Globally, stroke is the third-leading cause of death and disability, and lower-income and lower-middle–income countries carry the absolute majority of the stroke burden.1 Low socioeconomic status (SES), regardless of whether it is measured between or within countries, has repeatedly been linked to an increased risk of stroke, more severe strokes, and poor outcomes, including higher mortality and increased disability.2,-,4

Although it has also been demonstrated that patients with low SES have higher prevalence of cardiovascular risk factors and are underprivileged regarding quality of stroke care, access to stroke care, and secondary prevention after stroke,5,-,10 the mechanisms by which SES affects adverse stroke outcomes remain largely unknown. Previous studies using mediation analysis to explain the SES-adverse outcome relationship have found that stroke severity seems to be an important mediator—both in terms of short-term mortality11 and long-term disability,12 while quality of acute care was found to explain very little of the effect of SES on short-term mortality and readmission.13

In this nationwide register-based cohort study on stroke patients in Sweden, we use novel mediation analysis methods that allow for the evaluation of multiple mediators at once14,15 to investigate the connections between SES and adverse outcome (death and dependency) 3 months after stroke. We explore the extent to which SES disparities would remain if we could perform interventions to eliminate differences in comorbidity, stroke severity, and/or acute treatment.

We identified 31,807 eligible patients, of whom 26,983 (84.8%) were either followed up or dead 3 months after stroke (Figure 1). Of them, 1,137 (4.2%) were excluded because of missing values on 1 or more of the analysis variables (except for the NIHSS). This left a final study population of 25,846 patients, with an average age of 74.4 years (SD = 11.9), 46.7% of whom were female. A total of 6,798 (26.3%) patients were dead or ADL dependent 3 months after stroke, with a higher risk of adverse outcome among patients with low SES, compared with mid or high SES (Table 1, eTable 2, links.lww.com/WNL/D208).

The proportion of female patients decreased with increasing SES (Table 1), while the average age was highest in the low SES group. Patients in lower SES groups had higher proportions of diabetes and atrial fibrillation and were more often prescribed antihypertensive and antiplatelet drugs than those in the high SES group, while differences in statin and anticoagulant treatments were negligible between different SES levels (Table 1, eTable 2, links.lww.com/WNL/D208).

Lower SES was associated with a higher risk of moderate-to-severe strokes (Table 1), and the proportions remained similar after imputation: median 35.1% (min–max 34.0%–36.5%) for patients with low SES, 27.1% (26.7%–27.6%) for patients with mid SES, and 22.9% (22.1%–23.8%) for patients with high SES. Acute care measurements including reperfusion therapy and treatment at stroke unit increased with increasing SES, although the differences for the latter were small (Table 1, eTable 2, links.lww.com/WNL/D208).

After adjustment for the baseline confounders sex and age (age + age-squared), low SES was associated with a higher risk of death or ADL dependency at 3 months, compared with both mid and high SES (Table 2, column 1). The associations remained but were reduced after further adjustments were made for mediators (Table 2, column 2).

Logistic Regression Models of the Associations Between Exposure, Mediators, and Outcome

Low SES was associated with an increased risk of most comorbidities compared with those with mid and high SES, except for atrial fibrillation and treatment with statins and anticoagulants (Table 2, column 1). For the other mediators, low SES was associated with an increased risk of more severe strokes and a decreased chance of reperfusion therapy, while effects pertaining to the stroke unit care variable were small.

Independent of sex, SES, age, and other mediators, smoking, a medical history of diabetes, atrial fibrillation, and previous stroke were associated with an increased risk of death or ADL dependency at 3 months, while the effects of prescribed medications and stroke unit care were smaller (Table 2, column 2). Stroke severity was associated with a strong independent increase in the risk of death or ADL dependency, while reperfusion therapy was associated with a decreased risk.

After adjustment for sex and age (age + age-squared), low SES was associated with an increased absolute risk of death or ADL dependency at 3 months of 5.4% (95% CI 3.9%–6.9%) compared with mid SES and of 10.1% (8.1%–12.2%) compared with high SES, and just more than 60% of this increased risk would remain if all mediators were shifted to have the same distribution among patients with low SES as that of the more privileged patients (Table 3).

Estimated Adjusted Total Association and Interventional Disparity Direct and Indirect Effects

If we could intervene to shift the distribution of all mediators among patients with low SES to the distributions of those with higher SES, the absolute risk reduction in death or ADL dependency would be 2.2% (95% CI 1.2%–3.2%) compared with patients with mid SES and 4.0% (95% CI 2.6%–5.5%) compared with patients with high SES (Table 3). Much of this reduction among patients with low SES would be accomplished by intervening on stroke severity accounting for 1.5% (95% CI 0.6%–2.3%) and 2.6% (95% CI 1.5%–3.8%), respectively, of the increased absolute risk, compared with those with mid and high SES. Interventions focused on shifting the distributions of comorbidities, reperfusion therapy, and the dependence between mediators would yield smaller decreases in the absolute risk difference, while the indirect effects of stroke unit care were close to zero (Table 3).

This nationwide study showed that low SES was associated with a 5% increase in the absolute risk of death or ADL dependency 3 months after ischemic stroke compared with mid SES and a 10% increase compared with high SES. Approximately 40% of these excess risks were mediated through factors in the causal pathway, including comorbidities, stroke severity, and reperfusion therapy. This suggests that it could be possible to save 40 of every 1,000 patients in the low SES group from dying or becoming ADL dependent if we could equalize SES differences in comorbidity, stroke severity, and reperfusion therapy.

The increase in the risk of death and dependency for patients with low SES is in line with findings from previous studies on short-term mortality,2,32,-,35 disability,36 and the composite outcome of death or disability.37

Stroke severity was by far the most important mediator in this study. Previous studies have found that initial stroke severity explained approximately 40% of income inequalities in 3-month case fatality11 and more than 60% of income inequalities in long-term disability after ischemic stroke.12 We have previously studied the link between education level and stroke severity and found that nearly 30% of the effect was an indirect effect mediated through cardiovascular disease (CVD) risk factors (including smoking, diabetes, atrial fibrillation, previous stroke, and ADL dependency before the stroke).38 In this study, we found that part of the effect of SES on adverse outcome could be eliminated by only equalizing the distribution of comorbidities (risk factors and prescribed medications). Together with the likely importance of risk factors in the SES-stroke severity relationship, this means that an obvious target for clinical interventions aiming to reduce disparities in stroke outcomes would be to reduce disparities in comorbidities and risk factors of stroke among patients with low SES. Here, hypertension, atrial fibrillation, and diabetes are all important components related to lifestyle factors. The risks of hypertension and diabetes have been found to be modifiable by regular physical activity, a healthy diet, and weight loss, and hence, lifestyle changes including smoking cessation should be aggressively promoted, especially among those with low SES.39,40 Apart from physical inactivity and obesity, hypertension and diabetes can increase the risk of atrial fibrillation, and hence medication for hypertension and diabetes together with lifestyle changes are important in reducing the risk of atrial fibrillation.41

Previous studies have found that there is unequal access to acute stroke care across SES groups,5,6,8,9 and while differences in access to stroke unit care were small in our study, we found that there was unequal access to reperfusion treatment with patients with low SES less likely to receive reperfusion therapy. However, our results suggest that inequalities in adverse outcome at 3 months are not driven by inequalities in acute care. This is in line with a Danish mediation study on income inequalities in 30-day mortality and readmission, which found no mediating effect of quality of early care.13

The method used in the study relies on an assumption that there was no unobserved confounding of the mediator-outcome relationships (see the eMethods, links.lww.com/WNL/D211 for more details). Through linking individual registers, we were able to include several confounding factors and possible mediators. We were, however, limited to the variables collected by the registers. Functional outcome at 3 months is patient reported and based on a questionnaire and does not include the modified Rankin scale. However, ADL dependency based on questions in Riksstroke has shown good agreement with the modified Rankin scale,42 and with Barthel index,43 and we do not expect that this would have any major effects on the findings.

We had access to information on prescribed medications at the time of stroke (e.g., antihypertensives, statins) and CVD-related comorbidities (atrial fibrillation, diabetes, and previous stroke), but no or limited information on postacute care, patient preferences, lifestyle (e.g., alcohol consumption, physical activity), compliance with medications, other comorbidities (e.g., renal disease, heart failure, dementia, and cancer), other social determinants (e.g., occupation, neighborhood-level SES), or clinical measurements such as blood pressure or cholesterol levels. Furthermore, we did not consider stroke awareness and help-seeking behavior, factors that may lead to increased onset-to-door times and reduced benefit of reperfusion therapy. A previous review suggested that help-seeking behavior is more dependent on perceived severity of symptoms than on actual knowledge of symptoms and that delays were not related to sociodemographic factors.44 We were able to adjust for the major baseline confounders sex and age but cannot rule out residual confounding. Additionally adjusting for the hospital where the patient was treated as a sensitivity analysis for possible confounding by region did not have a major impact on the estimated logistic regression model parameters (eTable 3, links.lww.com/WNL/D209). An aim of future studies should be to broaden the included variables and mediators to further elucidate the complex relationship between SES and outcomes after stroke.

Stroke severity was measured using the NIHSS, dichotomized into mild stroke (0–5) and moderate-to-severe (>5) stroke. This dichotomization has been used in other studies, both as a predictor45 and as an outcome measure.46 While our results indicate that a substantial reduction in the death or ADL dependency disparity could be achieved by shifting the distribution of mild stroke (NIHSS 0–5) vs moderate-to-severe stroke (NIHSS >5) in patients with low SES to that of more privileged patient groups, it is possible that shifts based on a different cutoff or on a more fine-grained scale of the NIHSS could lead to reductions of a different magnitude. A more objective measure would be achieved by using imaging to gain information on infarct volume and location. Such information is not currently available in Riksstroke. NIHSS was missing for nearly half of the patients. However, we had extensive information on patient characteristics, level of consciousness at hospital admission, treatment, and outcome, which in combination with the choice to impute NIHSS to 2 categories rather than the full scale of measurement makes us expect no major deviations from the missing-at-random assumption. In addition, we used a flexible imputation model to reduce the risk of model misspecification. Under these conditions, a previous simulation study has shown that multiple imputation offers unbiased results, even with large proportions of missing data (up to 90% missing).47 Less than 5% of patients were missing data on variables other than the NIHSS. These patients were excluded.

The study is based on a nationwide quality register with high coverage. The analysis was restricted to patients who died or responded to the 3-month follow-up questionnaire. Nonresponders were more likely to be younger, have low SES, be smokers, have diabetes, or to have had a previous stroke, but were less likely to be prescribed antihypertensives or to receive reperfusion therapy (eTable 4, links.lww.com/WNL/D210). Selection may have biased the estimated absolute risk of death and dependency but is unlikely to have had a major impact on the main findings.

Although SES lacks a standard classification, it generally incorporates assessments of income, education, and/or occupation. These determinants are correlated but not interchangeable, and each measures different aspects of SES.48 For example, education is often established early in life and is considered a strong determinant of future income and occupation, while economic measures have been found to be more sensitive in detecting associations between SES and health, particularly in the nonelderly individuals.49 Composite measures have the potential to overcome some of the limitations of a single determinant. In this study, we used a composite measure of SES based on attained education and income. The combination of education and income into a composite measure has been shown to produce more comprehensive estimates of social inequalities in health.50

A limitation of our study is that no data on occupation were available, and hence, we could not capture, for example, aspects related to work-based psychosocial processes and environmental exposures.48 However, most of the patients were elderly individuals and likely to be retired, making occupation less important as a determinant of SES in our cohort. Finally, both education and income were obtained by register data and were therefore not subject to recall bias.

We used an approach to mediation analysis, which focuses on the reduction in observed SES disparities that could be accomplished by intervening to equalize the distributions of intermediate variables.14 One strength of this is that we shift the focus from infeasible interventions on SES itself to intervention targets that are more informative from a policy standpoint. In addition, the methods we used allow us to investigate effects of multiple mediators without making strong assumptions about the directions of associations between the mediators.14,15 The method used to estimate the effects relies on the specification of parametric regression models for the outcome and for the mediators. These models are subject to model misspecification bias. We tried to mitigate this issue by making the models as flexible as allowed by the data through the inclusion of interactions and age-squared.

Finally, the study setting was Sweden, a high-income country with publicly financed education and health care systems, and the generalizability of the study findings may be restricted to similar settings.

In our nationwide cohort study using prospectively collected data, we found that low SES was associated with an increased absolute risk of death and ADL dependency 3 months after stroke by 5%–10%, compared with higher SES. If we could intervene to minimize SES differences in comorbidity, stroke severity, and acute care, up to 40 of every 1,000 patients with low SES could potentially be saved from dying or becoming ADL dependent.