Association between psychosocial work characteristics and health functioning in American women: prospective studyBMJ 2000; 320 doi: https://doi.org/10.1136/bmj.320.7247.1432 (Published 27 May 2000) Cite this as: BMJ 2000;320:1432
- Yawen Cheng, research fellowa,
- Ichiro Kawachi, director ()b,
- Eugenie H Coakley, research fellowc,
- Joel Schwartz, associate professord,
- Graham Colditz, professore
- a Department of Health and Social Behavior, Harvard School of Public Health, Boston, MA 02115, USA
- b Harvard Center for Society and Health, Harvard School of Public Health
- c Department of Nutrition, Harvard School of Public Health
- d Department of Environmental Health, Harvard School of Public Health
- e Channing Laboratory, Department of Medicine, Brigham and Women's Hospital, Boston MA 02115, USA
- Correspondence to: I Kawachi
- Accepted 7 February 2000
Objective: To examine prospectively the relation between psychosocial work characteristics and changes in health related quality of life over four years in a cohort of working women in the United States.
Design: Longitudinal cohort study.
Setting: United States.
Participants: 21 290 female registered nurses who completed the Karasek's job content questionnaire and a modified version of the short form 36 questionnaire (SF-36) as used for a survey of health status by the medical outcomes study.
Main outcome measures: Seven dimensions of health status: physical functioning, role limitations due to physical health problems, bodily pain, vitality, social functioning, role limitations due to emotional problems, and mental health.
Results: Examined separately low job control, high job demands, and low work related social support were associated with poor health status at baseline as well as greater functional declines over the four year follow up period. Examined in combination, women with low job control, high job demands, and low work related social support (“iso-strain” jobs) had the greatest functional declines. These associations could not be explained by age, body mass index, comorbid disease status, alcohol consumption, smoking status, education level, exercise level, employment status, marital status, or presence of a confidant.
Conclusions: Adverse psychosocial work conditions are important predictors of poor functional status and its decline over time.
The job strain model as conceptualised by Karasek and Theorell postulates that a combination of high psychological demands with low control at work leads to mental and physical illnesses.1–3 Previous studies have linked job strain to hypertension, cardiovascular disease, cigarette smoking, psychosomatic symptoms, depression, and adverse birth outcomes.1 3–9
With few exceptions,10 11 previous studies have assessed job strain on only one occasion. Although older workers have been reported to be more susceptible to job strain, it is unclear whether age is a proxy for cumulative exposure.4 12 Moreover, most studies have focused on conventional measures of mortality and morbidity. The impact of job strain on health functioning and sense of wellbeing have been reported in only a few recent studies.13–15
We studied the cumulative effects of job strain on health status in a large cohort of women in the United States, with repeated measures of job characteristics. We hypothesised that job strain not only predicts poor health status but also accelerates functional decline over time.
Participants and methods
Participants were from the nurses' health study, an ongoing cohort of female registered nurses in the United States. This study began in 1976 when 121 700 women (98% white) aged 30-55 years completed a mailed questionnaire regarding risk factors for major disease. Follow up questionnaires are sent biennially to update information on exposure and the incidence of new diseases.
The assessment of job characteristics and general health status began in 1992, when 75 434 women (69% response rate) returned a questionnaire that included the Karasek's job content questionnaire and the short form 36 health questionnaire (SF-36). Of these women, 34 558 were in the paid work force and were free of coronary heart disease, stroke, and cancer. Of these, 31 330 returned the follow up questionnaire in 1996. We excluded 1540 women who developed coronary heart disease, cancer, or stroke between 1992 and 1996. We further excluded 6814 women who retired and 1686 women with missing data on either any item in the job content questionnaire or more than half the items in a subscale of the SF-36. This left 21 290 women available for our study, 16 287 (76.5%) of whom were working as nurses in 1996.
Psychosocial job characteristics
Psychosocial job conditions were measured with the job content questionnaire.16 The job demands subscale is the sum of five items inquiring about excessive work, conflicting demands, insufficient time to work, fast pace, and working hard. The job control scale is the sum of two subscales: skill discretion as measured by six items (learning new things on the job, ability to develop new skills, job requiring skill, task variety, work not repetitious, job requiring creativity) and decision authority as measured by three items (freedom to make decisions, choice about how to perform work, having a lot of say in the job). The work related social support scale is the sum of two subscales: support from coworkers (four items) and supervisors (four). For each item the respondents could choose from one of four responses ranging from strongly disagree to strongly agree.
Health status was assessed by a modified version of the SF-36 as used for a survey of health status by the medical outcomes study.17 We examined seven of the eight dimensions of SF-36: physical functioning, measuring the ability to perform a variety of daily activities and tasks that require physical effort (10 items); role limitations due to physical health problems (four); freedom from bodily pain (two); vitality, measuring the perceived level of energy and fatigue (four); social functioning (two); role limitations due to emotional problems (three); and mental health, assessing both positive and negative emotional states (five). All participants responded to at least half the items in a scale. An average score was used to estimate the missing values (less than 1%), following procedures described in the SF-36 manual.17 All scales were transformed to range between 0 (worst health) and 100 (best health).
Potential confounding factors
Data on several biological and behavioural factors were obtained by questionnaire: age (years), body mass index (kg/m2), smoking status (never, former, and current smoker), alcohol intake (g/day), presence of comorbid conditions including diabetes, hypertension, hypercholesterolaemia, osteoarthritis, and rheumatoid arthritis (yes or no), employment status (part time or full time), presence of a confidant (yes or no), marital status (married or not married), educational attainment (advanced or graduate degree, registered nurse or bachelor degree), and exercise level (metabolic equivalent of work hours per week).
We examined the effects of psychosocial job characteristics averaged between 1992 and 1996 on the SF-36 measured in 1996. Additionally, we examined change in SF-36 scores over a four year period as an outcome. As expected, changes in SF-36 scores were negatively correlated with the baseline scores.18 19 We adjusted for baseline SF-36 scores measured in 1992, with the intercept terms constrained to be zero.
Multivariate linear regression models were used to control for potential confounding factors. To aid interpretability, the job control and job demands subscales were divided into thirds, whereas work related social support was split into two groups on the median. For continuous covariables, generalised additive models were applied to examine non-parametrically the shapes of their associations with SF-36 outcomes,20 and any necessary transformation was carried out. Age was treated as a continuous variable, and an additional age squared term was included to accommodate polynomial associations between age and most of the SF-36 outcomes. All analyses were conducted with the SAS program.
Table 1 compares the personal characteristics, job content scores, and SF-36 scores for 21 290 participants in 1992 with those of 13 268 participants included in a previous cross sectional study 15 but excluded from our study. Compared with women who were excluded, the participants were younger, better educated, less likely to have a chronic disease, and more likely to be a full time worker. The participants also reported higher levels of job demands (31.6 versus 30.8) and job control (73.1 versus 71.4). The mental health subscales were comparable between the two groups, although in terms of physical health subscales, women included in our study seemed to be healthier.
During the four year period the overall job demands of the group decreased slightly (31.6 in 1992 to 31.3 in 1996) whereas job control increased slightly (73.1 to 73.7). Job content assessed at two points of time showed a moderate degree of stability, with correlation coefficients for job control, job demands, and social support of 0.60, 0.54, and 0.41 respectively. The demands and control scores were not correlated with each other (correlation coefficients less than 0.10). Work related social support was positively correlated with job control (r=0.20 to 0.37) but negatively correlated with job demand (r=−0.10 to −0.22).
Having a body mass index greater than 29 kg/m2, not being married (including never married, separated, divorced, and widowed), having a lower level of educational attainment, and being a current smoker were all associated with lower SF-36 scores in all the seven subscales.
The follow up SF-36 scores assessed in 1996 were regressed on job content scores averaged between 1992 and 1996 (table 2). In all the models, significantly better health status was found in women with higher levels of job control, lower levels of job demands, and higher levels of work related social support after adjustment for potential confounders. The effect sizes were greater when the average scores of job strain over the four year period were used compared with either 1992 or 1996 job content scores alone.
The joint effects of job demands and control on SF-36 were next examined by including dummy variables made up of combinations of each subscale divided into thirds. Although results of only three subscales (physical functioning, vitality, and mental health) are shown in table 3 owing to space limitations, all the seven subscales were analysed, and the results showed a similar pattern. Women in the highest third of job demands and the lowest third of job control (reference group, “high strain” job) had the worst health status, whereas those in jobs with the highest control and lowest demands (“low strain” job) had the best health status. When analyses were stratified by level of work related social support, women reporting lower social support had lower scores in all the SF-36 subscales compared with women reporting higher social support.
All effect estimates changed little when analyses were restricted to women who remained in nursing (data not shown). The SF-36 scales assessed at two points of time were moderately correlated, with correlation coefficients ranging from 0.32 for the role emotional subscale to 0.63 for the vitality subscale. Over the four years of follow up, scores in the physical health domain deteriorated, whereas vitality and mental health status improved (table 4). Women reporting low job control, high job demand, and low work related social support had greater declines in the physical health subscales and less improvement in mental health subscales. The patterns remained after adjustment of other covariables.
Women in jobs with high demands, low control, and low social support (“iso-strain” jobs) showed the greatest declines in health status. Our findings supported our hypotheses that cumulative measures of job demands and control provided a more stable assessment of the associations than concurrent measures. The effects of job strain were independent of a variety of potential confounders.
A comparison between participants and those who were excluded from analysis because of major illness or retirement suggested a healthy worker effect. Healthier employees are more likely to remain working whereas those with health problems may shift to jobs with lower strain or quit work altogether. If participants shifted to jobs with lower strain because of their health problems there would be a selection bias leading to an underestimation of the effects of job strain on health.
There is some debate about whether the job demand dimension predicts health. A review by Schnall et al 4 found significant associations between job control and cardiovascular outcomes in 17 of 25 studies (68%), whereas associations with job demands were found in only eight of 23 studies (35%). Several recent substudies from the Whitehall II study of British civil servants also showed that poor health was associated with lower job control but not with high job demands.11 14 21 Questions have been raised regarding colinearity between work characteristics and socioeconomic status. Does the better predictive power of job control reflect a closer correlation to socioeconomic position than job demands? In the Whitehall II study, high job demands and high job control were highly correlated—both were more prevalent in higher employment grades.22 In the nurses' health study, the low correlation between job control and job demands meant more power to examine their independent effects on health.
A previous report from this cohort showed that social networks were positively associated with mental functioning.23 Our study extends those findings by indicating the deleterious effects of low social support in the workplace. This finding is consistent with other studies.7 12 24
An important issue concerns the validity of self reported data. Since both the exposure and outcomes were assessed by self report, there was potential for bias by people reporting in a negative manner on both. People with negative affectivity may perceive their environment more negatively, creating an artificial correlation between work environment and health outcomes. Reserve causation is another major issue in research into job stress.25 It has been argued that unhealthy or less socially competent people may drift to worse jobs with lower levels of control over decision making. Both these sources of bias may partly account for the observed associations in our study. Objective measures of job content and information on personality are needed to clarify this issue.
Currently, most solutions proposed to reduce job stress, such as relaxation therapy and modification of lifestyle, target individuals rather than their social environment and tend to deal with symptoms instead of causes. If the psychosocial work environment contributes to the quality of life of the workforce—as suggested by our study—hospitals and medical practices will need to focus their strategies for health promotion on the redesign of jobs.
What is already known on this topic
Stressful job conditions, characterised by low control, high demands, and low social support, increase the risk of cardiovascular disease
Previous cross sectional studies suggested that job strain is associated with low functional health status
What this study adds
A prospective study of 21 290 female nurses in the United States found that low control in their jobs predicted significant declines in physical function and mental health
The effects of job strain on functioning were independent of socioeconomic status, baseline functioning, and other confounders
The declines in health functioning associated with job strain were as large as those associated with smoking and sedentary lifestyles
We thank the members of the nurses' health study for their continuing participation and support.
Contributors: YC participated in the study design, carried out the data analyses, and took primary responsibility for writing the paper. IK formulated the primary study hypotheses, designed the study, and contributed to the interpretation of the findings and the preparation of the manuscript; he will act as guarantor for the paper. EHC provided statistical and programming advice, documented the data, and participated in the preparation of the manuscript. JS provided statistical advice and helped in interpretation of the data and editing the manuscript. GC is the principal investigator of the study. He helped develop the primary study hypotheses, took part in the study design, and contributed to the interpretation of the data and editing of the manuscript.
Funding This research was funded by a grant from the National Institute of Aging Research (R01-AG-12806). The nurses' health study is funded by a grant from the National Institutes of Health (R01-CA-40356). IK was supported by the MacArthur Foundation Network on Socioeconomic Status and Health and a career development award from the National Institutes of Health (HL03453).
Competing interests None declared.