The Socioeconomic Gradient in the Incidence of Stroke
A Prospective Study in Middle-Aged Women in Sweden
Background and Purpose— A socioeconomic gradient in stroke has been demonstrated in a variety of settings, but mostly in men. Our purpose was to establish whether a socioeconomic gradient in stroke existed in a group of Swedish women and whether this gradient could be explained by established stroke risk factors or psychosocial factors.
Methods— The Women’s Lifestyle and Health Cohort Study includes 49 259 women from Sweden aged 30 to 50 years at baseline (1991 to 1992). The women completed an extensive questionnaire and were traced through linkages to national registries until the end of 2002. Among the 47 942 women included in these analyses, there were 200 cases of incident stroke during follow up (121 ischemic stroke, 47 hemorrhagic stroke, and 32 of unknown origin). Statistical analysis was through the Cox proportional hazards model.
Results— The risk of stroke was significantly inversely related to years of education completed, our proxy for socioeconomic status (hazard ratio comparing lowest with highest education group=2.1, 95% CI: 1.4 to 2.9, P for trend <0.001). This association was reduced after adjustment for established risk factors, although remaining significant (1.5, 1.0 to 2.2, P for trend=0.04). The gradient was more pronounced for ischemic stroke (2.9, 1.8 to 4.7, P for trend <0.001) than for hemorrhagic stroke (1.4, 0.7 to 2.9, P for trend=0.35). Job strain and social support were unrelated to risk of stroke. Self-rated health was strongly related to risk of stroke mediated by established risk factors. Psychosocial factors did not contribute toward the socioeconomic gradient in stroke.
Conclusions— There was a strong gradient in risk of stroke by years of education, especially for ischemic stroke. Most of the social gradient was explained by established risk factors, particularly smoking and alcohol, but not by psychosocial factors.
Stroke is a dominant cause of mortality and morbidity throughout the world.1 Each year, approximately 15 million people will have a stroke, which will be fatal for 5 million people and permanently disabling for a further 5 million.2 Stroke is a leading cause of death and disability in Sweden, third only to ischemic heart disease and depression/neurosis in terms of disability-adjusted life-years lost.3 People in low socioeconomic groups are at increased risk of stroke,4 even in egalitarian Sweden.3 The social disparity in stroke has persisted over time, despite the overall fall in mortality from stroke.5
Most studies that have tried to explain the basis for the socioeconomic gradient in stroke show that it is reduced after taking account of conventional stroke risk factors,6–11 although a significant excess risk in the lower socioeconomic classes often remains.10,12–15 The persistent excess risk may be attributable to residual confounding, because adjustment for risk factors is usually incomplete. Psychosocial factors such as stress at work may also play a role because they are associated with cardiovascular disease,16 stroke risk factors,17 and socioeconomic status,18 and men with poor adaptation to stress have an increased risk of stroke.13,19
The aim of this study is to establish whether a socioeconomic gradient in stroke existed in a group of Swedish women,20 and whether this gradient could be explained by established stroke risk factors or psychosocial factors using comprehensive assessment of risk factors and validated stroke outcome measures.
Materials and Methods
The Women’s Lifestyle and Health cohort was enrolled during 1991 and 1992. A sample of 96 000 women born between 1943 and 1962 (aged 30 to 49 years) residing in the Uppsala Health Care Region were randomly selected from the Swedish Central Population Registry at Statistics Sweden and sent a survey questionnaire.20 A total of 49 259 women returned a completed mailed questionnaire (response rate 51%). Each woman is identified by a unique 10-digit national registration number, which encodes information on date of birth and gender.21 For the current study, we excluded 1091 women because they did not provide information about their years of education (the primary exposure for the study). We excluded a further five women who had emigrated before the start of follow up and 221 women with a history of stroke or myocardial infarction before the start of follow up. The final study population included 47 942 women.
At baseline, the women completed a detailed self-administered questionnaire. Socioeconomic status was estimated using self-reported total years of school attendance in four categories22:
Seven to 9 years (primary school with at most 2 years of additional professional education);
Ten to 12 years (completed secondary school or up to 5 years of professional training);
Thirteen to 15 years (university bachelor degree or several professional training sessions); and
Sixteen or more years (usually corresponds to university master’s level degree or higher).
Work characteristics were measured using established questionnaires for the central components of the job strain model, that is, job demands (five questions), job control (six questions), and social support at work (six questions).23 Scores for each scale were calculated as the sum of the item scores and the population was divided into tertiles for each characteristic based on the responses across all the women. The few subjects missing one or two items in a scale were assigned an average score based on the items that they did answer, and those missing more than two items on a scale were excluded from the analyses (419 women for job demands, 284 for job control, 674 for social support at work). Four quadrants of job strain (“active work”—high demand and high control, “high strain”—high demand and low control, “low strain”—low demand and high control, “passive work”—low demand and low control) were constructed by crosstabulating job demands and job control, both divided at the median.
Social support in general was measured using six questions. A score was created and the women were divided into tertiles, although 351 women were excluded because they did not answer more than two questions. Women were also asked to give a personal assessment of their health. Health-seeking behavior was measured by enquiring about the frequency of breast self-examination, mammography screening, and gynecologic checkups.
Conventional Stroke Risk Factors
Participants reported on established stroke risk factors, which were: cigarette smoking (never smoker, <5, 5 to <10, ≥10 pack-years), physical activity (very low, low, normal, high, very high), alcohol consumption (0, <1.7, 1.7 to 4.4, ≥4.4 g/d), body mass index (<18.5, 18.5 to <25, 25 to <30, ≥30 kg/m2), diabetes (yes/no), and high blood pressure (yes/no). The women with missing data for previous stroke (n=3543), previous myocardial infarction (n=3558), diabetes (n=3363), or hypertension (n=2313) were assumed not to have prevalent disease.
Follow Up and Stroke End Points
The cohort was followed up through linkages with existing nationwide health registers using the unique national registration number of the women so that follow up was virtually complete with respect to death, emigration, and stroke. Information on stroke was collected through linkage to the National Hospital Discharge Register International Classification of Diseases (ICD), 9th Revision from 1987 to 1996 and the 10th version thereafter. We considered cases in the Inpatient Register with any of the following main diagnoses: ischemic stroke (occlusion of cerebral arteries, IS) (ICD-7: 332; ICD-8: 433 to 434; ICD-9: 434; ICD-10: I63.3 to I63.9), intracerebral hemorrhage (ICD-7: 331; ICD-8, 9: 431; IC-D10: I61), and undefined stroke (ICD-7: 334; ICD-8, 9: 436; ICD-10: I64). Because some patients might have experienced sudden death attributable to stroke without hospitalization and recording in the Inpatient Register, we also linked our cohort to the nationwide Causes of Death Register. If a subject was found to have different diagnoses of stroke within 28 days after index diagnosis, the subtype was defined by the latest hospital discharge. We obtained information on date of death from other diseases from the Causes of Death Register and on date of emigration out of Sweden from the Emigration Register.
The start of follow up was defined as the date of receipt of the returned questionnaire and person-years were calculated from the start of follow up to the primary diagnosis of fatal or nonfatal stroke, date of emigration or death, or the end of follow up (December 31, 2002), whichever came first. The average length of follow up was 11.2 years. In total, there were 200 events (121 ischemic, 47 hemorrhagic, and 32 of unknown origin).
We calculated relative hazards using the Cox proportional hazards model to assess whether years of education was associated with the age-adjusted incidence of stroke. The association between incidence of stroke and, in turn, general social support, self-rated health, health-seeking behavior, and job variables (among women in full- or part-time employment) was also modeled. We interpreted hazards ratios as estimates of relative risk, and these were reported with 95% CIs. We tested for trends across categories of variables by assigning equally spaced values (eg, 1, 2, 3, or 4) to the categories and treating the variables as continuous variables in the Cox proportional hazards model. All analyses were adjusted for age at baseline, which was categorized by 5-year intervals. The models were successively adjusted for established stroke risk factors: cigarette smoking, body mass index, alcohol consumption, self-reported diabetes, self-reported high blood pressure, and exercise.
The study was approved by the Data Inspection Board in Sweden and by the regional Ethical Committee. Consent was assumed by the return of the postal questionnaire.
Of a total of 47 942 women included in the study, 19.7% had completed less than 10 years of education, 39.1% had completed 10 to 12 years, 32.9% had completed 13 to 15 years, and 8.4% had completed at least 16 years (Table 1). Women were on average 40.3 years (±5.8 SD) at baseline. Longer education was associated with a lower prevalence of stroke risk factors, including fewer pack-years of smoking, lower average body mass index, a lower prevalence of diabetes and hypertension, and more frequent exercise.
The incidence of all stroke was strongly related to years of education (hazard ratio comparing the lowest with highest level of education=2.1, 95% CI=1.4 to 2.9, P for trend <0.001) (Table 2). For these analyses, we combined group 3 (13 to 15 years of education) and group 4 (≥16 years of education) because only 12 events occurred in the highest education group. This association was more marked for ischemic stroke (2.9, 1.8 to 4.7, P for trend <0.001) than for hemorrhagic stroke (1.4, 0.7 to 2.9, P for trend=0.35). Adjustment for smoking explained some of the educational gradient for stroke, as did adjustment for alcohol, but adjustment for the other stroke risk factors had less influence on the effect estimates. After adjusting for all stroke risk factors, the association between education and all stroke (1.5, 1.0 to 2.2, P for trend=0.04) and ischemic stroke (2.2, 1.3 to 3.7, P for trend=0.003) was weakened, although it remained statistically significant.
The analyses for work characteristics were restricted to the 35 471 women in full-time (n=19 533) or part-time (n=15 938) employment. The educational gradient in risk of stroke was still apparent after the sample was restricted to women in full- and part-time employment (hazard ratio comparing the lowest with the highest education group=2.2, 1.4 to 3.3, P for trend <0.001) (Table 3). The gradient remained more pronounced for ischemic stroke (2.7, 1.6 to 4.6, P for trend <0.001) than for hemorrhagic stroke (1.9, 0.8 to 4.3, P for trend=0.16). Adjustment for the psychosocial factors individually, or together, did not explain the educational gradient in all stroke (2.2, 1.4 to 3.5, P for trend <0.001), ischemic stroke (2.5, 1.4 to 4.6, P for trend=0.002), or hemorrhagic stroke (1.7, 0.7 to 4.2, P for trend=0.26).
Job control, job demands, job strain, and social support at work were essentially unrelated to risk of all stroke or hemorrhagic stroke during follow up (Table 4). There was a borderline significant increased risk of ischemic stroke among women who had low job control (hazard ratio for low versus high job control=1.4, 0.9 to 2.4, P for trend=0.08) or high job strain (hazard ratio for high versus low job strain=1.6, 0.9 to 3.0). Ischemic stroke was not associated with job demands or social support at work.
Self-rated health was strongly associated with risk of all stroke (hazard ratio low versus high self-rated health=3.4, 2.1 to 5.6, P for trend <0.001) and ischemic stroke (4.5, 2.4 to 8.4, P for trend <0.001), but not hemorrhagic stroke (1.2, 0.3 to 5.1, P for trend=0.27) (Table 5). Most of the excess risk of stroke among women with low self-rated health disappeared after adjustment for stroke risk factors, although the associations for all stroke and ischemic stroke remained statistically significant. Low social support was associated with risk of all stroke (hazard ratio for low versus high social support=1.6, 1.1 to 2.2, P for trend=0.006), ischemic stroke (1.5, 1.0 to 2.4, P for trend=0.05), and hemorrhagic stroke (2.0, 1.0 to 4.1, P for trend=0.05), although most of the excess risk was explained by adjustment for stroke risk factors. Low health-seeking behaviors were inversely related to the risk of stroke, and this was independent of stroke risk factors.
In this large prospective study, we found a strong inverse association between educational attainment and stroke among middle-aged Swedish women. The gradient was particularly pronounced for ischemic stroke. Most of the educational gradient was explained by established stroke risk factors, particularly smoking and alcohol, whereas psychosocial factors did not contribute toward the gradient.
The social gradient in stroke could be driven by variation in stroke risk factors, health-seeking behaviors, or psychosocial risk factors by social stratus. Most of the studies that tried to explain the basis for the socioeconomic gradient in stroke have shown that the gradient is reduced after taking account of conventional stroke risk factors,6–11 although a significant excess risk remained in the lower socioeconomic classes.10,12–15 The results from the current study support this finding and are consistent with a recent study that showed a strong association between educational level and healthy lifestyle in women.24 Health-seeking behavior and psychosocial factors could not explain the educational gradient in stroke. The evidence suggests that the social gradient in stroke was largely driven by conventional stroke risk factors. The stronger social gradient in ischemic stroke than in hemorrhagic stroke (consistent with the findings of other studies)10 would support this finding, because ischemic stroke is related to conventional stroke risk factors, whereas hemorrhagic stroke is more often caused by structural abnormalities. Any excess risk in the lower socioeconomic groups that persists after adjusting for risk factors may be attributable to residual confounding or unmeasured confounding, because risk factors were only measured through self-report and at one point in time.
Few studies have investigated psychosocial factors as a cause of stroke or as mediators of the socioeconomic gradient in stroke,11 and no previous cohort studies have investigated the association between work stress and stroke. The results of the present study do not support the existence of an association between work stress and overall risk of stroke, although the association between low job control and job strain and ischemic stroke approached statistical significance. We found some association between risk of stroke and social support, although this was largely explained by established stroke risk factors. Self-rated health consistently predicts overall mortality,25 because it may capture subtle symptoms of subclinical disease, and so the association between self-rated health and stroke may be attributable to incomplete control for baseline health status.
There were some limitations to the study. Only half of the women who were contacted agreed to participate in the study, and this may have introduced the possibility of a selection bias. However, a validation exercise from the Norwegian part of the cohort indicated that the cohort participants were representative of the general population,26 reducing the potential for bias. The exposure variables were only measured at baseline, which may have resulted in residual confounding of the association between education and stroke. This would not have had an important effect for education, our main exposure variable, because this is unlikely to change among women after the age of 30. The misclassification of the psychosocial variables is expected to be nondifferential with respect to outcome and so this could have led to an underestimation of effects.27 We were not able to measure access to health care in this study, although we used a proxy measure for health-seeking behavior, and in Sweden, everyone has free access to high-quality medical services. Some potentially important stroke risk factors were not measured in the study such as childhood socioeconomic conditions, cholesterol and fibrinogen levels, migraine with aura, atrial fibrillation, drug use, and cerebrovascular disorders; and these unmeasured confounders may have contributed toward explanation of the social gradient in stroke. There were relatively few cases of hemorrhagic stroke, and this could limit interpretation of the data and result in wide confidence intervals in the adjusted models.
There were also strengths. This study simultaneously assessed the role of conventional stroke risk factors and psychosocial risk factors in explaining the socioeconomic gradient in stroke and also investigated the effect of job strain on stroke in women. We used validated outcomes to measure incident stroke among women who were disease-free at baseline and measured job strain through standardized questionnaires. Disease end points were obtained through the In-Patient Register and Mortality Register, which allowed complete follow up of the cohort. This study was large and had extended follow up, and the women in the cohort were likely to be representative of the general population.26
Stroke is a dominant cause of morbidity and mortality throughout the world.1,2 The social gradient in stroke offers us an insight into prevention of stroke because we can aim for the incidence in the lowest social strata to approach that of the highest social strata. The study results indicate that in this cohort of women aged 30 to 50 at baseline, most of the social gradient is attributable to established risk factors, particularly smoking and alcohol. Because socioeconomic status may be a good proxy to identify individuals at increased risk for stroke, a health promotion campaign targeting lower educational groups may reduce the population incidence of stroke, although other correlates of low education need further investigation.
In this cohort of middle-aged Swedish women, followed on average for more than 11 years, we found a strong gradient in risk of stroke by education level. Most of this gradient was explained by established conventional risk factors, but not by job characteristics, other psychosocial variables, or health-seeking behavior.
The authors thank all the women who contributed to this study.
Sources of Funding
In Sweden, the survey was supported by the Swedish Council for Planning and Co-ordination of Research, Swedish Cancer Society, STINT (The Swedish Foundation for International Cooperation in Research and Higher Education) Organon, Pharmacia, Medical Products Agency and Schering-Plough. The travel costs for Hannah Kuper and Elisabete Weiderpass were supported by a joint program grant from the Royal Society.
- Received June 21, 2006.
- Revision received September 1, 2006.
- Accepted September 5, 2006.
Mackay J, Mensah G. Atlas of Heart Disease and Stroke. Geneva: World Health Organization, 2004.
Avendano M, Kunst AE, van Lenthe F, Bos V, Costa G, Valkonen T, Cardano M, Harding S, Borgan JK, Glickman M, Reid A, Mackenbach JP. Trends in socioeconomic disparities in stroke mortality in six European countries between 1981–1985 and 1991–1995. Am J Epidemiol. 2005; 161: 52–61.
Shaper AG, Phillips AN, Pocock SJ, Walker M, Macfarlane PW. Risk factors for stroke in middle aged British men. BMJ. 1991; 302: 1111–1115.
Hart CL, Hole DJ, Smith GD. Influence of socioeconomic circumstances in early and later life on stroke risk among men in a Scottish cohort study. Stroke. 2000; 31: 2093–2097.
Avendano M, Kawachi I, Van Lenthe F, Boshuizen HC, Mackenbach JP, Van den Bos GA, Fay ME, Berkman LF. Socioeconomic status and stroke incidence in the us elderly: the role of risk factors in the EPESE study. Stroke. 2006; 37: 1368–1373.
Salonen JT. Socioeconomic status and risk of cancer, cerebral stroke, and death due to coronary heart disease and any disease: a longitudinal study in eastern Finland. J Epidemiol Community Health. 1982; 36: 294–297.
Everson SA, Lynch JW, Kaplan GA, Lakka TA, Sivenius J, Salonen JT. Stress-induced blood pressure reactivity and incident stroke in middle-aged men. Stroke. 2001; 32: 1263–1270.
Morris RW, Whincup PH, Emberson JR, Lampe FC, Walker M, Shaper AG. North–south gradients in Britain for stroke and CHD: are they explained by the same factors? Stroke. 2003; 34: 2604–2609.
Kivimaki M, Leino-Arjas P, Luukkonen R, Riihimaki H, Vahtera J, Kirjonen J. Work stress and risk of cardiovascular mortality: prospective cohort study of industrial employees. BMJ. 2002; 325: 857.
Kuper H, Marmot M. Job strain, job demands, decision latitude, and risk of coronary heart disease within the Whitehall II study. J Epidemiol Community Health. 2003; 57: 147–153.
Andre-Petersson L, Engstrom G, Hagberg B, Janzon L, Steen G. Adaptive behavior in stressful situations and stroke incidence in hypertensive men: results from prospective cohort study ‘Men born in 1914’ in Malmo, Sweden. Stroke. 2001; 32: 1712–1720.
Kumle M, Weiderpass E, Braaten T, Persson I, Adami HO, Lund E. Use of oral contraceptives and breast cancer risk: the Norwegian–Swedish women’s lifestyle and health cohort study. Cancer Epidemiol Biomarkers Prev. 2002; 11: 1375–1381.
Lunde AS, Lundeborg S, Lettenstrom GS, Thygesen L, Huebner J. The person-number systems of Sweden, Norway, Denmark, and Israel. Vital Health Stat 2. 1980; 2: 1–59.
Kuper H, Adami HO, Theorell T, Weiderpass E. Psychosocial determinants of coronary heart disease in middle-aged women: a prospective study in Sweden. Am J Epidemiol. 2006; 164: 349–357.
Karasek RA, Theorell T. Healthy Work: Stress, Productivity and the Reconstruction of Working Life. New York: Basic Books, 1990.
Gerstman BB. Inaccuracy in Epidemiologic Studies, Part II (bias). Epidemiology Kept Simple: An Introduction to Classic and Modern Epidemiology. New York: Wiley-Liss, 1998: 183–200.