Data

We used repeated cross-sectional individual data from the European Union Statistics on Income and Living Conditions (EU-SILC) database.[19] We included all countries with yearly data available from 2005 to 2014 at the moment we started the study (Austria, Belgium, Cyprus, Czech Republic, Denmark, Germany, Greece, Estonia, Spain, Finland, France, Hungary, Ireland, Iceland, Italy, Latvia, Lithuania, Luxembourg, Netherlands, Norway, Poland, Portugal, Sweden, Slovenia, Slovak Republic, and United Kingdom).

We only considered adults aged 25 to 79 years old; we excluded people younger than 25 as most of them might not have concluded their education (our measure of socioeconomic status) (556,938 observations excluded, 12.79% of the sample), and people over 79 because the EU-SILC sample does not include people living in collective households and institutions[19] (215,670 observations excluded, 4.95% of the sample). The inclusion of this latest group could introduce a bias[3] since institutionalized groups tend to have a worse health status than not institutionalized groups[20], and the proportion of institutionalized people varies from country to country[21]. Moreover, EU-SILC codes all respondents over eighty years-old as having 80 or 81 years-old, which would hinder age standardization.[3] Observations with missing data on SRH were also excluded (550,165 observations, i.e., 15.4% of cases). The final sample included 3,030,595 observations. The number of observations per country-year ranged from 2,316 (Iceland, 2006) to 39,843 (Italy, 2005).

‘Poor SRH’ was selected as dependent variable. This variable has been considered as a reliable proxy of health status and mortality[22,23] and has been widely used for the measurement of HI.[3,4,22,24] Five items (‘very good’, ‘good, ‘fair’, ‘bad’ and ‘very bad’) were available to answer the question ‘How is your health in general?’. We grouped the ‘bad’ and ‘very bad’ answers into the ‘poor SRH’ category.

We stratified the population by education level. This is a stable measure of socioeconomic status across life, less subject to reverse causation than other measures (such as income), easily comparable across countries, and, thus, widely used to measure HI.[2–4,8,14,15,25] We classified the seven International Standard Classification of Education categories of the highest level attained into three groups: ‘1’–no education, pre-primary, primary and lower secondary education; ‘2’–upper secondary and post-secondary education; and ‘3’–tertiary education.[3] This simplification allowed the comparison of HI across countries, while ensuring enough observations per education category in each country/year sub-group.

We included sex and age as covariates. Age was included as a continuous variable.

In order to understand how the social welfare context may influence HI, we followed the welfare state typology most frequently used in public health studies,[9–11,26–30] and grouped the 26 countries into six welfare regimes: Scandinavian, Southern, Bismarckian, Anglo-Saxon, Former USSR, and Post-communist. These are described in Table 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Organization and description of European welfare state regimes [9–11,26–31].

https://doi.org/10.1371/journal.pone.0193165.t001

We further considered unemployment rate, the annual percentage growth of the Gross Domestic Product (GDP), and the Gini index (of income inequalities) as indicators of economic cycles. GDP growth measures the national productivity, which may not be sufficient to capture the unequal effects of the Great Recession and the policy responses to it. This is why we completed this indicator by unemployment and income inequality measures, used as complementary means to address the possible effects of the Great Recession. Yearly and per country GDP growth, unemployment rates and Gini index were obtained from the World Bank[32] and Eurostat[33] databases, respectively.

Statistical methods

We first calculated HI among educational groups, stratifying the observations per year (globally), welfare-year, and country-year units (S1 Table). We used absolute (slope index of inequality, SII) and relative measures (relative index of inequality, RII), which allow to account for the distribution of health status across all socioeconomic groups, while taking into account the distribution of the population across socioeconomic status.[24,34,35] SII may be interpreted as the difference in rates between the two extremes of the education hierarchy, and allows observe the impact of a policy in absolute terms, which is more relevant for public health (e.g., number of cases prevented). As SII is sensitive to changes of the prevalence of a certain condition on a population, [34,35] it must be complemented by the RII. That is, when the prevalence lowers, absolute variations become smaller but still relevant, calling for using variation measures in relative terms. The RII can be interpreted as the ratio of the morbidity between the least and most educated groups, and it is relevant to assess the impact of a reform or intervention in low-prevalence contexts. As both measures take into account the distribution of health among intermediate groups and their dimension, their rise can be attributed to a widening of health inequalities across education groups, and/or to a widening of the distribution of the population across education groups[34,35]. Because these two measures behave differently in high and low prevalent contexts, they must be complemented to better express the impact of the macroeconomic and policy changes on HI.[34,35]

Country-year SII and RII were calculated controlling for age, and sex. Personal cross sectional weights, provided by the EU-SILC database,[19] were used to adjust for non-response. Global- and welfare-year SII and RII were further adjusted for country fixed effects, and were plotted in Fig 1 and in S1 Fig.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Evolution of health inequalities between 2005 and 2014.

Absolute (SII) and relative (RII) inequalities on poor SRH plotted by year, globally and by social welfare regime (adjusted for sex, age and country fixed effects).

https://doi.org/10.1371/journal.pone.0193165.g001

Secondly, we created a database compiling the 260 country-year calculated RII and SII. In order to estimate HI time trends, we stratified the countries per welfare, and regressed these country-year RII and SII on a yearly time trend, adjusting for country fixed effects (Model 1, below). Finally, to test whether variations in HI were explained by economic cycles, we additionally adjusted for GDP growth, unemployment, Gini index, and all three proxies together (Models 2, 3, 4, and 5, respectively).

1. Model 1: HIct = α + Trendt + Countryc + ε
2. Model 2: HIct = α + Trendt + Countryc + GDPct + ε
3. Model 3: HIct = α + Trendt + Countryc + UNEMPct + ε
4. Model 4: HIct = α + Trendt + Countryc + GINIct + ε
5. Model 5: HIct = α + Trendt + Countryc + GDPct + UNEMPct + GINIct + ε

where ‘HI’ is either RII or SII, that is, these five equations were estimated for both SII and RII indicators. The indices ‘t’ and ‘c’ refer to the year and country, respectively. The ‘Country’ variable stands for country fixed effects, and the ‘Trend’ for the yearly linear trend. These models were estimated globally and then by the six welfare regimes, so we finally ran 5x2x(1+6) = 70 regressions.

Key findings

Our analysis shows that poor SRH decreased in Europe between 2005 and 2014, but its distribution remained concentrated among the least educated groups. Over the period, the highest RII values were mostly found in Post-Communist and Southern countries, surpassed in 2006, 2009, and 2012 by the Anglosaxon countries. As for SII, Post-Communist and Former-USSR countries showed the largest rate differences between least and most educated groups.

Over the 10-year period and for all countries taken together, absolute inequalities persisted and relative inequalities slightly widened. However, there were substantial differences across welfare regimes in the evolution of HI. Anglo-Saxon countries experienced the largest increase in absolute inequalities, followed by Bismarckian countries, while a statistically significant reduction was observed in Post-Communist countries. Post-Communist countries also experienced significant changes in relative inequalities, with a sharp widening trend across this period. In contrast, relative inequalities were stable in all remaining welfare regimes.

Finally, variations in GDP and unemployment did not explain any of the time trends in inequalities, and slightly obscured the rising trend in relative inequalities in Bismarckian countries. Variations on the Gini index turned the rising trend of absolute inequalities in Anglo-Saxon countries non-significant.

Interpretation

The wide HI in this large population sample of 26 European countries is consistent with the evidence collected over time and locations.[3,4,8] The differences recorded across welfare regimes are partially in line with the results by Eikemo et al,[8] where Post-Communist, Former-USSR and Southern countries were among the welfare regimes with higher SII values, and Southern countries exhibited higher RII.

The apparent paradox of large absolute inequalities coupled with smaller relative inequalities in Post-communist and Former-USSR countries may be explained by the higher prevalence of poor SRH in these regions: the higher the prevalence of a certain condition, the higher the absolute differences between socioeconomic groups, and the lower the relative ones[3]. The different magnitudes of absolute inequalities between welfare regimes may be explained by their socioeconomic characteristics, as Former-USSR and Post-Communist countries have traditionally lower social protection expenditure as a percentage of GDP.[36,37]

The global trends of HI are in line with earlier studies by Kunst et al[4] and Hu et al,[3] confirming the persistence of HI over the last decades. However, the different patterns on the evolution across welfare regimes deserves further reflection. On the one hand, only Post-Communist countries presented a rise in relative inequalities, which confirms the trends previously described for Czech Republic, Poland, and Hungary.[3,15] On the other hand, the significant but small increases in absolute inequalities found in Bismarckian and Anglo-Saxon countries confirm some of the (non-significant) data from Hu et al.[3] Unexpectedly, Post-Communist countries showed a small narrowing trend of their absolute inequalities, which has not been previously noted in the literature. The remaining welfare regimes did not experience any significant change in the pattern of RII or SII, which goes in line with recent findings about the evolution of inequalities in SRH.[3,24]

The highest increase in absolute inequalities among Anglo-Saxon countries may be interpreted as the result of a more liberal welfare regime, and also of the response to the Great Recession. The substantial cuts to social protection, public health, and education services,[38,39] linked with the raise in income inequality may explain the rising absolute inequalities in Anglo-Saxon countries, during this period. In some Bismarckian countries, the recession followed by changes in labour market policies[24] and reduction of health budgets,[38,39] that may have led to the slight increase in absolute inequalities. These results seem to show that to prevent HI increasing during times of economic downturn, there is a minimum level to which the social safety net cannot go below.[13] Differently, in Post-Communist countries the prevalence of poor SRH decreased at a faster pace in the least educated group, leading to a reduction in absolute inequalities and an increase in relative ones. During this period, Post-Communist countries increased public spending on health, fostered government budget transfers to the health insurance scheme, and strengthened primary care and health promotion and prevention,[38,39] possibly with a higher effect on least educated groups. In comparison, Scandinavian and Former-USSR countries maintained a stable public health budget, despite the recession,[39] leading to stable HI. The stable trend in Former-USSR may also be linked to the fact that in these countries the public social spending is historically low and did not expressively change in the last decade.[40]

Surprisingly, and contrasting with previous studies[41], Southern countries, highly affected by substantial plunges in GDP, rises in unemployment rates, and under pressure to implement austerity measures, including sweeping public sector budget cuts, experienced stable trends in HI. This may be explained, firstly, by the characteristics of this welfare regime, where social support strongly relies on the family and charitable sector.[9] Secondly, budget cuts in healthcare mostly focused on renegotiations of drug expenditure[18] and corrections of longstanding sources of inequity in financing,[38] which may have actually contributed to reducing HI, instead of increasing them. Thirdly, in these countries the budget crisis was so extensive that it may have affected the population as a whole, and not only the least educated. Fourthly, unemployment particularly affected the youngest age groups, who are also the most resilient in terms of experiencing short term ill-health effects. In the same line, Spanish data has shown that good SRH has been improving at a faster pace among unemployed groups, and that unemployment during the economic crisis did not worsen SRH more than before it. [41,42]

Finally, we originally posited that trends in HI could be explained by economic cycles, proxied by variations in GDP, unemployment, and income inequality. However, our findings only confirmed this hypothesis amongst Anglo-Saxon countries, where rising absolute inequalities were explained by the growth in income inequality. Long-lasting and recently implemented counter-cyclical policies in some countries may have attenuated the effect of recession on the most vulnerable people.[18,38,39] It may well be also that GDP and unemployment rates, being only available at the national level, may have failed to capture geographical variations within countries, which may be more relevant to understand how people are affected by economic cycles.[43]

Strengths and limitations

The analysis of the EU-SILC database allowed a comprehensive and recent overview of the evolution of HI. We studied the trends of HI among 26 European countries, analysing more than 3 million cross-sectional observations collected from 2005 to 2014, using socio-political and economic lenses. Also, this is one of the few studies that used comparable international data, collected using comparable methodology in all countries, over a 10-year period, and additionally adjusted using weighting methods to accurately reflect the population of each country.

Yet, some limitations must be considered. First, only countries with data from 2005 to 2014 were included, leaving Bulgaria, Croatia, Macedonia, Malta, Romania, Switzerland, Serbia and Turkey out of the analysis. Such limitation does not seem to have constrained the applicability of the results of this study, considering the size and variety of our sample. Second, we used as outcome a subjective measure of health. However, as referred before, SRH reflects both mental and physical health,[44] providing a more comprehensive and immediate perspective of population health than diagnosed conditions or mortality, and is widely considered a good predictor of health status and mortality.[22,44] In addition, this study focused on the time trend analysis of HI in SRH, not just on a one-off observation, which decreases the risk of bias due to cultural differences in health perception. Finally, in order to get a broad perspective on the evolution of HI in all these 26 European countries, we grouped them into six welfare regimes. Though the value of the ‘welfare state regimes approach’ is sometimes questioned[30,45], the typology grouping allows a theoretically-based analysis of the influence of socio-political and cultural characteristics of these countries on health outcomes, from the older theories of Esping-Andersen[46] and Ferrera,[27] to the more recent works of Bambra,[28] Aidukaite,[29] and Campos-Matos.[9] Moreover, welfare regimes have been considered important policy contexts in the aetiology of health inequalities[47] that should be taken into account in the study of the impact of debt crisis[5,11,48,49] and cannot be excluded from the explanation of the evolution of HI.[8,11,48,50]