Combined effects of job strain and social isolation on cardiovascular disease morbidity and mortality in a random sample of the Swedish male working population.

T. Combined effects of job strain and social isolation on cardi ovascular disease morbidity and mortality in a random sample of the Swedish male working population. Scand J Work Environ Health 1989;15:271-279. This study examined the impact of psychosocial work organization on cardiovascular disease (CVD) morbidity and the nine-year cumulative mortality incidence for a random sample of the male Swedish working population (N = 7219). A multiplicative measure was constructed to model the combined effects of psychological job demands, work-related social support and work control. Highly strained and isolated workers had a higher age-adjusted prevalence ratio for CVD morbidity and a higher age-adjusted risk ratio for CVD mortality when compared with those work ing under less strained and more collective conditions. Blue-collar workers showed the greatest risk for both morbidity and mortality when groups with highly isolated and strained conditions were compared to those in more collective and less strained conditions. Strained and isolated workers also had a substan tially higher probability of developing and dying of CVD at a younger age than did those in less adverse environments.

Cardiovascular and circulatory disorders cont inue to be the major causes of death in the Western world . Although much is unde rstood about cardiovascular disease (CVD) with respect to the etiologic significance of biomedical and behavioral risk factors (such as diet, smoking, exercise, and hypertension), a significan t proportion of CVD occur rence remains unexplain ed. In exploring the etiolo gic factors which might contribute to CVD, researchers have examined both environmental conditions and individual predi sposing factors which may contribute to the development of cardiovascular disorders. In recent years an increasing body of literature has pro vided evidence that certain properties of work (such as occupational status, repetitiveness, pacing, and organizational structure) are associated with changes in cardio vascular health . For example, Buring et al (I) found, after contro lling for biological risk factors, that blue-collar workers are at increased risk of CVD when compared with whitecollar workers.
In att empting to und erstand why an individual 's occup ation or the organizational characteristics of work to which he or she is exposed might contribute to the development of CVD, researchers have inves-tigated the con tent of the daily experience of worklife. It has been found , in both laboratory and field studies, that certain conditions produce measurable physiological chan ges. Hormonal elevations (notably the catecholamines) have been found to be associated with specific work properties (such as time pressure, loud noise, distractions and repetitiveness), and lack of control over task cont ent and execution (2)(3)(4)(5)(6)(7)(8) .
For example , recent studies of prison personnel in Sweden have shown that employees in prisons with a low level of decision latitude have an elevated urinary excretion of noradrenaline (9) and also tend to have an elevated prevalence of ventricular ectop ic heart beats during workhours (0).
There is additional evidence that job strain can induce excessive and long-lastin g sympathoadrenal arou sal, which leads to sustained blood pressure elevation (4,8). Adrenaline causes the heart muscle to beat faster to increase blood flow, and other hormones cause periphe ral vasoconstr iction to minimize blood flow. Due to these hormonal elevations and vasoconstriction, blood pressure rises (II ). In a five-year study of 380 air traffic controllers repeated measures during various job ta sks demonstrated that high work load produced significant increases in blood pressure (12). These physiological reactions -if they are prolonged over an extended period of time -may result in further damage to the cardiovascular system. Theorell et al (13) found that tho se men who show a propensity to blood pressure elevation earlier in life and who work in " strain occupations" have much more marked blood pressure elevat ions 'at work than other men in the same 28-year-old age group . Anothe r recently completed longitudinal study performed by Theorell and his colleagues evaluated the health of working men and women in six service occupations (14). Each person was examin ed four times. A significant positive average association between increa sing job strain and increasing blood pressure at work was shown. These observation s point to the possibility of a direct relation ship between the psychosocial work enviro nme nt and risk of heart disease (15).
We have cause to hypothesize, assuming other relevant factors to be equ al, that those working for extended period s in jobs that are stressful could be expected to age more rapidly and to die earlier when they are compared to similar wor kers with less expo sure to adversive psychosocial conditions. Th is connection between chro nic work stress and th e long-term health con sequences of such expo sure has been the focus of a series of investigations in the United States (16)(17)(18)(19) and Sweden (20)(21)(22)(23)(24)(25)(26) . Karasek and his colleagues have predicted that biologi cally adversive strain wiII occur under a specific condition , ie, when the psychological demands of the job exceed the resources for cont rol over task content (20, 21. 23). A number of studies support this hypoth esis and have demonstrated th at workers in jobs characterized by high demands and low contro l ar e at increased risk of developing CVD (16)(17)(18)(19)(20)(21)(22)(23)(24)(25)(26).
Johnson (27)(28)(29) and John son & Hall (30,3 I) have developed and tested an expa nded version of this formulation. In investigatin g wheth er social support from co-workers is another str uctural factor which moderates the impact of psychological job demands, it was found that social isolati on at work fun ction s as an ind ependent risk factor. When isolation is coupled with lack of control and high demands at work, there is a substantial increa se in observed CVD prevalence risk. Thus in epidemiologic investigations there is increasing evidence that these three factors -demands, control and support -may function as a psychosocial risk complex with respect to health outcomes such as CVD.
In an effort to examine the effect of combined exposure to low control, low support , and high demands, a new variable, iso-strain (ie, combination of social isolation and job strain), was constru cted in the present investigation. This composite repre sent s the joint effects of the three job character istics. It was constructed with the pr evious findin g of the investigators, ie, that the thr ee factors " demands, " "contro l," and "support " interact in a mult iplicative mann er in relat ion to CVD prevalence risk, as the starting point (29). One of the purposes of the present study was to test this new varia ble on a large repre sen tat ive sample in an att empt to evaluate its usefulness in stress research. Because iso-strain is a univariate measure, it is a more parsimonious instrument for measur ing and anal yzing workplace conditions. It is easier to use and interpret in statistical anal yses th an the thr ee-fa cto r demand -control-support mod el with its 27 po ssible combinations (ie, high , medium, and low conditions for each of the factors). 272 Th is study addressed th e following resear ch questions: I . Do wor kers exposed to psychosocially adverse job conditions (high iso-strain) have a greater risk for CVD morbidity and mortality than those not so exposed (low iso-strain)?
2. Do workers exposed to psychosocially adverse job conditions (high iso-strain) develop CVD at a younger age and die of CVD earlier than tho se exposed to more favorable work conditions (low iso-stra in)?
3. Is there evidence of a dose-respon se relation ship between the intensit y of iso-strain exposure, CVD risk, and age-related trends in the development of CVD morbidity and mortality?

Materials and methods
Our study used the Swedish Central Bureau of Statistics' survey of living conditions. Data were collected in a I-h personal int erview performed by pr ofessional interviewers in 1976 and 1977. The Bureau obta ined a systema tic random sample from the Nat ional Register o f Births in 1976 and 1977. Th e respon se rat es were 80 010 in the two years used in our study. We used a subsample of 7219 emplo yed men . As the measures used in this study have been extensively described elsewhere (27)(28)(29)(30) , the y wiII be discussed onl y briefly.
The categor ies of blue-and white-collar workers were defined according to the Bureau 's system. Whitecollar workers were offi ce and pro fessional employees, and blue-collar workers were production , distribu tion , and service workers .
The scales for the work characteristics were based on a principal component factor analysis of the Swedish questionn aire items relating to work , which were identically word ed and scored in both survey years. They app ear in the appendix. The psychological job-demand indicator was constructed from two items, which asked if work was hectic and/or psychologically demanding. The wor k control scale measures decision-making auth ority, task variety, and personal freedom on the job . The social support scale consisted of five dichotomou s items which measured the availability of social inter action in the workplace and whether these interactio ns carryover into non workl ife.

Construction of the iso-strain indicator
One of our earlier studies examined the sta tistical interactions o f demands, control, and social suppo rt in relatio n to the prevalence of CVD. Using Rothman' s interaction ratio technique (32)(33), we found that the pr esence of combined exposure to all of the theoretically adverse conditions resulted in a CVD prevalence ratio that was 9 0J0 greater than that predicted in a mult iplicat ive model. A univariat e measure was constructed on the basis of the approximat ely mult iplica-tive character of this interaction. We re fer to this as "iso-strain" since it represent s the combination of socia l isolation and job strain . The variable was constructed by rescoring the scales so that a high score represented an adverse condition. The three subcomponent scales were standar dized to a mean of zero and a standard deviation of one. A con stant term was then added to bring all scores to above one, and the three scales were multiplied together. On the basi s of the frequency distribution of the resulting scale scores, the high iso-strain gro up was defined as the highest quintile, low iso-strain as the lowest quintile, and medium iso-strain as the middle three quintiles of the continuous indicator.

Cardio vascular disease indicators
In this study we examined two outcomes , the prevalence of CYD morbidity and cardiovascular-specific mortality over a nine-year follow-up period. The measur e of CYD pre valence was based on the Swedish Stati stical Bureau 's health classification system, which obtained information on all types of long-term illness and disability during the personal interview with the followin g que stion concerning health status: "Do you suffer an y long standing illness, effects of an injury, any disability or weakness?" Given an affirmative respon se, the interviewer probes: "Could you explain that a little mor e?" "What did the doctor say it was?" " What part of the bod y or organ system is affe cted?" The subjects were also questioned concerning their usc of regular medication so that symptom-free conditions not covered by the previous question s could be included . The description of illness was coded by a central unit of the Bureau according to a coding system developed and tested by two consultant physicians. They matched reported symptoms and illness descriptions with the International Classification of Diseases, eighth revision (lCD) . A general category for diseases of the cardiovascular system, develop ed by the Bureau, was used to classify 407 men as ha ving CYD, a prevalen ce rate of 5.64 0,70 . Of these cases, 93 % were coded as havin g art erio sclerotic heart disease , cerebrovascula r disease, or peripheral vascular disease (lCD code s 400--404; 410--414; 427, 430--436; and 440--445 respectively). Of the remaining 7 %, 1.4 % were coded as having a venous or pulmonary embolism or thrombosis, 2 010 were coded as ha ving an ill-defined heart disease (ICD code 429), and 3.6 % were coded as having other or unspecified forms of circulatory system disea se (lCD codes 454--458). Although a small proportion of these lCD categories do not have a plausible stress-related etiology, a decision was made to use the broader CYD indicator because of the findings from prospecti ve validation studies of th e Bureau's chronic disease classification system (34). These studies had indic ated that the survey-based dete ction system was very accurate in ascertaining broad categories of disease, such as " cardiovascular disease. " As we ha ve reported elsewhere (18) the CYD indicator used in this study was found to have good predictive and discriminant va lidity, for it was strongly associated with follow-up CYD mortality and not significantly associated with non-CYD mortality.
CYD mortality was assessed through the use of th e offici al Swedish personal identification number. This number, assigned to all individuals at birth, provided the basis for linking the survey responses with mortality data from the National Death Register for the years 1976--1986. Those inte rviewed in 1976 were followed until 1985, and those interviewed in 1977 were followed until 1986, for a total of nine years of followup for the entire study group. We analyzed the incidence of CYD-related mortality by combining all deaths for arteriosclerotic heart disease, cerebrovascular disease, and peripheral vascular disease (lCD code s 400--404; 410--414; 427, 430--436; 440--445 respectively). ln the period 1977--1986 there were 193 deaths related to CYD. The nine-year cumulative incidence rate was 2.67 % .

Statistical analysis
The data were analyzed with the Statistical Analysis System (35) and a series of epidemiologic programs developed by Rothman & Boice (36). The measures of association were the relative prevalence and relative risk ratios. The Mantel-Haenszel chi was used for hypothesis testin g (37), and the confidence limit s were constructed from the point estimates of the prevalence ratio (PR) and risk ratio (RR) and the Mantel-Haenszel test stati stic with the use of test-based interval estimation (38--40). Age adjustment was per formed according to Mantel-HaenszeI's and Miettinen's techniques (37--39).
In our examination of the change in CYD risk across the three iso-strain levels, the Mantel extension test for overall trend was calculated, stratified on age (41). The standardized risk ratio (SRR) values were computed with Miettinen's procedure , the low iso-strain group serving as the standard or reference category (42). A heterogeneity chi-square was used to examine the uniformity of effect across the age strata (36).
The logistic cur ve analysis for age trends was calculated according to the methods proposed by Mac-Mahon & Worcester (43). The adequacy of the logistic model was indicated in the goodness of fit chi-square, which in all cases was sta tistically non significant and therefore indicated that there were no significant differences between the observed rates of CYD and those probabilities predicted by the logistic model. The logistic cur ves have been described by two parameters, ie, the 50 % midpoint , in this analysis the year of age at which 50 % of the specific iso-strain exposure group would be expected to develop CYD, and the logistic slope. By looking at the trends in CYD prevalence across variou s age strata, we made the assumption that age was a rough approximation for expo sure time. The logistic curves were calculated sepa rately for each isostrain group. Since the probability of developing and dying of CVD inevitably increases with age, our interest focused on the possibility that the manifestation of CVD would increase at a differential rate in the separate iso-strain groups.

Results
For the total study group, the workers with a high isostrain level were found to have an age-adjusted PR of 1.77 [95 OJo confidence interval (95 % CI) 1. 28-2.44] for CVD morbidity, when compared with the low isostrain group. Moreover, the Mantel extension test for overall trend, which included the middle category of exposure, was statistically significant (P = 0.(01). There was a stepwise increase in risk for each increase in isostrain, ie, there was an SRR of 1.00 for the low, 1.38 for the medium, and 1.75 for the high exposure groups.
A similar and even more pronounced pattern was found among the blue-collar male sample, the high isostrain group having an age-adjusted PR of 2.04 (95 % CI 1.24-3.36) when compared with the low iso-strain group. The Mantel extension test for trend was statistically significant (P = 0.002), the SRR values increasing from 1.00 for the low to 1.47 for the medium and 2.16 for the high iso-strain categories of exposure.
A different pattern was found for the white-collar sample. Those with high iso-strain had a statistically nonsignificant age-adjusted PR of 1.49 (95 OJo CI 0.91-2.43) when compared with those with low isostrain. The age-adjusted Mantel extension test was not statistically significant (P = 0.07), although the SRR values for both the medium (SRR 1.41) and high (SRR 1.34) groups were elevated above 1.00. However, since there was an indication of heterogeneity of effect across age strata, separate analyses were performed for the workers aged 44 years and younger and for those above 44 years of age. When older workers were considered separately, there was an indication of a significant trend (P = 0.03), and some stepwise increase in CVD prevalence risk, the medium iso-strain group having an SRR of 1.64 and the high iso-strain group an SRR of 1.74. It should be noted that the differential between the medium and high iso-strain levels for white-collar workers was not as substantial as that found for the total or blue-collar samples. The trend in the sample of younger white-collar workers, although not significant (P = 0.20), was actually reversed, the medium (SRR 0.88) and high iso-strain groups (SRR 0.38) having SRR values below that of the low iso-strain group.
In the analysis of the CVD mortality of the total sample, the high iso-strain group was found to have an age-adjusted RR of 1.92 (95 % CI 1.15-3.21) when compared with the low iso-strain category. The Man-teI extension test was statistically significant (P = 0.02) and therefore indicated some degree of trend. However, the SRR values for the medium (SRR 1.71) and 274 high (SRR 1.90) iso-strain categories were not markedly different.
Of the blue-collar subsample, the high iso-strain group was found to have an age-adjusted RR of 2.58 (95 % CI 1.06-6.28) when compared with the low isostrain group . The Mantel extension test was only of borderline statistical significance (P = 0.08) . Although the SRR for the medium iso-strain group (SRR 2.21) was elevated relative to that of the low iso-strain group, it was of a similar magnitude as the SRR of the high iso-strain group (SRR 2.54).
For the white-collar sample, the high iso-strain group had an age-adjusted RR of 1.31 (95 Ofo CI 0.58-2.96) for CVD mortality when compared with the low isostrain group. The Mantel extension test was not statistically significant (P = 0.23), and the trend of effect, an SRR of 1.39 for the medium and of 1.26 for the high iso-strain group, provided no indication of a doseresponse relationship.
The results of the logistic analysis of age trends for the prevalence of CVD morbidity in the total population is displayed in figure 1. The low iso-strain group had a markedly slower age progression of cardiovascular disease morbidity than did the other iso-strain groups . Although there was a greater magnitude of CVD morbidity at each age within the high iso-strain category, the age trend was not markedly different from that seen in the medium iso-strain group.
As in the PR, RR and SRR analyses , the analysis of age trends for CVD morbidity showed a more pronounced effect for the blue-collar workers than for either the total sample or the white-collar sample, as indicated by a comparison of figure 2 with figures I and 3. There was a much stronger suggestion of a doseresponse relationship in this subgroup, as there was a stepwise increase in the age-related progression of CVD morbidity with increasing levels of iso-strain.
For the white-collar workers, as is shown in figure 3, the most marked finding was the difference in the age trends for CVD morbidity between the high and medium iso-strain group on the one hand and the low isostrain group on the other. However, there was no noticeable difference in trend between the high and medium iso-strain groups.
The logistic analysis of age trends in CVD mortality incidence for the total male population is displayed in figure 4. Although there was little difference in the age-related progression of CVD mortality between the high and medium iso-strain groups, the trends for these two groups were considerably more pronounced than that found for the low iso-strain category, which had a much slower age progression of CVD death.
A similar pattern was found among the blue-collar workers, as is shown in figure 5. Again, the most significant finding was the much slower age progression of CVD mortality in the low iso-strain group, when compared with the other two groups, which demonstrated rather similar trends .
Finally. figure 6 sh o ws the r esults o f the age trends in C VD mortal i t y f or th e w h i t e-collar w orke rs. The o ver all difference in t r ends bet w een the hi gh and medium grou ps on the one h and an d t h e low iso-strain gro u p on the other was less pronounced than that found f o r the blue-collar grou p or for the total p op u -     among white-collar workers were very large in the mortality analysis. The refore these curves can only be considered app roxima tions of possible underl ying tren ds.

Discussion
The main finding in the present stud y was that the combined mea sure of iso-strain identi fied a group with remarkably slow cardiovasc ular aging and late CVD death risk -the low iso-stra in gro up. Th ese findings were consistent from morbidity to mortalit y and from blue-collar worker s to white-collar worker s. They were also consistent with the pro spective findings in a previous study of Swedish men (23), am ong whom the absence of self-repo rted job strain was associated with a decreased risk of heart disease developm ent.
The present stud y provides support for the hypothesis that workers exposed to adverse psychosocial work organization (represented in this study by the iso-stra in variable) have an elevated risk for CVD morb idity and mort ality. In term s of a dose-respon se relationship between iso-str ain and CVD risk, the result s were more uncertain. In the CVD morbidity and mor talit yanalysis the Mantel extension test was statistically significant, and the SRR values increased with increa sing levels of iso-strain in the ana lysis of the to tal sample. Thi s same pat tern was repeated in a more pronounced form for the blue-collar sample (thou gh the Mantel exten sion test was only of borderline sta tistical significance in the mortality analysis). For the white-collar workers, however, there was only a mode rate degree 276 .20 of support for a dose-respon se interpretation and then only for CVD morbidity in the older group.
In the logistic analysis o f the age-related progression of CVD, the clearest indicat ion of a dose-respon se relation ship was found for CVD morbid ity in the bluecollar subsample. There was a similar , if less pronounced , dose-response effect fou nd for CVD morbidity in the total sample. For the white-collar workers, and for CVD mortality in all the groups in the study, there was relatively little , if any, indicat ion o f a stepwise increase in the age-relat ed pr ogression of CVD with increa sing levels of iso-strain.
It is important to distin guish bet ween the overall CVD risk magnitude repre sented by the age-adjusted PR, RR, and Mantel extension test, and the age-related progression of CVD. As is evident in figure 5, for example, workers with a high level of iso-strain evidenced a not abl y greater probability of dying of CVD from the age of 35 to 60 years than did the medium iso-strain group. This greater risk was reflected in the difference in the SRR values, 2.54 versus 2.21. How ever, the trend, as represented by the logistic slopes of the two curves for these two groups, was similar. A further distinction can be made between the rate of progression of CVD (the slope o f the curve) and the degree to which iso-strain accelerates the on set of the disea se. As can be observed in figure s 3 to 6, although there was an identical or very similar rate of progression for the high and medium iso-st rain exposure categories, pro gression began at an earlier age in the high iso-strai n group.
Although both the overall risk of disea se and the patt ern of its developm ent should be taken into ac-count in considerations of the effects of environmental exposure, we would argue that , in the present investigation, the pattern of age progre ssion is perhaps more important. Though there is some indication of a doserelated acceleration of disease onset, the most marked differences were observed when the workers with either high or medium iso-strain were compared with thos e with low iso-strain, Rather than a clear do serespon se effect that can be attributed to levels of isostrain exposure, this stud y has identified a remarkably healthy group -those with low demands, high contro l, and high support who evidence a very slow progression of CVD morbidity and mortality.
The lack of consistent observed differences between the medium and high iso-strain categories may be due to an error in classification. The iso-strain measure may onl y be able to distingui sh tho se with some exposure from those with little or no exposure. The measure may need to be refined further in order to delineate between the high and medium exposure categories.
The present study has a number of methodological strengths and weaknesses. The studied sample is fully representati ve of the Swedish male work for ce. The participation in the interviews was good. Loss to follow-up in the death register was negligible. Thu s, the findings should be generalizable for Swedish working men.
CVD morbidity was studied by means of an indicator based on interviews, not on actual medical examination . Althou gh the jud gment regarding caselno case was made following the medical instructions of an expert panel, an element of self-report bias may still exist. Self-report bias can increase the risk of spurious associations since the job characteristics were also based upon self-reports. The CVD death diagnosis, on the other hand , has a relati vely high degree of accuracy due to the high autopsy rate in Sweden.
When the patterns observed for the two end points are compared, there is some overlap -approximately one-third of the men who died of cardiovascular disease had been classified as "cases" in the prevalence study. The difference between the low iso-strain group and the other groups was larger for CVD mortality than for morbidity. This finding speaks in favor of a true difference. Had the CVD mortality findings been due to the workers in the case group only, they would have been weaker than the morbidity findin gs.
Hou se et al (44) has pointed out another methodological concern in regard to the still common pra ctice in the epidemiolog y of occupational stress to measur e job characteristics at only one point in time. The subjects' reports on this single occasion may reflect a temporary phenomenon which has no significance for cardiovasc ular pathology. Thu s it is possible that the reported associations may be spuriously weak. In a future analysis of the same samples using an occupationall y based psychosocial exposure matri x, we will be able to utilize indi vidual work histori es in order to 3 relate mortality to the duration of exposure to variou s categories of iso-strain, The following explanations of the observed decreased risk of CVD amon g low iso-strain workers are possible: I . The association may be spurious and simply due to the self-selection of already sick or vulnerable persons int o jobs with moderate or high degrees of isostrain. This explanation seems unlikely . Sick or vulnerable person s would rather avoid than select jobs that are demanding, isolated, and uncontrollable, particularl y in a country with an advanced social security system. 2. Self-report bias may have influenced the morbidity associations, but it is unlikely to explain the mortalit y associat ions. (See the preceding discussion.) 3. Accepted biomedical risk factors may have been lower in the low iso-strain group. Because of the limited information available on our sample, we were not able to examine explicitly the potential confounding effects. However, other studies (15,30,45,46) have shown that biomedi cal risk factors do not "explain away" the associat ion between job strain, social isolation, and CVD. Biomedical risk factor s, in particular smoking habit s, blood pressure, and possibly serum cholesterol, may in fact be influenced by job conditions (45,46). A recent report (47) has also shown that extremely adverse job conditions may influence lipoproteins, although in general lipoproteins do not seem to be influenced to any great degree by psychosocial factors. Therefore, one important distinction should be mad e between variance in biomedical risk factors caused by expo sure to psychosocial job conditions and biomedical risk factor variance not influenced by these job factors. Finally, psychosocial job conditions could influ ence illness risk dir ectly outside the accepted biomedi cal risk factors. 4. The three componen ts of iso-strain all correlate with social class. Thus, in this study, psychological job demands were significantly higher for the white-colla r workers than for the blue-colla r workers, while both control and support were significantly lower among the blue-collar workers. Thu s the comb ined iso-strain measure was influenced by social class and was higher among the blue-collar workers since two of the components were predom inatel y "blue-collar problems. " There were only a few white-collar men in the high isostrain category as compared with the number of bluecollar men in th is category . However, the association between reduced CVD risk and low iso-strain was not explain ed by the class gradient since it appeared both in the white-collar group and in the blue-collar group. Although on different levels, the white-collar workers, on the whole, had a lower risk than the blue-collar workers. These observation s do , however, point to the need for a more class-specific anal ysis that considers the physical properties of work a s additional potential ri sk factors . In a future analysi s, a class -specific strategy will be introduced to our analysis of CVD risk.