Income and education as predictors of return to working life among younger stroke patients

The sample

In the study 7,081 inpatients (out of 11,864 registered) aged 40-59, who had survived a first stroke and had been discharged from a hospital in the period 1996-2000 in Sweden were analysed. These persons had not previously suffered from ischemic heart disease and had a paid work before the stroke. Linking inpatient registers to census data gave information about these individuals' level of income and their educational attainment.

The inclusion/exclusion criteria

All major stroke subtypes were included, i.e., subarachnoid haemorrhage, cerebral infarction, intracerebral haemorrhage and 'stroke, not specified' (ICD 10 = I60, I61, I63 and I64). Due to less reliable data TIA (transient cerebral ischaemic attack, ICD 10 = G45.8 and G45.9) was excluded. To avoid possible confounding from comorbidity, people suffering from ischemic heart disease (ICD 10 = I20-I25) prior to the first stroke were excluded from the initial sampling.

To exclude student groups and young parents on leave after childbirth the lower age limit was set at age 40 (note that stroke incidence is also low before that age). The upper age limit of 59 was motivated by the fact that 65 is the official retirement age in Sweden, and the follow-up required a margin of some years before retirement.

Since we wanted to study return to work, only those with paid work before the stroke were included. Consequently, people who had only received some form of social security or unemployment benefit before the stroke were excluded. The inclusion criteria were based on registered income from work. The limit was set to an income from work amounting to at least € 6,600 in the year prior to the stroke. This sum represents an income comparable to approximately 25 percent of the average pay for a Swedish full-time worker in 2000. Twenty-five percent is the lowest amount of work time required to qualify for sick leave benefits in the Swedish social insurance system.

The selection frame

The selected population consisted of all first time stroke patients in Sweden discharged from a hospital in the period 1996-2000 aged 40-59. The stroke patients were identified from the population register of in-patient care at the Swedish National Board of Health and Welfare and collated with a population-based data register from Statistics Sweden (SCB). The number of missing observations in the sample and the reasons to why some observations had to be disregarded is shown in the flowchart (Figure 1).

The measures

The dependent variable, return to work, was measured in terms of patients' income from work after discharge from hospital. The lowest amount of income accepted was the same as the lowest amount accepted prior to the stroke (approximately € 6,600). Return to work implies returning to any paid job and represents a documented indicator of work capability. It was not possible to determine whether the return was to the original job or if a change of job had taken place due to the stroke

In contrast to the selection criteria for the dependent variable, the independent variable "level of income" includes all types of income - not only income from work. Level of income was divided into equally sized quartiles. Education was divided into three levels: nine-year compulsory school, upper secondary school and university studies. Since the 1950's, compulsory education in the Swedish school system has been of nine years' duration (age 7-16). Over the years, an increasing number of youths have continued their studies at upper secondary school, which means that in general, younger stroke patients have more years of education. University studies was used as the highest educational category, including those who had completed at least one semester of full-time studies; who were included since they shared more similarities with people who had longer university education in terms of return to work and mortality than those who had only completed upper secondary school.

Consequently used variables for models adjustment were age, sex, stroke subtype and length of in-patient care.

Statistical analysis

Data were analysed with discrete time hazard regressions with time-varying baseline intensity, controlling for sex, age, age-square, days of in-patient care, days-square, and stroke subtype. The variables age and days of in-patient care were given both linear and quadratic representations to capture the curvilinear relationships.

The results are presented as relative risks (RRs). The follow-up period was four calendar years after discharge from hospital. Note that since our income data are based on calendar years, the 'first' year also includes the remaining months of the year of discharge. If stroke discharge is random over the year, this means that the average first year is 18 months (years two, three and four are each twelve months). Persons who died were assumed to die in the middle of that calendar year and be censored after half of the year. A time-varying baseline intensity was used to account for variations in return to work between the years and for the fact that the first 'year' is longer than the other years. The baseline intensity is assumed to be constant within each year [27]. The discrete time hazard regressions were performed with the STATA command STPIECE.

Descriptive statistics

A majority of 69 percent returned to work. Table 1 shows the descriptives. For example, a majority of patients were men (n = 4548, total n = 7081), and the proportion of men returning to work (71.3 percent) during the whole follow-up period of four years was higher than among women (64.2 percent). Educational and income differences are also indicated by the table. The proportion returning to work in different age groups decreases somewhat with age. Patients with a short period of in-patient care returned to work more often than those with longer periods. Cerebral infarction was the most common stroke diagnosis, but the proportion of this stroke subtype that returned to work did not differ from the average.

Patients with intracerebral haemorrhage had the longest periods of in- patient care (not shown) and returned to work less frequently (Table 1). Those with subarachnoid haemorrhage also had longer periods of in-patient care than the average (not shown) but returned to work more often (Table 1). Persons with subarachnoid haemorrhage were younger on average (51 years vs. 53 for the other stroke groups) and larger proportions were women (56 percent compared to 33 percent for other of stroke subtypes).

Average follow-up time until death, return to work or censoring was 2.4 years. As is shown in Table 1, 399 persons (5.6%) died without returning to work. Another 106 persons (1.5%) returned to work but died during the follow-up period. The average time before death in these cases was 2.5 years.

Regression analyses

The results of the hazard regressions in Table 2 shows that people with higher education returned to work significantly more often than those with compulsory education only, and people with higher income returned to work significantly more often than people with low income. Individuals with university education were about 30 percent more likely (RR = 1.33 in Model 1) to return to work than those with compulsory education only and those in the highest income quartile about twice as likely as those in the lowest income quartile (RR = 2.02 in Model 2). When education and income were included in the same model, i.e., controlling for each other, the relative risks for education became attenuated, while the relative risks for income were about the same.

Women had a significantly lower relative risk of return to work in the age-adjusted model and in Model 1, when controlling for education, a difference which became small and insignificant when additionally controlling for income in Model 3. This suggests that men and women have about the same probability of return to work, when the gender differences in income level have been accounted for.

Results also show that, controlling for stroke subtype, there are strong associations between length of hospital care and the probability for return to work; the shorter the period of hospital care, the better the prognosis for return to work.

Those with subarachnoid haemorrhage did not differ from the other stroke subtypes (age-adjusted analyses in Table 2) with controls for age. But with additional variables included (i.e., sex, length of in-patient care, and education) those with subarachnoid haemorrhage had significantly higher probability of return to work than on average.

Table 3 shows that the association between socioeconomic position and return to work was similar for the different stroke subtypes. There were no significant differences between the different stroke subtypes (modelling not shown).

Limitations

The generalisability of results to other times and countries is unclear. Since we arrived at our contribution by using population-based data, we should at least be able to give strong evidence from Sweden. Nevertheless, even for Sweden, recent policy reforms may affect the generalisability of the results over time. There has been an almost aggressive return-to-work policy in Sweden with activation measures and benefit reductions during recent years. This could possibly change the association between stroke and return to work. It is also possible that such changes do affect groups with different education and income differently.

This study is based on registers generally considered to be of high quality that provide the basic information for Swedish health care development and comparisons (see e.g. [41]). Very few cases on income and education are missing in our study. Statistics Sweden also takes active part in the methodological discussion on usage of administrative registers (see e.g [42]). Our study is also in line with the recommendations of a recent review of stroke outcomes suggesting that return-to-work studies should be based on population data and the measurement of paid work [6]. Nevertheless, there are some important limitations in our study design since we do not know the clinical condition of the stroke-affected persons. We avoided using TIA diagnosis due to less reliable data and we excluded those with prior ischemic heart disease, but there might be other co-morbidities or confounding factors due to unmeasured variables such as health behaviour and quality of care. We do not know the type of work returned to, or whether the subjects remained in the labour force after the initial phase of return. Income from work was used as a proxy for work capacity and we only studied stroke survivors who had an income from work prior to the stroke. We tried modelling other income levels (≈ €1100 and ≈ €11000). The proportion of people returning was dependent on the income level chosen, but the socioeconomic pattern remained the same (results not shown). There is also a risk of overestimating return to work in high income groups (and underestimating those in the lowest income group) because our income level was based on the average Swedish income, corresponding to 25 percent of full time work. Someone in the highest income group may work less than 25 percent, but still have an income from work equivalent to the average Swedish income.

By using population data we tried to avoid a selection bias. However, there is a potential bias from fatal events. We have no information about people who did not visit a doctor or died before reaching hospital care. The socioeconomic mortality pattern after a stroke, both before and after discharge from hospital, corresponds to the pattern of not returning to work (based on preliminary analyses from the same data, not shown). Persons with low socioeconomic position are both more likely to die and less likely to return to work. As a consequence, our results could have underestimated the association between socioeconomic position and return to work.

Since the study is based on register data, no information is available on lifestyle factors. We used all relevant covariates available. While this limitation may have resulted in omitted variable bias/unobserved heterogeneity, it was the best possible modelling given the available data.

While the highest income quartile had twice the odds of returning to work compared with the lowest quartile, those with a university education were only 13 percent more likely to return to work than those with an elementary education. This may indicate that economic resources are more decisive than educational ones (e.g. how well-informed someone is), but whether this is the case remains unclear. In the future, it would be interesting to examine more closely the determinants captured by different socioeconomic indicators.