Association between ambulatory blood pressure and alternative formulations of job strain.

Association between ambulatory blood pressure and alternative formulations of job strain. Scand J Work Environ Health 1994;20:349-63. OBJECfIVES - The goal of the study was to determine whether alternative formulations of Karasek & Theorell' s job-strain construct are associated with ambulatory blood pressure and the risk of hyper tension. M~;THODS - Full-time male employees (N = 262) in eight worksites completed a casual blood pres sure screening, medical examinations, and questionnaires and wore an ambulatory blood pressure mon itor for 24 h on a workday. Cases of hypertension were ascertained from casual blood pressure read ings for a case-referent analysis. A cross-sectional analysis was also conducted, ambulatory (continu ous) blood pressure measurements being used as the outcome. RESULTS - All formulations of job strain exhibited significant associations with systolic blood pres sure at work and home, but not with diastolic blood pressure. Employees experiencing job strain had a systolic blood pressure that was 6.7 mm Hg (~ 0.89 kPa) higher and a diastolic blood pressure that was 2.7 mm Hg (~ 0.36 kPa) higher at work than other employees, and the odds of hypertension were increased [odds ratio (OR) 2.9, 95% confidence interval (95% CI) 1.3-6.6]. Using national means for decision latitude and demands to define job strain increased the systolic and diastolic blood pres sure associations to 11.5 mm Hg (~1.53 kPa) and 4.1 mm Hg (~ 0.54 kPa), respectively. Adding or ganizational influence to the task-level decision latitude variable produced a stronger association for hypertension with job strain (OR 3.7,95% CI 1.6--8.5). Adding social support to the job-strain mod el also slightly increased the hypertension risk. CONCLUSIONS - The impact of job strain, at least on systolic blood pressure, is consistent and robust across alternative formulations, more restrictive cut points tending to produce stronger effects.

The association between j ob strain and blood pressure has also been examined to ascertain a possible biological mechanism linking job strain and CVO. Nin e studies (11)(12)(13)(14)(15)(16)(17)(18)(19) hav e examined job strain and casual blood pre ssure me asured in a clinic setting (in which typicall y on ly a few readings are taken), but only one fo und a sig nificant association (17) . However, an ambulatory blood pressure monitor has been used in nine job-strain studies (15,18,(20)(21)(22)(23)(24)(25)(26) sin ce it is both a more reliable (no obse rver bias and increased number of readings) and a more valid measure of average blood pre ssure (blood pre ssure measured during a person's normal daily activities) than casual measurements of blood pr essure.
Variou s forms of the independent variable that sugge st a robustness to the job-strain concept have been reported in these studies. However, it might be that this is an artifact of the selection process of reporting the forms that have positive associations with outcome and ignorin g negative ones. Therefore , the analy sis reported in this paper was designed to examine various operat ionalization s of the job-strain variabl e and their associati on with ambulatory blood pressure (measured continuously) and hyperten sive status. The result s are presented from an ongoin g prospecti ve study being conducted in New York City throu gh the Hypertension Cente r of New York Hospital-Cornell University Medical College (22,23,27,28).

Formulations of j ob stra in
Combination ofdemands and decision latitude. Since 1979, the interaction between job demands and job decision latitude which defines job strain has been operat ionalized primarily in four ways (2). The most comm on procedure has been to define emplo yees both above the median on demands and below the medi an on latitude as a "high strain" group (henceforth called the job-strain "quadrant" definition). Such employees can be identified after self-reports of demands and latitude at either the sample medians (figure 1a) or national median s, or the means are dichotomized (figure lb).
Second, a newer approach [used in six stud ies with positi ve result s (2)], has been to create a continuou s independent variable -demand s divided by latitude (figure l c). Since this quotient term is nonlinear and tend s to give more weight to latitude (the denominator ) than demand s, a simil ar continuous (but linear) formulation has been propo sed in which the contributions of low latitude and high demands are equally weighted (figure Id). 3 Third, in five studies (8,12,13,17,29), a multiplicative interaction term controlling for main effects (demands x latitude, controlling for demands and latitude) was used. However, in only two of these studies ( 17, 29) did the interaction term significantly improve the variance explained by the model beyond what could be obtained with only the main effects.
Fourth , exploratory graphic approa ches have been used (29,31) to exam ine interaction effects and thresholds (nonlinear effects). For example, demands 3 In one study (30), a similar formulation was used to isolate 20% of the sample as "high strain." The "high strain" group was defined as (demands +I) x (latitude -I) < -1.2, where demands and latitude are standardized scores. Tests of this formulation using the current data set produced results similar to those of other formulations in figure I , and therefore, for the sake of brevity, they are not presentedhere. 350 and latitude have been divided into tertiles or quartiles, and the patterns of heart disease risk then examined across the resulting nine (29) or sixtee n (3 1) exposure cells.
Thresholds. If a population threshold exists for the effect of job strain, the proportion of subjects exceeding this thresh old will typically vary across nonpopulation-ba sed samples . For example , if only 10% of a given sample is actually experiencing "high" (biologically relevant) job strain, but 25% of the sample is classified as "high strain" accordin g to the quadrant defin ition, misclassification and a dilution of the effect estimate occurs . One appro ach to explore possible threshold s, or nonlinear or nonmonotonic dose-response curves , is to dichotomize a continuous exposure variable at various points and plot the resulting odds ratios (32). This approach has been applied in the following presentation .
Main effects. Several studies have also examin ed the main effects of job demand s and j ob decision latitude and found some significant association s (2). As in studies that use multiplicative interaction formulations, studie s that use other formulations to model an interaction implicitly (eg, median cut point s or quotient terms) need to examine main effects carefully to determine whether the effect attributed to j ob strain is due solely to one or both main effects, rather than synergy.
Compon ents of decision latitude. Job decision latitude has traditionall y been defined as the combination of skill use and task-level decision-m aking authority. The two components are often highly correlated (33,34), although not always (3 1). Therefore, one of these components may be a better predictor of CVD risk. In addition, job decision latitude has traditionally been operationalized as a task-level variable. However, it has been suggested that control not only over task characteristics but over work-group, departmental, or organizational functionin g (35,36) may also be important in modifying (buffering) the effect of job demands.
Addition of social support. Workplace social support has also been examined in conjunction with the jobstrain model in several studies of CVD (7,9,10,29), as well as in studies of psychological strain outcomes (37,38). Evidence has been provided for both a main effect of social support ( 10,29) and a buffe ring of the effects of j ob strain by social support (9,10,29). A third approac h (7), which does not try to disentangle main effects and interaction, simpl y hypothesizes that socially isolated (ie, without supportive co-workers or supervisors) "high strain" work ("iso-a) Job   strain " rather than simply "job stra in") carries the highest risk. This approach was proposed since "isostrain" is a univariate measure, "a more parsimonious instrument for measuring and analyzin g workplace conditions [p 272]" (7) .

Socio economic status
The four CVD studies that compared social class groups found stronger associations between job strain and outcome for blue-collar workers (7,29), workers with low education (31), or female clerical workers (3) than for higher socioeconomic status groups. Therefore, potential effect modification by socioeconomic status was examined in the present study.

Hypoth eses
With the preceding as background, this paper focuses on the following seven specific hypotheses: (i) job strain, utilizing a variety of formulations , is associated with higher blood pressure and with increased risk of hypertension; (ii) the main effects of demands and decision latitude are less strongly associated with higher blood pressure or with increased risk of hypertension than formulation s of job strain; (iii) the components of decision latitude -skill utilization and decision authority -in combination with demands, are each associated with higher blood pressure and with increased risk of hypertension; (iv) the addition of organizational influen ce to the task-lev-Scant! J Work Environ Health 1994, vol 20, no 5 el decision latitude component of the job-strain model increases effect estimates; (v) social support is associated with lower blood pressure and with reduced risk of hypertension; (vi) social support modifies (buffers) the effect of job strain on blood pressure and risk of hyperten sion [buffering is assessed primarily through multiplicative interaction; an increase in effect estimates when social support is added to the job-strain model (resulting in the "isostrain" model) may result from main effect s or buffering], and (vii) higher socioeconomic status modifies (buffers ) the effect of job strain on blood pressure and risk of hypertens ion.

Subjects
The subject selection and blood pressure measurement have been described in detail elsewhere (22,27,28). Briefly, after an initial casual blood pressure screening at eight New York City worksites, each employing at least 150 men, subjects were eligible to be selected for the initial case-refe rent study (27,28) if they were male, were between 30 and 60 years of age, were employed more than 30 h per week, were able to read English, had a body mass index of less than 32.5 kg . rrr", had no second job of~15 h per week, and had been at their current worksite for at least three years before being approached for this study or before being diagno sed as having high blood pressure, whichever came first. (At the eighth worksite the employment criterion was relaxed to one year.) Subjects were excluded if they had a history of CVD, secondary hypertension , systolic blood pressure greater than 160 mm Hg (",21.28 kPa] or diastolic blood pressure greater than 105 mm Hg ('" 13.97 kPa). Subjects with diagno sed hypertension were eligible only if they could have their medication stopped for at least three weeks before wearing the ambulatory blood pressure monitor and still maintain a diastolic blood pressure below 105 mm Hg ('" 13.97 kPa).
Eligible subjects having both screening and recruitment casual diastolic blood pressure (four to six weeks later) above 85 mm Hg ('" 11.31 kPa) or those taking antihyp erten sive medication and agreeing to participate were classified as hyperten sive and became study cases (N = 85). Eligible subjects having screening and recruitment diastolic blood pressures of~85 mm Hg (~1 1.3 1 kPa) and agreeing to participate were classified as nonhypertensive and became study referents (N = 177). In addition to the initial case-referent study, a subsequent cross-sectional analysis of this sample was conducted with the ambulatory blood pressure readings as a continuou s outcome measure (22). (Subsequent to the cross-sectional analysis, it was discovered that two subjects failed to meet the original eligibility criteria and were dropped from the current analysis.)

Procedures
The subjects wore an ambulatory blood pressure monitor for 24 h during a normal workday, using procedures described previously (39). Diary information (ie, whether the subjects reported being at work, at home, or asleep) was used to calculate the average ambulat ory blood pressures for each location category. The subject s were also given a routine medical examination , which included a history, a physical examination, blood testing, 24-h urine collection, an electr ocardiogram, and an M-mode echocardiogram . In addition, they completed a questionn aire which included the Job Content Questionnaire (JCQ), a 42-item survey based, in part, on questions drawn from the United States (US) Quality of Employment Surveys (QES) (40).
The following job characteristics scales were constructed from the JCQ: Job decision latitude was the sum of two equally weighted subscales: (i) skill utilization, measured by six items (keep learning new things, can develop skills, job requires skill, task variety, repetitiou s (reverse scored), job requires creativity) and (ii) decision authority, measured by three items (have freedom to make decision s, can choose how to perform work, have a lot of say on the job). Organizational influence was defined as the sum of three items (influence over group decisions, my work group's ideas about compan y policy will be considered, chances for career development/promotions are good). Composite latitude was the sum of three equally weighted subscales: skill discretion, decision authority, and organizational influence. Psychologicaljob demands were defined by five items (excessive work, conflicting demands, insufficient time to do work, fast work, hard work). Social support at work was the sum of the followin g two equally weighted subscales: (i) co-worker support, measured by four items (competent co-workers, friendly coworkers, co-workers helpful in getting the job done, co-workers take a personal interest in me) and (ii) supervisor support, measured by four items (supervisor successful in getting people to work together, supervisor helpful in getting the job done, supervisor pays attent ion to what I am saying, supervisor concerned about welfare of those under him or her). Physical exertion was measured by a single item (job requires lots of physical effort). All of these JCQ items were scored on a Likert scale of I to 4. Decision latitude, psychological demands, and composite latitude were constructed to have a range of 12 to 48. The internal consistency was above 0.70 for all of the scales, except for organizational influen ce (Cronbach's alpha =0.62) (table I).
The followin g questionnaire scales and items were also analyzed. Type A behavior was evalu ated by the 52-item Jenkins Activity Survey (4 1). Subjects were classified as "Type A" if their standardized scores were above O. Demographic items included years of education , race, and age. Race was classified as Caucasian versus other (including His- panics). Alcoho l and smoking beha vior were assessed by interview at the time of the medical examination. The subjects were classified as nondrinkers if they reported they drank not at all or occasionally or as drinkers if they report ed daily consumption or binge drinkin g. They were classified as smokers if they currently smoked.
The association between various formulations of job strain and ambulatory blood pressure was tested by multipl e regression, controlli ng for known or suspected risk factors (age, education, race , body mass index, smoking, alcohol, 24-h urine sodium excretion, type A behavior, physical exertion) and a potential confounder (worksite). Quadr ant terms, quotient terms (demands divided by latitude), continuous linear terms (demands minus latitude) , and main effects were assessed from an examination of the regression coefficients. Multipli cative inter action was assessed by removing the product term from the full model with main effects and examining the F-statistic for the change in the squared multiple correlation (R2 l . The previously reported (22,27) job-strain quadrant term was computed by dichotomizing demands and latitude at their respective sample medians. Fifty-six subjects (2 1% of the sample) reported jo b decision latitude of less than 37 and job demands of greater than 32. A similar proportion of subjects was classified as being under "high strain" (22 and 21%, respectively) when skill or authority of less than 19 was combined with demand s of greater than 32.
The influence of the main effects of demands and decision latitude relative to the job-strain quadrant formulation was tested in a comparison of the change in R2 obtained when the quadra nt formulation was added to a model which included both main effects with the change in R2 obtained when both main effects were added to a model which includ ed the quadrant formulation.
An examination of the possible thresholds of the effect of job strain on blood pressure was conducted. First, cut points were chosen which classified a smaller proportion of subjects as being under "high strain" (latitude < 36 and demands >3 2, latitude < 37 and demands > 33, skill or authority < 18 and demands >32), includ ing the use of the national US QES averages for men (demands > 30, latitude < 32, skill <18, and authority < 16). The national average for latitude was far lower than our sample' s average; however, the national average for demands was also somewhat lower than our sample's average. Ther efore, while using national cut points produced a more restricted definition of "high strain" (8% of the sample), some addition al subjects with lower demand s were classified as being under "high strain" with this method (figure lb ).
Second, the quotient term was dichotomi zed at varying cut points and the pattern of effect sizes (unstandardized regression coefficients) was examined.
Third , a nine-cell model of job strain was created with tertiles of latitude and demands being used as the cut points (dema nds <29 , 29-34, >34; latitud e < 34, 34-3 8, > 38) and the model being tested in an analy sis of covari ance. The numbers of subjects per cell for low, medium , and high demands, respecti vely, were 41, 25 and 17 for low latitude, 22, 40 and 28 for medium latitude, and 18, 32 and 39 for high latitude.
Three composite latitude quadrant terms were constructed with composite latitude below 35, 36 and 37.
Combined with demands above 32, they classified 20,22, and 28%, respectively, as being under "high strain." Finally, the following three "isostrain" terms were computed: (i) job strain (latitude < 37 and demands >32) combined with social support below 23.5 (dichotomized at the median); (ii) composite latitude below 36, demands above 32, and social support below 23.5; and (iii) composite latitude below 37, demands above 32, and social support below 23.5. These three terms defined 14, IS, and 18%, respectively, as being under high "isostrain." To examine the effect modification by social support, we dichotomized the social support variable at its median. However, since a large number of subjects (13% of the sample) had the median value, two separate dichotomizations were examined; they defined 40 and 53% as having low social support. To examine effect modification by education, we compared subjects with a high school diploma or less education (34% of the sample) with those with some college education or a college degree.

Risk of hypertension (dichotomous outcome).
The risk of hypertension (ie, of being a case) was assessed in a test of the defined job-strain formulations with the use of unconditional logistic regression and control for the covariates used for ambulatory blood pressure. The multiplicative interaction was assessed by subtracting the -2 log likelihood (-2LL) of the saturated model from the model without the interaction term and comparing the result with that of the chi-square distribution. However, instead of the continuous linear term already described (demands-latitude), demands and latitude were first dichotomized and a three-level term was constructed by subtracting latitude from demands.
Missing values. Means were substituted for missing values for several covariates. In addition, missing values for social support for three subjects and for organizational influence for 20 subjects were replaced by worksite-specific means (since these two variables differed significantly by worksite). An analysis with these subjects removed produced similar results.

Sample characteristics
Descriptive statistics for the sample of 262 men are shown in table I. They averaged 44.3 years of age and 14.4 years of education (44% had at least a fouryear college degree) and were 84% Caucasian. They also averaged 8. I years with their current employer. Age, body mass index, alcohol use, and job strain had significant bivariate associations with blood pressure.

354
Decision latitude and demands were significantly correlated [correlation coefficient (r) = 0.33, P<O.OOI], as were the two subscales of job decision latitude, skill utilization and decision authority (r = 0.65). Job strain was not associated with education (14.3 years of education for "high strain" workers and 14.4 years for other workers). However, job strain was associated with lower social support [21.9 for "high strain" workers and 23.0 for other workers, t = 1.975, degrees of freedom (df) =260, P =0.049].

Association between job strain and ambulatory blood pressure (continuous outcome)
The associations between the various formulations of job strain and ambulatory blood pressure are listed in table 2.
Job-strain quadrant formulation. The 2 I % of employees defined as being under "high strain" with the use of the median cut points in this sample (figure I a) had systolic blood pressure at work that was significantly higher (6.7 mm Hg; "" 0.89 kPa) than that of the other employees. Work diastolic blood pressure was also significantly higher for these employees (2.7 mm Hg; ""0.36 kPa).
Other job-strain formulations. The other three formulations (multiplicative interaction, quotient, continuous linear) all exhibited significant associations with work and home systolic blood pressure, although the associations with diastolic blood pressure were not consistently significant.
Main effects. None of the main effects of job demands or job decision latitude on blood pressure were significant at P<0.05, except for job demands and work systolic blood pressure (regression coefficient = 0.40, P = 0.015) (data not shown). In addition, adding the previously reported (22,27) quadrant term to a model containing both main effects significantly increased the explained variance for work and home systolic blood pressure (R2 change = 0.018, P<0.05 for both), although the associations with diastolic blood pressure were not significant. In contrast, the increase in explained variance due to adding main effects to a model containing the quadrant term was much smaller and not significant.
Threshold. Choosing cut points that defined 14 and 18% of the workers as being under "high strain" produced similar associations to the previously reported quadrant term. However, when US national means for latitude and demands were applied (classifying 8% of the workers as being under "high strain," figure Ib), substantially greater associations with Table 2. Associations between formulations of job strai n and wor k and home ambulatory blood pressure (mm Hg) for 262 male employees from eight worksites in New York City (unstandardized regress ion coefficients a All regression models, and therefore (unstandardized) regression coeffic ients controlled for age, race, education, body mass index, smoking, alcohol, physical exertion, urine sodium, type A behavior, and worksite. Models with quadrant terms ind icate differences in ambulatory blood pressure between employees reporting job stra in and those not report ing jo b strain. b 1 mm Hg ",0.133 kPa. , P<0.05, , . P <0.01. '" P<0.OO1 . # P<0. 10. work blood pressure were observed [11.5 mm Hg ('" 1.53 kPa) systolic, 4.1 mm Hg ("'0.54 kPa) diastolic] than previou sly reported (22).
Dichotomizing the cont inuous quotient term (demands divided by latitude) at varying cut points along its range ( Proportion of subjects classified as exposed to job strain Figure 2. Distribution of associations between exposure to job strain and blood pressure among 262 male employees from eight worksites in New York City after dichotomization of the continuous job-strain variable (demands divided by latitude) at varying cut points. Associations controlled for age, race, education, body mass index, smoking, physical exertion, alcohol, urine sodium, type A, and worksite. (1 mm Hg =~O.133 kPa) subjects became too small.) Similar patterns were observed for home blood pressure.
Trichotomizing demands and latitude (figures 3 and 4) produced a slightly more complex threshold pattern. While the 6.5% of the sample in the highest strain cell (high demands plus low latitude) had a work systolic blood pressure of about 9-12 mm Hg (1.20-1.50 kPa) higher than those in low demand or high latitude cells, one other cell exhibited marked elevation. Employees with high demands and medium levels of decision latitude exhibited a 4-8 mm Hg (0.53-1.06 kPa) elevation. The patterns for diastolic blood pressure were not as clear. While the three highest job-strain cells had a significantly higher diastolic blood pressure than the rest of the sample, some lower strain cells exhibited similar high levels of diastolic blood pressure. The patterns of home blood pressure were similar to those of work blood pressure for this nine-cell model of job strain.
Latitude components. When low skill utilization and low decision authority were each combined with high demands, both were significantly associated with systolic blood pressure, although neither was consistently significantly associated with diastolic blood pressure. Slightly stronger associations were seen for low skill utilization, except when national means were used (table 2).
Composite latitude. Adding the organizational influence subscale to the decision latitude component of the job-strain quadrant term (table 2) resulted in marginally stronger associations with systolic blood pressure, except when a larger proportion (28%) was defined as exposed to job strain.
Social support. None of the main effects of any social support variable on blood pressure were signif-  Social support buffering. The associati on of job strain with diastolic blood pressure wa s stro nger for the subject s with low socia l support (table 3). Howe ver, the patterns for sys tolic blood pressure varie d considerably with small changes in the cut point, and the interac tion between job strain and socia l support was not statistica lly significa nt. Consistent with the aforementi oned finding, eac h of the "isostrain" measures slig htly increased the assoc iation s with diastolic blood pressure, but did not increase the assoc iations with systolic blood press ure (table 2).
Socioeconomic statu s. The association of jo b strain with work blood pressure was twic e as grea t for the subjects with no more than a high school education than it was for subjects with at least some college education (table 3). However, the interaction between job stra in and educatio n was not statistically significant. (Education was not sig nificantly associated with blood pressure. ) Asso ciation between j ob strain and hypertensive statu s (dic hotomous outcome) Job-strain quadrant formu lations . With the median cut-poi nt qu adrant term, j ob strain was associated with significa ntly increased odds of hypert ension [odd s ratio (OR) 2.9, 95% co nfidence inte rval (95% CI) 1.3-6.6] when controlled for the same covariates as in the precedin g analysis.
Other jo b-strain form ulations . Th e quotient term (dichotomized at 1.03 to defin e also 2 1% of the sample as high strain) produced a similar odds ratio (OR 2.8, 95% CI 1.2-6.2) as did the dichotomized linear term. Signi ficant associations with risk of hypertension were also observed for the continuous quotie nt and linear job-strain term s with the use of decision latitud e, as well as those with the use of skill utilization and decision authori ty.
However, the mult iplicative interacti on bet ween demands and latitud e was not significa nt (difference in -2LL =0.211, df = I) when controlled for the two main effec ts. Since hypertension was defin ed by casual diastolic blood pressure readings, this finding was consistent with the earlier observation of no multiplicative interaction effect on ambulatory diasto lic blood pressure.
Main eff ects. The risk of hypertension was sig nificantly associated with work-load demands above the median (OR 2.4, 95% CI 1.1-5.0) and decision latitude belo w the median (OR 2.1, 95% CI 1.0-4.4). Along with the nonsignificant multipli cati ve interaction between demands and latitud e, this fin ding sugges ts that risk was elevated as a res ult of the two main effec ts, rather than interaction. Th e three-Ievel linear (dem and s minus latitude) term produ ced a doubling of the risk (OR 2.2, 95% CI 1.3-3.6) for eac h unit change in the variable. Other measures of latitude were assoc iate d with slightly reduc ed ris k, but not sig nifica ntly.
Threshold . Choosi ng a cut point which defined 18% as being under "high strain" produced similar asso- Prop or tio n of s u b j ects clas sifie d as exposed to job s t r a in ciatio ns to those of the previou sly reported quad rant term (OR 2.8, 95% CI 1.2--fJ.7). However, quadrant ter ms using more restri cti ve cut poin ts, incl udi ng a term using national mean s, produ ced smaller odds ratios, with confidence intervals which included unity.
Dichotomizing the continuous quotient term across its range revealed odd s ratios fairly consistently above 2.0 and significant beginning at about the median (figure 5) . However, the sma ll number of subjec ts at the highest cut points (less than 19% of the sample) produced unstable estimate s and confidence intervals that included unity.
Latitude components. The low-authori ty co mponent of decision latitud e in co mbination with high demands was associated with increased risk of hyperten sion (OR 2.7, 95% CI 1.2-6.1). A similar odds ratio was obs erv ed for author ity when the nation al means were used; however, the confide nce interva ls included unity. Low skill utilization (plus high deman ds) produ ced smaller and nonsignificant odds ratios with either the sample medians or the national means.
Compo site latitude. The co mposite latit ude quadrant ter ms were associated with risk of hypertension (ta-ble 4), includi ng a term (co mpos ite latitude <36 and dem and s >32) with a substantially stro nger associatio n than the previously reported job-strai n ter m (OR 4.0,95% CI 1.7-9.2).
Social support. The social support measures were associated with a slightly, but not significa ntly reduced risk of hypertension (data not shown) .
Social suppor t buffering. Strain and social support did not sign ifica ntly interact (difference in -2LL = 0.066, df = I), and the buffering effects were not consistent, varyi ng at diffe rent cut points (table 5). However , the three dichotomous "isostrain" exposure variables were all associated with greater risk than previously reported (22 , 27) (table 4) , a findi ng sugge sting a combi nation of the main effects of jo b strai n and low soc ial support.
Socioeconomic status. The association of jo b strain with hypert ension risk among subject s with only a high school educa tion or less was substantially greater than for the subjects with at least so me co llege ed ucation (table 5) . However, stra in and education did not sign ificantly interac t (difference in -2LL =