European Journal of Epidemiology (2019) 34:1131–1142 https://doi.org/10.1007/s10654-019-00580-9 MORTALITY Progress against inequalities in mortality: register‑based study of 15 European countries between 1990 and 2015 Johan P. Mackenbach1 · José Rubio Valverde1 · Matthias Bopp2 · Henrik Brønnum‑Hansen3 · Giuseppe Costa4 · Patrick Deboosere5 · Ramune Kalediene6 · Katalin Kovács7 · Mall Leinsalu8,9 · Pekka Martikainen10 · Gwenn Menvielle11 · Maica Rodriguez‑Sanz12,13 · Wilma J. Nusselder1 Received: 8 May 2019 / Accepted: 5 November 2019 / Published online: 15 November 2019 © The Author(s) 2019 Abstract Socioeconomic inequalities in mortality are a challenge for public health around the world, but appear to be resistant to policy-making. We aimed to identify European countries which have been more successful than others in narrowing inequali- ties in mortality, and the factors associated with narrowing inequalities. We collected and harmonised mortality data by educational level in 15 European countries over the last 25 years, and quantified changes in inequalities in mortality using a range of measures capturing different perspectives on inequality (e.g., ‘relative’ and ‘absolute’ inequalities, inequalities in ‘attainment’ and ‘shortfall’). We determined which causes of death contributed to narrowing of inequalities, and conducted country- and period-fixed effects analyses to assess which country-level factors were associated with narrowing of inequalities in mortality. Mortality among the low educated has declined rapidly in all European countries, and a narrowing of absolute, but not relative inequalities was seen in many countries. Best performers were Austria, Italy (Turin) and Switzerland among men, and Spain (Barcelona), England and Wales, and Austria among women. Ischemic heart disease, smoking-related causes (men) and amenable causes often contributed to narrowing inequalities. Trends in income inequality, level of democracy and smoking were associated with widening inequalities, but rising health care expenditure was associated with narrowing inequalities. Trends in inequalities in mortality have not been as unfavourable as often claimed. Our results suggest that health care expansion has counteracted the inequalities widening effect of other influences. Keywords Mortality · Social inequality · Trends · Europe Electronic supplementary material The online version of this article (https​://doi.org/10.1007/s1065​4-019-00580​-9) contains supplementary material, which is available to authorized users. 7 * Johan P. Mackenbach Demographic Research Institute, Budapest, Hungary j.mackenbach@erasmusmc.nl 8 Stockholm Centre for Health and Social Change, Södertörn 1 University, Stockholm, Sweden Department of Public Health, Erasmus MC, P.O. Box 2040, 9 3000 CA Rotterdam, The Netherlands Department of Epidemiology and Biostatistics, National 2 Institute for Health Development, Tallinn, Estonia Epidemiology, Biostatistics and Prevention Institute, 10 University of Zürich, Zurich, Switzerland Department of Sociology, University of Helsinki, Helsinki, 3 Finland Institute of Public Health, Copenhagen University, 11 Copenhagen, Denmark INSERM, Sorbonne Universités, Institut Pierre Louis 4 d’Epidémiologie et de Santé Publique (IPLESP UMRS Department of Clinical Medicine and Biology, University 1136), Paris, France of Turin, Turin, Italy 12 5 Agència de Salut Pública de Barcelona, Barcelona, Spain Department of Sociology, Vrije Universiteit Brussel, 13 Brussels, Belgium CIBER de Epidemiología y Salud Pública (CIBERESP), 6 Madrid, Spain Lithuanian University of Health Sciences, Kaunas, Lithuania 13 Vol.:(0123456789) 1132 J. P. Mackenbach et al. Introduction Socioeconomic position was indicated by highest level of completed education, in three categories: ‘low’, ‘mid’, Inequalities in mortality by socioeconomic position are and ‘high’ (corresponding to ISCED 1997 categories 0–2, among the most consistently reproduced findings in public 3–4 and 5–6, respectively). We focused on educational health research: rates of mortality are higher among those inequalities (instead of, e.g., occupational inequalities) with a lower education, occupational class or income in because comparable data were available for a wider range all countries that have taken the trouble to collect the nec- of countries, and because education is normally com- essary data [1–3]. The ubiquity and persistence of health pleted in early adulthood, which avoids most problems of inequalities, despite the rise of the welfare state and some reverse causation [6]. The analyses were restricted to ages countries’ explicit attempts to narrow them, even give the 35–79 years because at advanced ages the proportion of impression that health inequalities are resistant to policy- high educated becomes very small, and education becomes making [4]. a less sensitive indicator of socioeconomic position. However, with data on more countries and longer For our analysis of possible determinants of trends in periods of time becoming available, it is increasingly mortality inequalities we extracted four national-level vari- becoming clear that these inequalities are more variable ables from harmonised international databanks, each of and dynamic than is often appreciated. Within Europe, which has been shown to be associated with mortality in inequalities in mortality are smaller in the Mediterranean cross-national analyses in Europe: national income (natu- region, and larger in the former East bloc, with Northern ral logarithm of Gross Domestic Product (GDP) per capita, Europe also having surprisingly large inequalities in mor- measured in purchasing power parity adjusted US$; source: tality, particularly among women [1]. In many countries, World Health Organization Health for All Database; higher relative inequalities in mortality (as indicated by, e.g., income is associated with lower mortality [7]), income ine- the rate ratio of mortality among lower as compared to quality (Gini index of net equivalised household income; higher socioeconomic groups) have risen strongly, whereas source: Standardizing the World Income Inequality Data- absolute inequalities (as indicated by, e.g., the difference base; higher income inequality is associated with higher between the mortality rates of lower and higher socioeco- mortality [8]), level of democracy (combined Freedom nomic groups) have followed a more variable course [5]. House—Polity2 democracy score; source: Quality of Gov- These variations between countries and over time are ernment Dataset; transition to democracy is associated with important because they show that the magnitude of ine- a temporary rise in mortality [9]), and health care expendi- qualities in mortality is far from fixed, and that there is, in ture (current expenditure on health as a percentage of GDP; principle, large scope for reducing health inequalities. In source: World Health Organization Health for All Database; this paper, we aim to identify the European countries which higher health care expenditure is associated with lower mor- have been more successful than others in containing or even tality [10]). reducing inequalities in mortality, and we exploit variations We also collected and harmonized data on two determi- between countries to identify the factors associated with nar- nants of mortality that were available by level of education: rowing and widening of inequalities in mortality. smoking (current cigarette smoking; source: national health interview surveys; higher prevalence of smoking is associ- ated with higher mortality [11]) and material deprivation (inability to pay for at least three out of nine items deemed Methods necessary for a normal life; source: EU Statistics on Income and Living Conditions survey; higher prevalence of material Data deprivation is associated with higher mortality [11]). We collected and harmonized register-based mortality data Analysis from 15 European countries covering all parts of the sub- continent. Most data cover complete national populations Mortality rates by educational level were directly age-stand- with the exceptions of England and Wales and France (1% ardized using the European Standard Population [12] and representative samples) and Spain and Italy (Barcelona and five-year age-groups. We also calculated partial life expec- Turin only). Most data stem from a longitudinal mortality tancies between the ages of 35 and 80 (years of life lived, as follow-up after a census. An overview of the mortality data a measure of ‘attainment’) and differences between partial sources and some key characteristics of the data are given life expectancies and the theoretical maximum of 45 years in Web appendix Table A1. Web appendix Table A2 gives (years of life lost, as a measure of ‘shortfall’) [13]. details on the causes of death used in the analyses. With these outcomes we calculated a total of eight measures of inequalities in mortality, capturing different perspectives on 13 Progress against inequalities in mortality: register-based study of 15 European countries… 1133 inequality: Rate Differences and Rate Ratios of mortality rates, Results Slope Indices of Inequality and Relative Indices of Inequality (which, in contrast to simple Rate Ratios and Rate Differences, Trends in mortality have generally been favourable, both for make an adjustment for shifts in the educational distribution the low and the high educated, and regardless of whether [14, 15]), Differences and Ratios of years of life lived, and we study mortality rates, years of life lived or years of life Differences and Ratios of years of life lost. We refrained from lost. We illustrate this in Fig. 1 with data on years of life adding even more measures, such as the recently developed lost between the ages of 35 and 80. In Northern, Western Slope Index of Inequality for years of life lost [16]. Please and Southern Europe, these shortfalls in life expectancy note that absolute differences of years of life lived and years have steadily declined over this time-period, both among of life lost are numerically identical. We used a wide range of the low and high educated, but in Eastern Europe recent measures of inequalities in mortality to counter recent criti- improvements represent a reversal as compared to unfavour- cism against reliance on a more restricted set of measures that able trends up to around the year 2000 (full details in Web is biased towards finding widening inequalities [13]. appendix Table A3). In order to analyse the relationship between determinants Mortality has always been higher, and life expectancy and inequalities in mortality we conducted panel (or pooled lower, among the low than the high educated, but due to cross-sectional time series) analyses in which mortality was differences in the rate of improvement between low and modelled as a function of national-level determinants, using high educated, gaps have sometimes narrowed, sometimes country- and period-fixed effects, in the following way: widened over time. We illustrate this in Fig. 2 with Slope and Relative Indices of Mortality. Among men, the Slope Mect = α + β1 ∗ C + β2 ∗ P + β3 ∗ Dct + β4 ∗ E Index of Inequality (SII), which measures absolute differ- + β5 ∗ Dct ∗ E + u0ct + εect (1) ences in mortality rates between education groups taking in which M = age-standardized mortality rate, C = a set of into account their relative sizes, has mostly gone down, dummy variables representing country fixed effects, P = a whereas the Relative Index of Inequality (RII) has mostly set of dummy variables representing period fixed effects, gone up. Among women, the trends are more variable, but D = determinant, E = level of education, α = intercept, β1 and here too the trends in absolute inequalities (SII) are much β2 = parameters indicating differences in levels of M between more favourable than trends in relative inequalities (RII). countries and periods, β 3 = parameter indicating main Whether a narrowing or widening has occurred is strongly effect of determinant on M, β4 = parameter indicating main dependent on the measure used. Table 1 presents the changes effect of education on M, β5 = parameter indicating effect in all eight measures of inequality, and while it confirms of interaction between determinant and education on M, that trends in absolute inequalities have usually been more u = country- and period-specific random residual, ε = edu- favourable than trends in relative inequalities, it also shows cation-specific random residual, e = subscript denoting level that the reverse is the case for years of life lived, for which of education, c = subscript denoting country, t = subscript relative differences sometimes narrowed even though abso- denoting calendar-year. lute differences widened. For the determinants that were available by level of edu- However, ranking of countries in terms of more or less cation, i.e., smoking and material deprivation, we conducted narrowing of inequalities in mortality is hardly depend- similar analyses, in the following way: ent on the measure used. This ranking is presented in Web appendix Table A4 and the last column of Table 1. Because Mect = α + β1 ∗ C + β2 ∗ P + β3 ∗ Dect + β4 ∗ E + u0ct + εect mortality in Eastern Europe started to decline only recently, (2) these countries have been ranked separately. All regression analyses were done with the SPSS 24 routine Combining all eight measures of inequalities in mortal- for linear mixed models. This allowed us to apply a multilevel ity, Austria, Italy (Turin) and Switzerland have had the most framework and thus to take into account dependency of data favourable trends among men, whereas Spain (Barcelona), within countries and time-periods. We used an autoregressive England and Wales, and Austria have had the most favour- model of order 1 to take into account the year-on-year serial able trends among women. On the other end of the scale, autocorrelation in the mortality rates. Finland, Norway, and Denmark have seen the least narrow- ing of inequalities in mortality. In Eastern Europe, Hungary Role of the funding source (men) and Estonia (women) performed best, and Lithuania performed worst. The study sponsor had no role in study design, collection, What drives differences between low and high educated analysis, and interpretation of data, writing of the report, or in the pace of mortality decline? As a first step, we have in the decision to submit the paper for publication. identified the causes of death for which mortality decline 13 1134 J. P. Mackenbach et al. Fig. 1  Trends in years of life lost between the ages of 35 and and 80 years) minus 45 years. Dashed lines: Eastern European coun- 80 years, by country, gender and level of education. Notes Years of tries; dotted lines: Northern European countries. Scale for men and life lost calculated as partial life expectancy (between the ages of 35 women different has been faster or slower among the low than among the Differences between countries in the extent of narrow- high educated (Fig. 3). The 13 causes of death included in ing of inequalities in total mortality are partly dependent on this analysis present a highly variable picture. For example, whether or not there are causes of death for which declines absolute declines in mortality from ischemic heart disease in mortality have been larger among the high than the low and amenable causes were mostly faster among the low educated, instead of vice versa. For example, Austria’s very than among the high educated (shown as negative values in good performance is due to almost uniformly larger declines Fig. 3), whereas declines in mortality from breast cancer and in mortality among the low educated. At the other end of the other cancers have mostly been faster among the high than spectrum, Denmark’s bad performance is due to the fact that among the low educated (shown as positive values in Fig. 3). for many causes of death declines in mortality have been 13 Progress against inequalities in mortality: register-based study of 15 European countries… 1135 Fig. 1  (continued) larger among the high than the low educated. The excellent Nordic countries, however, is not only due to more unfa- performance of Hungary (men) is based on massive reduc- vourable trends among the low educated in smoking-related tions among the low as compared to the high educated in mortality, but also to more unfavourable trends for other mortality from, among others, amenable causes and alcohol- causes, including alcohol-related mortality and (in the case related causes. of Denmark) amenable causes. As we have seen, trends in inequalities in mortality are As a second step, Table 2 presents the results of regres- less favourable among women than among men. Figure 3 sion analyses in which we analysed which determinants shows that this is partly due to the fact that declines in mor- of mortality were associated with a widening or narrow- tality from smoking-related causes are mostly larger among ing of absolute inequalities in mortality. Trends in national the high than the low educated among women, whereas income, income inequality, levels of democracy and health the reverse is true for men. The worse performance of the care expenditure have, on average, all been upwards, but 13 1136 J. P. Mackenbach et al. Fig. 2  Changes in Slope Index of Inequality and in Relative Index of Inequality for mortality, by country and gender. Notes: Start = around 1990; end = around 2012. *Start = around 2000; end = around 2012. For exact timing of observation periods, see Table 1 and Web appendix Table A1 13 Progress against inequalities in mortality: register-based study of 15 European countries… 1137 Fig. 2  (continued) their associations with mortality and mortality inequalities This is not only seen in a simple correlation diagram as in have been different. Rising national income was associated Fig. 4, but also in country- and period-fixed effects analy- with declining mortality, but in absolute terms the relation- ses (Table 2). Web Web appendix Table A6 shows that this ship was not statistically significantly different between low effect of health care expenditure is seen for many causes of and high educated, implying that absolute inequalities in death, but particularly for mortality from amenable causes mortality are not affected by rising national income. Rising and cardiovascular diseases. income inequality was associated with rising mortality, par- ticularly among men, and this relationship was stronger for the low educated, suggesting that rising income inequality Discussion may have contributed to widening inequalities in mortality. Rising levels of democracy were associated with rising mor- Strengths and limitations tality among the low educated, again suggesting a mortality widening effect. By contrast, rising health care expenditure This study provides a robust picture of trends in mortality was associated with stronger mortality decline among the inequalities in a large number of European countries and low educated, implying that this may have contributed to a over a long period of time. Combining different measures of narrowing of inequalities in mortality. inequalities made us less dependent on the arbitrary choice As Table 2 shows, trends in smoking have also differently of a more limited set of measures as used in previous stud- impacted mortality among the low and the high educated. ies. It has recently been argued that when studying trends in Both smoking and material deprivation are strongly associ- health inequalities measures of ‘attainment’ should receive ated with mortality, but only the gap in smoking has widened equal attention as measures of ‘shortfall’, which are more over time, implying that trends in smoking have probably commonly used and tend to show widening inequalities [13]. contributed to widening inequalities in mortality. Because ‘attainments’ and ‘shortfalls’ of mortality are more These results show that the narrowing of absolute ine- readily conceptualized in terms of years of life lived and qualities in mortality that has occurred in at least some coun- lost, we have studied these together with mortality rates. The tries, particularly among men, has occurred despite several relationship between life expectancy and age-standardized unfavourable trends, including rising income inequality, mortality rates is mathematically complex, and includes a transitions towards democracy, and widening inequalities difference in the weight given to different age-groups [17, in smoking. The only factor that may have contributed to a 18]. narrowing of inequalities in mortality has been rising health Our study shows that trends in ratios of years of life lived care expenditure. This is illustrated in Fig. 4, with years of are much more favourable than trends in ratios of years lost life lost as the measure of mortality. (Table 1). This is unsurprising in view of the fact that the At each level of health care expenditure the low edu- number of years lived between ages 35 and 80 is much larger cated lose more years than the high educated, but the gap is than the number of years lost, but our results nevertheless considerably smaller at higher levels of health expenditure. illustrate that whether inequalities in mortality narrow or 13 1138 J. P. Mackenbach et al. Table 1  Annual changes in eight quantitative measures for the magnitude of inequalities in mortality A. Men Period covered Inequality measure Average Mortality Life expectancy rank Rate Rate ratio (%) Slope index of Relative index Difference Ratio years of Difference Ratio years of difference inequality (%) of inequality years of life life lived (%) years of life life lost (%) (%) (%) lived (%) lost (%) North Finland ca. 1993–2012 – 0.29 3.60 – 0.72 1.67 0.58 0.03 0.58 1.72 8.5 Sweden ca. 1992–2007 – 0.50 1.99 – 0.58 1.53 – 0.06 – 0.02 – 0.06 0.93 5.5 Norway ca. 1988–2008 – 0.14 3.85 – 0.10 6.81 0.47 0.02 0.47 1.70 9.3 Denmark ca. 1997–2012 0.42 5.08 0.25 5.01 1.21 0.08 1.21 2.40 10.9 West England & W ca. 1993–2012 – 1.62 1.72 – 1.62 1.87 – 0.76 – 0.06 – 0.76 0.82 3.6 Belgium ca. 1994–2009 – 1.00 1.46 – 1.18 0.90 0.02 – 0.01 0.02 0.81 4.3 Austria ca. 1991–2012 – 1.53 0.14 – 0.92 3.12 – 1.20 – 0.17 – 1.20 0.13 2.6 Switzerland ca. 1993–2012 – 1.39 1.88 – 1.66 2.28 – 1.16 – 0.15 – 1.16 0.91 3.5 France ca. 1988–2005 – 0.43 2.49 – 0.60 1.61 0.48 0.02 0.48 1.43 7.5 South Spain (B) ca. 1994–2012 – 0.26 3.34 – 0.41 2.53 0.17 – 0.01 0.17 1.39 7.8 Italy (T) ca. 1989–2012 – 1.51 1.24 – 1.51 1.23 – 0.96 – 0.07 – 0.96 0.47 2.6 East Hungary ca. 2000–2011 – 2.14 – 0.48 – 2.25 – 0.22 – 1.86 – 0.52 – 1.86 – 0.12 1.4 Poland ca. 2002–2011 – 1.22 0.40 – 1.39 – 0.63 – 0.53 – 0.14 – 0.53 0.62 2.5 Lithuania ca. 2003–2013 0.05 2.74 – 0.39 1.87 0.25 – 0.05 0.25 1.76 4.0 Estonia ca. 2000–2013 – 1.87 1.76 – 2.18 3.01 – 1.73 – 0.72 – 1.73 0.90 2.1 B. Women Period covered Inequality measure Average rank Mortality Life expectancy Rate Rate ratio (%) Slope index of Relative index Difference Ratio years of Difference Ratio years of difference inequality (%) of inequality years of life life lived (%) years of life life lost (%) (%) (%) lived (%) lost (%) North Finland ca. 1993–2012 1.94 6.75 0.56 4.44 3.52 0.12 3.52 2.72 10.6 Sweden ca. 1992–2007 0.51 1.93 0.23 1.79 1.02 0.03 1.02 1.04 6.0 Norway ca. 1988–2008 1.84 5.78 1.64 7.41 2.41 0.08 2.41 2.05 9.9 Denmark ca. 1997–2012 1.53 6.72 0.50 6.02 2.02 0.09 2.02 2.57 9.5 West England & W ca. 1993–2012 – 1.21 1.08 – 1.21 1.75 – 1.03 – 0.04 – 1.03 0.24 2.5 Belgium ca. 1994–2009 0.56 3.05 – 0.23 1.49 1.95 0.05 1.95 1.16 7.3 Austria ca. 1991–2012 – 1.21 0.65 – 1.01 1.02 – 0.16 – 0.01 – 0.16 0.65 2.9 Switzerland ca. 1993–2012 – 0.30 2.16 – 0.90 1.34 0.91 0.02 0.91 1.09 5.0 France ca. 1988–2005 – 0.03 2.00 – 0.20 2.01 1.59 0.05 1.59 1.38 7.0 South Spain (B) ca. 1994–2012 – 1.83 – 0.81 – 2.31 – 1.93 0.10 0.00 0.10 0.43 1.9 Italy (T) ca. 1989–2012 – 1.39 0.51 – 1.96 – 0.89 1.49 0.02 1.49 0.66 3.5 East Hungary ca. 2000–2011 – 0.31 2.94 – 1.48 0.43 – 0.42 – 0.05 – 0.42 1.00 2.3 Poland ca. 2002–2011 – 0.70 1.17 – 0.64 1.69 – 0.27 – 0.03 – 0.27 0.60 2.3 Lithuania ca. 2003–2013 0.88 3.96 0.00 4.78 0.49 0.04 0.49 1.87 4.0 Estonia ca. 2000–2013 – 1.71 1.94 – 2.60 3.79 – 1.74 – 0.27 – 1.74 0.82 1.5 For each inequality measure, this table presents per cent per annum changes calculated between beginning and end of the observation period. For Eastern Europe (in italics) changes have been calculated over a more recent period, with declining mortality, only. In green: narrowing of inequalities in mortality; in red, widening of inequalities in mortality. Each inequality measure was also used to calculate a country’s rank (with 1 representing the country with most narrowing or least widening of inequalities), and then to calculate a country’s average rank over all eight inequality measures (with lower values indicating more narrowing or less widening; see Web appendix Table A4). In green: best performing countries; in red: worst performing countries widen over time is dependent on the measure used. There is Another major strength of our study is that we have been no consensus among researchers or policy-makers on what able to study possible determinants of the magnitude of ine- measure of health inequalities to use, and apart from tech- qualities in mortality. Studies of determinants of the mag- nical considerations value judgements are inevitable when nitude of health inequalities at the country-level are almost choosing between measures [19]. We have therefore based non-existent, and the only previous study we are aware of our rank order of countries on a range of different meas- studied one point in time only [11]. As our current study ures, helped by the fact that this ranking is not too different exploits time-series data using country- and period-fixed between measures (Web appendix table A4). effects, it controls for a wide range of potential confounders. 13 Progress against inequalities in mortality: register-based study of 15 European countries… 1139 Fig. 3  Contribution of 13 causes of death to changes in absolute Eastern Europe changes have been calculated over the most recent inequalities in mortality between low and high educated. a Men, b observation period, with declining mortality. Excluding all other Women. Notes: Difference between low and high expressed in annual causes of death mortality change, expressed as deaths per 100,000 person-years. For However, because our analyses are ecological in nature, the countries were collected in a longitudinal design. How- risk of ecological fallacy prevents us from making causal ever, as our main concern is with changes in mortality over inferences, particularly about relationships between deter- time, and as the study design has remained identical, this is minants and mortality at the individual level [20]. unlikely to have biased our findings. Our study has several limitations. For Spain and Italy, We used level of education as our measure of socioeco- only urban and relatively prosperous populations could be nomic position, and this does not cover all dimensions of included. However, recent national-level studies from Spain social inequality. However, recent studies of European coun- and Italy found similarly small inequalities in mortality in tries for which more indicators are available, have shown the whole country as in these regional populations [21, 22]. that trends of mortality inequalities by occupational class are Data from Spain, Hungary and Poland were collected in a largely similar to those by educational level [23]. Our study cross-sectional unlinked design, whereas data from all other covers adult ages between 35 and 80 years, implying that the 13 1140 J. P. Mackenbach et al. Table 2  Results of regression analyses Effect of time on determi- Main effect of determinant on Interaction effect between Effect of trend in nant in whole population mortality determinant and low educa- determinant on mortality tion on mortality inequalites Men Women Men Women Men Women Men Women National income 0.043*** 0.043*** − 338*** − 78 (n.s.) − 69.4 (n.s.) 15.0 (n.s.) None None Income inequality 0.158*** 0.158*** 16.7*** 0.6 (n.s.) 37.2*** 18.5*** Widening Widening Level of democracy 0.105*** 0.105*** 1.1 (n.s.) − 23.7*** 90.8*** 74.3*** Widening Widening Health expenditure 0.124*** 0.124*** − 29.2 (n.s.) − 8.9 (n.s.) − 168*** − 53*** Narrowing Narrowing Control variables c, e c, e c, p, e, g c, p, e, g c, p, e, g c, p, e, g Effect of time on determinant in Interaction effect between Effect of determinant on Effect of trend in deter- whole population time and low education mortality minant on mortality determinant inequalities Men Women Men Women Men Women Men Women Smoking − 0.004*** 0.000 (n.s.) 0.005*** 0.007*** 1274*** 361*** Widening Widening Material deprivation − 0.002 (n.s.) − 0.001 (n.s.) 0.001 (n.s.) 0.001 (n.s.) 1134*** 265*** None None Control variables c, e c, e c, e, y c, e, y c, e, y c, e, y Control variables: c = country, p = period, e = education, g = GDP, y = year. In bold: statistically significant (p < 0.05). ***p ≤ 0.001. n.s. = not statistically significant (p > 0.05). More extensive results in Web appendix Table A5 Interpretation The challenge facing countries that aim to reduce inequali- ties in mortality in a context of rapid mortality decline, is to make sure that the mortality decline in lower socioeco- nomic groups exceeds that of higher socioeconomic groups. The general rise of relative inequalities in mortality shows that almost no European country has achieved this for rela- tive (i.e., percentage) mortality declines. On the other hand, many European countries have achieved this for absolute mortality declines, particularly among men, as shown by a decline of absolute inequalities in mortality. It can also be mathematically demonstrated that a narrowing of absolute inequalities occurs under a much wider range of conditions, in terms of combinations of changes to mortality in lower and higher socioeconomic groups, than a narrowing of rela- tive inequalities in mortality [24]. It is encouraging to see that some European countries have achieved absolute reductions in mortality among the low educated strongly exceeding those among the high edu- cated. This applies, first of all, to some Eastern European countries, particularly Hungary and Estonia. Although the period of mortality decline covered by our analysis is still Fig. 4  Association between health care expenditure and years of life lost. Notes: Each dot represents one observation (years of life lost by rather short, the achievements of both countries are spec- country, period, gender, level of education). Similar results for other tacular. The recent declines in average mortality in Eastern mortality measures (mortality rates, life expectancy) Europe have been attributed to a combination of long-term declines in smoking, improvements in health care, dietary changes, road traffic safety measures and alcohol control many deaths occurring at higher ages are not included, and policies [25]. Apparently, Hungary and Estonia have man- neither are deaths occurring at younger ages for which trends aged to bring the benefits of these mortality-lowering poli- and contributing causes may partly be different. cies also to the low educated. 13 Progress against inequalities in mortality: register-based study of 15 European countries… 1141 However, some Western European countries have also and not to non-amenable causes [32]. These results were done very well, particularly Austria, England and Wales, obtained with regression models controlling for national Italy (Turin), Spain (Barcelona) and Switzerland. Our income and country- and period-fixed effects, and are there- cause-specific mortality results suggest that the reductions fore robust against various sources of confounding. in absolute inequalities in mortality have been achieved as This suggests that in many European countries health a by-product of population-wide improvements in preven- care expansion, in response to greater demand and enlarged tion and treatment, particularly for ischemic heart disease, opportunities for medical intervention, has not only contrib- smoking-related causes, and causes amenable to medical uted to average mortality decline as shown elsewhere [33], intervention. In the case of ischemic heart disease, these but may also have had a dampening effect on inequalities trends must be due to more favourable changes in either in mortality. Even though it would be better to tackle the ‘proximate’ determinants of ischemic heart disease, such as root causes of health inequalities [34], this is good news, health-related behaviours (e.g., smoking, diet, exercise) or and shows that the role of health care in reducing, or simply health care effectiveness (e.g., hypertension detection and containing, health inequalities should not be ignored. treatment, thrombolytic therapy), or both, among the low than the high educated. A recent narrowing of absolute ine- Acknowledgements This study was conducted as part of the LIFE- PATH project, which has received financial support from the Euro- qualities in cardiovascular disease mortality has previously pean Commission (Horizon 2020 Grant Number 633666). Data were been reported for both England and Scotland, and has been partly collected as part of the DEMETRIQ project, which also received attributed to an even distribution of treatment benefits rather support from the European Commission (Grant Number FP7-CP-FP than to risk factor changes [26]. Grant No. 278511). The work of Johan Mackenbach, Wilma Nusselder and José Rubio Valverde is also supported by Netspar (Network for The worst performers, both among men and among Studies on Pensions, Aging and Retirement). We gratefully acknowl- women, are three Nordic countries: Finland, Norway and edge the contributions of the following persons to data collection: Dr Denmark. We have demonstrated previously in cross-sec- Johannes Klotz (Statistics Austria, Vienna, Austria), Prof. Olle Lund- tional analyses that, unlike what one would expect on the berg (Center for Health Equity Studies, Stockholm, Sweden), Dr Bjørn Heine Strand (National Institute of Public Health, Oslo, Norway), and basis of their universal and generous welfare states, these Dr Bogdan Woytiniak (Department of Monitoring and Analyses of countries do not have smaller health inequalities than other Population Health, National Institute of Public Health-National Insti- European countries [27]. This new analysis shows that over tute of Hygiene, Warsaw). The permission of the Office for National the past decades, these countries have also experienced Statistics to use the Longitudinal Study is gratefully acknowledged, as is the help provided by staff of the Centre for Longitudinal Study Infor- relatively unfavourable trends in inequalities in mortality. mation and User Support (CeLSIUS). CeLSIUS is supported by the This ‘Nordic’ paradox has a complex explanation, includ- ESRC Census of Population Programme (Award Ref: ES/K000365/1). ing further advanced changes in social stratification and The authors alone are responsible for the interpretation of the data. This social mobility than elsewhere in Europe; further advanced work contains statistical data from ONS which is Crown Copyright. The use of the ONS statistical data in this work does not imply the changes in smoking, fertility behaviour and other proximal endorsement of the ONS in relation to the interpretation or analysis determinants of population health; and stubborn persistence of the statistical data. This work uses research datasets which may not of inequalities in social disadvantage [28]. exactly reproduce National Statistics aggregates. The mortality data for Our study of determinants of mortality inequalities Switzerland were obtained from the Swiss National Cohort, which is based on mortality and census data provided by the Federal Statistical showed that trends over the past 25 years in several deter- Office and supported by the Swiss National Science Foundation (Grant minants may have contributed to a widening of inequali- Nos. 3347CO-108806, 33CS30_134273 and 33CS30_148415). ties in mortality. Income inequality has gone up [29], and this has been associated with rising mortality, but more so Author contributions JPM conceptualized the study, participated in among the low than the high educated, suggesting an ine- the analysis, and wrote the first and final drafts of the paper. JRV and WJN participated in the analysis and commented on the paper. MB, qualities-widening effect. Levels of democracy have gone HB-H, GC, PD, RK, KK, ML, PM, GM and MR-S prepared data and up in Eastern Europe in the 1990s, and the accompanying commented on the paper. All authors approved the final version. JPM political and economic turbulence has been associated with a is guarantor of the paper. temporary rise in mortality, but more so among the low than the high educated, as a result of their greater vulnerability Funding The funders had no role in the study design; in the collection, analysis, and interpretation of data; in the writing of the report; and to increased economic stress and disruption of social and in the decision to submit the article for publication. The researchers health care services [30]. Smoking has mostly gone down, worked in independence from the study’s funders and all authors had but more so among the high than among the low educated full access to all of the data. [31], again implying an inequalities-widening effect. Only rising health care expenditure was associated with Compliance with ethical standards smaller absolute inequalities in mortality. In a more detailed analysis reported elsewhere we have shown that this effect Conflict of interest The authors declare that they have no conflict of interest. applies particularly to causes amenable to medical care, 13 1142 J. P. Mackenbach et al. Open Access This article is distributed under the terms of the Crea- 18. Vaupel JW. How change in age-specific mortality affects life tive Commons Attribution 4.0 International License (http://creat​iveco​ expectancy. Popul Stud. 1986;40(1):147–57. mmons​.org/licen​ses/by/4.0/), which permits unrestricted use, distribu- 19. Harper S, King NB, Meersman SC, Reichman ME, Breen N, tion, and reproduction in any medium, provided you give appropriate Lynch J. Implicit value judgments in the measurement of health credit to the original author(s) and the source, provide a link to the inequalities. Milbank Q. 2010;88(1):4–29. Creative Commons license, and indicate if changes were made. 20. Piantadosi S, Byar DP, Green SB. The ecological fallacy. Am J Epidemiol. 1988;127(5):893–904. 21. Marinacci C, Grippo F, Pappagallo M, et al. Social inequalities in total and cause-specific mortality of a sample of the Italian popu- References lation, from 1999 to 2007. Eur J Public Health. 2013;23(4):582–7. 22. Regidor E, Kunst AE, Rodriguez-Artalejo F, Mackenbach JP. 1. Mackenbach JP, Stirbu I, Roskam AJ, et al. Socioeconomic Small socio-economic differences in mortality in Spanish older inequalities in health in 22 European countries. N Engl J Med. people. Eur J Public Health. 2012;22(1):80–5. 2008;358(23):2468–81. 23. de Gelder R, Menvielle G, Costa G, et al. Long-term trends in 2. Toch-Marquardt M, Menvielle G, Eikemo TA, et al. Occupational socioeconomic inequalities in mortality in 6 European countries. class inequalities in all-cause and cause-specific mortality among Int J Public Health. 2017;62(1):127–41. middle-aged men in 14 European populations during the early 24. Mackenbach JP, Martikainen P, Menvielle G, de Gelder R. The 2000s. PLoS ONE. 2014;9(9):e108072. arithmetic of reducing relative and absolute inequalities in health: 3. Mortensen LH, Rehnberg J, Dahl E, et al. Shape of the associa- a theoretical analysis illustrated with European mortality data. J tion between income and mortality: a cohort study of Denmark, Epidemiol Community Health. 2016;70:730–6. Finland, Norway and Sweden in 1995 and 2003. BMJ Open. 25. Mackenbach JP, Karanikolos M, Lopez Bernal J, Mckee M. 2016;6(12):e010974. Why did life expectancy in Central and Eastern Europe suddenly 4. Mackenbach JP. The persistence of health inequalities in mod- improve in the 1990s? An analysis by cause of death. Scand J ern welfare states: the explanation of a paradox. Soc Sci Med. Public Health. 2015;43(8):796–801. 2012;75(4):761–9. 26. Bajekal M, Scholes S, Love H, et al. Analysing recent socio- 5. Mackenbach JP, Kulhánová I, Artnik B, et al. Changes in mortality economic trends in coronary heart disease mortality in Eng- inequalities over two decades: register based study of European land, 2000–2007: a population modelling study. PLoS Med. countries. Br Med J. 2016;353:i1732. 2012;9(6):e1001237. 6. Galobardes B, Lynch J, Smith GD. Measuring socioeconomic 27. Mackenbach JP, Kunst AE, Cavelaars AE, Groenhof F, Geurts JJ. position in health research. Br Med Bull. 2007;81–82:21–37. Socioeconomic inequalities in morbidity and mortality in western 7. Mackenbach JP, Looman CWN. Life expectancy and national Europe. Lancet. 1997;349(9066):1655–9. income in Europe, 1900-2008: an update of Preston’s analysis. 28. Mackenbach JP. Nordic paradox, southern miracle, eastern disas- Int J Epidemiol. 2013;42(4):1100–10. ter: persistence of inequalities in mortality in Europe. Eur J Public 8. Pickett KE, Wilkinson RG. Income inequality and health: a causal Health. 2017;27(suppl_4):14–7. review. Soc Sci Med. 2015;128:316–26. 29. OECD. Divided we stand: why inequality keeps rising. Paris: 9. Mackenbach JP, Hu Y, Looman CWN. Democratization and life OECD Publishing; 2011. expectancy in Europe, 1960–2008. Soc Sci Med. 2013;93:166–75. 30. Leinsalu M, Stirbu I, Vagero D, et al. Educational inequalities in 10. van Baal P, Obulqasim P, Brouwer W, Nusselder W, Mackenbach mortality in four Eastern European countries: divergence in trends J. The influence of health care spending on life expectancy. Til- during the post-communist transition from 1990 to 2000. Int J burg: Netspar; 2013. Epidemiol. 2009;38(2):512–25. 11. Mackenbach JP, Bopp M, Deboosere P, et al. Determinants of the 31. Giskes K, Kunst AE, Benach J, et al. Trends in smoking behaviour magnitude of socioeconomic inequalities in mortality: a study of between 1985 and 2000 in nine European countries by education. 17 European countries. Health Place. 2017;47:44–53. J Epidemiol Commun Health. 2005;59(5):395–401. 12. Ahmad OB, Boschi-Pinto C, Lopez AD, Murray CJL, Lozano 32. Mackenbach JP, Hu Y, Artnik B, et al. Trends In inequalities R, Inoue M. Age standardization of rates: a new WHO standard. in mortality amenable to health care in 17 European countries. Geneva: World Health Organization; 2001. Health Aff. 2017;36(6):1110–8. 13. Kjellsson G, Gerdtham U-G, Petrie D. Lies, damned lies, and 33. Heijink R, Koolman X, Westert GP. Spending more money, health inequality measurements: understanding the value judg- saving more lives? The relationship between avoidable mortal- ments. Epidemiology. 2015;26(5):673–80. ity and healthcare spending in 14 countries. Eur J Health Econ. 14. Mackenbach JP, Kunst AE. Measuring the magnitude of socio- 2013;14(3):527–38. economic inequalities in health: an overview of available meas- 34. Commission on Social Determinants of Health. Closing the gap ures illustrated with two examples from Europe. Soc Sci Med. in a generation. Health equity through the social determinants of 1997;44(6):757–71. health. Geneva: World Health Organization; 2008. 15. Moreno-Betancur M, Latouche A, Menvielle G, Kunst AE, Rey G. Relative index of inequality and slope index of inequality: a Publisher’s Note Springer Nature remains neutral with regard to structured regression framework for estimation. Epidemiology. jurisdictional claims in published maps and institutional affiliations. 2015;26(4):518–27. 16. Latouche A, Andersen PK, Rey G, Moreno-Betancur M. A note on the measurement of socioeconomic inequalities in life years lost by cause of death. Epidemiology. 2019;30(4):569–72. 17. Keyfitz N, Caswell H. Applied mathematical demography. New York: Springer; 2005. 13