophisticated human organizational decision making. In decision making the controllability of the stressor is critical, and it becomes more important as increasingly complex and integrated social organizations develop, with ever more complex limitations on individual behavior [p 78]". The model has two components: "psychological demands, and a combined measure of task control and skill use, or decision latitude [p 78]". Job strain occurs when the human organism is overloaded psychologically and at the same time deprived of control over the work environment, a combination which is predicted to give rise to increased risk of stress-related illness (3, p 78). The basic components of the two dimensions are summarized in table 1. A third dimension, social isolation, was later added to the model, with the worst situation being "iso-strain": high demands, low decision-making latitude, plus lack of social support (5).
Exposure to job strain can be assessed from self-report via a questionnaire, with the dimensions operationalized in the form of short, general instruments, most frequently the job content questionnaire (JCQ) or the psychosocial job strain questionnaire (PSJSQ) (6-8). These measures are feasibly administered in field and epidemiologic studies. Data linkage methods have been developed in the United States and Sweden, so that exposure to job strain (as well as to "iso-strain" in Sweden) can also be inferred from occupational titles alone [ie, the imputation method (9-10)]. External assessment of job characteristics (eg, expert observer) is yet another method for obtaining exposure data. [See the work of Greiner & Krause (11).] The job-strain model has been the model most widely used for evaluating the psychosocial work environment and its potential impact upon the cardiovascular system, with some studies incorporating the third dimension of social isolation as well. Since the introduction of the model, many empirical investigations have been published concerning the relation between job strain and CVD outcomes, including acute myocardial infarction (MI), coronary artery disease (CAD), and CVD-related mortality. Many of these studies report significant positive findings, and job strain is increasingly receiving attention as a potential contributor to . On the other hand, there have been several nonconfirmatory findings concerning job strain and CVD outcomes published in largescale studies. These results spurred some questions concerning the strength and consistency of the evidence.
Several in-depth reviews (13,(15)(16)(17)(18)(19) have been published concerning the empirical data on the etiologic role of psychosocial workplace factors (in some of these also outside work) and CVD. One of these reviews (16) focused explicitly on job strain. However, a comprehensive and systematic assessment of the internal validity of the current body of reported results is needed. A key issue yet to be addressed is the direction in which methodological issues would most likely be acting, delineating situations that would increase the likelihood of obtaining null results and those that could lead to an overestimation of any association.
In this paper, we have used a predefined set of criteria to examine each of the empirical investigations on job strain and CVD. The criteria were developed specifically to assess the methodological issues affecting the internal validity of studies on this topic, and, whenever possible, to identify the direction in which the results would most likely be affected. We have not only addressed the strength and consistency of the association, but also the other major elements of causal inference as well. Particular attention has been devoted to exploring the viability of alternative hypotheses, as well as the question of biological plausibility (ie, what are the possible mechanisms through which job strain could affect the cardiovascular system). This critical review has thereby sought to provide a more definitive answer to the question "Is job strain a major CVD risk factor?" The clinical implications of the conclusion have also been explored.

Identification and inclusion of relevant studies
Search strategy A computer-based search was carried out using Medline, via OVID, from 1966 to January 2002. The search terms were entered as text words in the title, abstract, or other searchable fields (mesh terms, etc). For the independent variable, the search terms were job strain, iso-strain, decision latitude, psychological demands, work control, job control, skill discretion, decision-making authority and intellectual discretion. These terms were combined with search terms for the dependent variable: myocardial infarction, angina pectoris, ischemic heart disease, coronary artery disease, and cardiovascular death. A senior medical information specialist replicated this strategy. Bibliographies of relevant articles and personal files were also reviewed.

Inclusion criteria and procedure
Articles were chosen for review if they fulfilled the following criteria: (i) exposure to job strain was assessed or imputed via its two major workplace dimensions, psychological demands plus any of the following: decision latitude, skill discretion, decision authority and decision control, as these directly relate to the dependent variable, (ii) any of the following were included as the dependent variable: CAD, manifestations of ischemic heart disease (IHD) (angina pectoris, MI) or mortality from cardiovascular causes, (iii) a case-control, cross-sectional or cohort design was used, (iv) the study was empirical, and (v) the complete study was published in English as a full-length article in a peer-reviewed journal. More than one publication by a given author or authors was included in the review insofar as either the group(s) under study, the endpoints, or the design differed. If two or more studies by the same author(s) offered complementary information but had the same design, endpoint, and study group, they were combined and analyzed together.
Whenever self-report tools other than the JCQ or PSJSQ (7, 20) or their earlier versions were used to assess the demand and control dimensions, two of the authors independently reviewed the described methods to determine whether they were sufficiently compatible with the job-strain model. This procedure was performed in a blinded fashion. Formulations focusing primarily on the individual's subjective reaction to the work environment (eg, "how stressed are you by ...") rather than on its objective characteristics were excluded. As a minimal guideline, it was required that at least one item from each of the major job-strain dimensions be included and that these queries be phrased identically to the original questionnaires or so closely as to be a measure of the original concept (eg, "hectic work" as a measure of psychological job demands). Insofar as the two reviewers disagreed, a third author served as an arbiter.

Included studies
A total of 35 articles was identified that met all the inclusion criteria and needed no further evaluation. Another five articles were independently reviewed by two of the authors, after which those by Hammar et al (21), Sihm et al (22),and Suadicani et al (23) were included, while those by Lynch et al (24) and Murphy (25) were excluded since they did not conform closely enough to the demands or the control dimensions of the job-strain model. Another three, by Billing et al (26), Karasek et al (27), and Messner & Sihm (28) 6 were omitted since only main effects were assessed, and not job strain in relation to the dependent variable. In four instances two or more papers by the same group of authors were combined. The full set of included articles is included in the bibliography.
Assessment of methodological quality 7

Internal validity assessment
We reviewed all studies fulfilling the aforementioned criteria with respect to their methodological strengths and weaknesses, focusing upon the aspects that would seriously compromise the internal validity of the reported results. The 15 validity criteria grouped into four categories were derived from the work of Stock (29). They are described in the appendix. The first category, assembly of the sample, includes the avoidance of selection bias, the avoidance of nonresponse bias, and the application of appropriate exclusion criteria. The validity of exposure variable assessment was evaluated by five criteria related to the assessment of point exposure to high psychological demands and to low control, the avoidance of recall bias, the analysis of job strain, the adequate range of variation, and the assessment of temporal aspects of exposure. The four criteria under the category for confounding and effect modification were adjustment for relevant demographic confounders, adjustment of relevant biomedical and behavioral confounders, appropriate consideration of gender as an effect modifier, and assessment of other dimensions of the work environment. The validity criteria for the outcome variable were related to the assessment of the endpoint itself, whether the assessment of outcome was blinded with respect to exposure status, and the adequate range of variation of the outcome variable.
In most cases, the maximum score was 3 points (optimal). For six of the criteria, there was a possibility for 6 Each of these studies (2 case-control, 1 cross-sectional) had some positive results, such that their omission does not represent a bias towards positive findings. However, because of a substantial number of methodological weaknesses, including confounding (26), low response rates (28), and problems with the assessment of the temporal aspects of exposure (26, 28) and with countermeasures against recall bias (26-28), among others, the positive results do not contribute major supporting evidence for the job strain hypothesis. 4 points, insofar as the authors had used innovative methods that served to advance the state of the art in this field of investigation. Thus the maximum total score possible was 51. The minimum score for each criterion was usually 1. There was the possibility of a score of 0 for situations that would seriously undermine the valid-ity of a study's conclusions. Two of us independently assessed each of the studies with respect to the 15 validity criteria. In cases of disagreement, a third served as arbiter.
The methodological ratings for each study are presented in tables 2-5 according to the 15 internal validity (continued) (37) 1 Unclear how follow-up was carried 2 Unclear whether self-report of out-3 Presumably all participants out, states "all patients were con-come, not explicitly blinded assess-followed-up regardless of tacted" at follow-up intervals to ment outcome "document out-come" [

Directionality
We next asked the question of how these various methodological issues would affect the results and in which direction they would most likely be acting. We examined each of the validity criteria from this perspective, delineating situations that would increase the likelihood of obtaining null results and those that could lead to an overestimation of association. These directionality issues roughly followed the order of the internal validity criteria, although there was no precise one-to-one correspondence between them, since some issues may affect the results in either direction, depending on the specific circumstances. There were also instances in which the way a methodological issue might affect the results could not be determined (eg, a low response rate without any description of the nonrespondents).
The issues that were considered to increase the likelihood of obtaining null results included (i) selection bias in the assembly of the sample, if the participants exposed to job strain but without CVD preferentially entered the study; (ii) selective attrition, if those exposed to job strain or related work stressors selectively stopped working during the follow-up period; (iii) survivor bias (healthy worker effect); (iv) nonexclusion of outcome at baseline leading to dilution of the results; (v) use of the imputation method (imprecise) to define job strain, leading to nondifferential misclassification; (vi) one to two items for assessing psychological demands if the imputation method was used, leading to nondifferential misclassification; (vii) use of a dichotomous variable to define job strain, leading to nondifferential misclassification; (viii) a low percentage of exposure to job strain, leading to a loss of power to detect an existing effect; (ix) single occupation or a limited range of variation of exposure; (x) assessment of exposure to job strain temporally distant from the outcome (studies with long follow-up periods without repeated assessment of exposure status); (xi) lack of a gender-stratified analysis; (xii) likely confounding by another factor, if the relationships were in the opposite direction of the tested association or if several important confounders were not taken into account.
The issues that were considered to increase the likelihood of an overestimation of association were (i) selection bias in the assembly of the sample, if the participants exposed to job strain and with CVD preferentially entered the study; (ii) selective attrition, if those not exposed to job strain or related work stressors selectively stopped working during the follow-up period; (iii) information bias if the outcome was known to the participant at the time of the self-report of exposure; (iv) likely confounding by another factor, if the relationships were in the direction of association. An alternative hypothesis is likely to be operative, whereby a factor other than job strain is the true effect modifier: (v) infor-mation bias if the outcome and exposure were both selfreported.
Next, a judgment was made about the overall direction in which the methodological issues were likely to affect the results concerning associations with job strain, as follows: (i) unequivocal bias to the null: several clear and strong biases to the null and no biases to overestimate; (ii) likely bias to the null: a few likely biases to the null and no clear bias to overestimate, (iii) minimal biases: nearly all of the potential sources of bias fully taken into account; (iv) bias possible in both directions; (v) likely bias to overestimate: one or more likely biases to overestimate and no clear bias to the null; and (vi) unequivocal bias to overestimate: several clear and strong biases to overestimate and no biases to the null.

Strength and consistency of the empirical findings with respect to job strain and cardiovascular disease, reviewed in light of the methodological issues affecting the results
The salient details with respect to the results of each of the reviewed longitudinal, case-control, and cross-sectional studies are presented in tables 6, 7, and 8, respectively. Table 9 provides a summary of the relationships between the results and the direction in which the methodological issues were likely to affect each study.

Longitudinal studies
The longitudinal studies (21-23, 30-47) had higher mean total validity ratings than the case-control and cross-sectional studies did. The mean scores of the studies among men were almost identical for the positive, nonsignificant positive, and null studies. The null studies had a somewhat lower mean total score for the women than those that were positive. Of the two longitudinal studies with the highest total scores (score 40), one yielded a significant positive effect estimate (21, 36), while null results were obtained in the other one (45).
Notwithstanding the high overall methodological quality of these investigations, in all but two, biases towards the null dominated. In 11 of the 17 studies, the biases were unequivocal. Biases towards the null were generally due to the use of the imputation method and long follow-up times, with no re-assessment of exposure or even employment status. Persons close to or even above usual retirement age were included in the baseline sample of several of the studies with protracted follow-up (21,35,36,(38)(39)(40)(44)(45)(46); this inclusion would have attenuated the effect estimates even further. The imputation method is particularly problematic for the psychological demand Table 6. Results of the reviewed longitudinal studies. a (BMI = body mass index, CAD = coronary artery disease, CHD = coronary heart disease, CVD = cardiovascular disease, HR = hazards ratio, HRT = hormone replacement therapy, IHD = ischemic heart disease, MI = myocardial infarction, NS = nonsignificant, O = observer-rated, OR = odds ratio, RR = relative risk, SBP = systolic blood pressure, SES = socioeconomic status, SMR = standardized mortality ratio, SR = self-rated, UK = United Kingdom, US = United States, 95% CI = 95% confidence interval)      (53) years of age, N=13 ischemic demand NS, decision latitude bias as a case-control cases, N=12 hyper-heart disease NS, skill discretion NS, control study, single occupation, tensive controls, NS (age) limited range of variation of all professional exposure; 1 bias to overdrivers estimate: outcome known to participant at time of self-report of exposure; total validity criteria score 33 a All available risk estimates with confidence intervals are shown. For an odds or risk ratio to be considered significant, the 95% CI had to exclude 1.0. b The relevant confounders that were either matched between groups or were adjusted are italicized and indicated in parentheses. c Issues that could affect the directionality of the results of each study are given. d Readers who would like to see additional details about the methodological issues of a particular study or studies, may find it helpful to examine the results in this table together with the corresponding validity assessments in tables 2-5. Table 8. Results of the reviewed cross-sectional studies. a (BMI = body mass index, CAD = coronary artery disease, CHD = coronary heart disease, CVD = cardiovascular disease, HANES = Health and Nutrition Examination Survey, HES = Health Examination Survey, HDL = high-density lipoprotein, MI = myocardial infarction, OR = odds ratio, PR = prevalence ratio, RR = risk ratio, SES = socioeconomic status, SOR = standardized odds ratio, 95% CI = 95% confidence interval)  (65) population-based, heart disease: (95% CI 1.23-4.92), possible bias as a cross-sectional England, age 20-angina 1.1%, MI OR 1.46 (95% CI 1.01-2.12), study, low percentage 64 years possible MI physician diagnosed heart (15%) job strain; 2 biases 6%, physi-disease OR 1.50 (95% CI 1.02-to overestimate: outcome cian-diagnos-2.20), any heart disease OR 1.60 known to participants at ed heart (95% CI 1.20-2.13) (age, age 2 , time of self-report of disease 5%, SES, diet, smoking, leisure-, exposure, self-report of any heart time cholesterol, BMI, diabetes exposure and outcome; disease 9% mellitus, blood pressure total validity score 31 Studies with positive results for job strain and CVD, but none of which were statistically significant e Netterstrøm et N=512 men, N=537 Self-reported Job strain OR 2.3 (95% CI Men: job strain OR 2.4 (95% CI 1 bias to null: survivor bias al, 1998 (64) women, Danish, angina pecto-1.2-4.4) (age, gender, work 0.5-11.5) (age, social status) as a cross-sectional study; 30-59 years of age, ris (N=25 hours, psychosocial factors, 2 attenuated biases to overpopulation-based men, N=10 social status, smoking, systolic estimate: outcome known women) blood pressure, HDL:cholesterol to participants at time of ratio) self-report of exposure, self-report of exposure & outcome (however, no association found between job strain and other somatic pains or between job satisfaction and angina pectoris); low response rate (unclear how this affects results); total validity criteria score 32 (continued)