Roots in Health Disparities among European Regions

Study Design: This is a descriptive and prediction approach to MINDEX.

Methods: We use factor analysis to measure MINDEX and blocks of predictors (competitiveness and urbanization of regions). Multiple regressions to predict the key dependent variables are used to predict the validity of the measurement model.

Results: Competitive and urbanized regions have a better health status. Even if one controls for these two families of factors, the country context counts significantly from the statistical point of view. Five categories of countries influence the health patterns of regional populations in the EU.

Conclusion: The key contribution of the article is to validate the new index by descriptive and predictive analysis.

Is there a health capital at the level of territorial units? The inventory made at the local community level [1] does not include health status among the seven forms of community capital (social, human, cultural, natural, built, financial and political) but only in the series of problems that may arise, related to poverty and health. Health could, however, be included in the series of capabilities that define human capital. And it is effectively included later in the series of defining capabilities for human capital [2]. Such capital can be identified for the population not only at the level of the localized community but also for regions, states, networks of localities, etc. Implicitly, such a conception is also adopted in the case of the Human Development Index promoted by UNDP since the 1990s [3]. Along with education and economic or material capital, health status, estimated by life expectancy at birth, is included as an essential component of the human development index at the societal level. What if you don't have an estimate of life expectancy at birth for the reference territories? In this case, we would like to estimate the health status capital by NUTS-type regions (from the French nomenclature des unites territorials statistiques) of level 3 in the European Union. EUROSTAT data give estimates of life expectancy (LE) for NUTS 2 tier 2 regions [4] but not for NUTS 3. The substitute that can be used, and we have resorted to in this case, was to identify a set of indicators strongly associated with life expectancy at birth but for which calculation data are available at NUTS 3 level. These are standardized mortality ratio (SMR), under-five mortality rates (Under5R), and overall mortality rate (GMR). From these indices, we generated a synthetic mortality index (MINDEX) that integrates the three indices. 

 In addition, it is important to be able to specify the validity of the mentioned indicators. To what extent can we explain the changes in population health status between NUTS 3 regions by reference to MRS and associated indicators such as overall mortality rate and under-5 mortality rate? If we calculate a mortality index as a factor score that aggregates SMR, overall mortality rate (GMR) and under-five mortality rate (Under5R), is this summary index (MINDEX) an appropriate measure of health status at NUTS 3 level? We will give such validation our interest in the material that follows. The focus will be on predictive or nomological validation [5] given by the ability of different indicators to predict the values of the new health status index. The enterprise deserves to be carried out to better understand the significance of different measures of population health status at the regional level. Once the quality of human life asocial and health status has been measured by MINDEX and associated indices, successive aggregations through population-weighted averages of the region can be switched to estimates for higher measurement levels, NUTS level 2 or national, if the initial estimate was made correctly for NUTS 3.

As expected, even with nationwide data, the increase in life expectancy at birth is accompanied by a decrease in the mortality index. The worst health status is recorded in former communist countries, with the status of New Member States in the European Union (EU). Especially in Bulgaria, Romania, Hungary had the worst health status in 2019, in terms of high mortality index and low life expectancy at birth. Slovakia, Hungary and Poland also have a problematic health situation, placed in the same quadrant of high mortality and low life expectancy at birth. They are also from the category of new EU Member States.

Figure 1: A spreadsheet diagram of the EU countries by life expectancy and index of mortality 2019.

Data source: EUROSTAT. Own computations. Values for East (DEEast) and West (DEWest) Germany were computed by the author, knowing what the NUTS 3 about each of the two German regions.

At the opposite end, with low mortality and increased life expectancy at birth, are old EU Member States. In this category, three groupings are distinguished. The first grouping consists of or close Nordic countries, namely Sweden, Finland, the Netherlands and Ireland. Spain, Italy and France form a second grouping, predominantly located in the South, with high life expectancy at birth, but with mortality rates close to the EU average. Denmark, West Germany and Austria form a third grouping, with high life expectancy at birth but mortality rates close to the European average. Otherwise, there are specific situations, and groupings with a small number of countries. Latvia and Lithuania, for example, have low mortality but relatively low life expectancy at birth. The regions of East Germany are in worse health than those of West Germany, but better than those of Bulgaria, Romania and Hungary, for example. We will see if country groupings maintain a similar configuration when the country effect can be isolated or controlled by multiple regression. Assumptions 1 and 2, mentioned in the methodology section, are supported by multiple regression analysis when the dependent variable is MINDEX. Competitiveness and high-level urbanization tend to lead to lower regional mortality (Table 1).

Table 1: Predicting health status at NUTS 3 level.

Data source: EUROSTAT. Own computations. OLS regressions in STATA 16. Robust standard errors. Colinearity effects, identified by VIF command in STATA, were avoided by eliminating dummies of counties generating VIF that are higher than 4. Very small countries like MT, EE, LV, CY were eliminated from the series of country predictors. West Germany was also eliminated from the list of predictors to avoid collinearity. Shadow marks significant coefficients for p less than 0.05. Urbanization and competitiveness are factor scores in a separate PCA. Urbanization factor aggregates urbanity of NUTS 3, density, and having not the capital city of the country in the reference NUTS region. All the variables for the factor score competitiveness are measured at the NUTS 2 level. It aggregates regional competitiveness, education of the population of more than 25 years old, net migration rate, and NEET measures the percentage of people 15-29 years old that are neither in employment nor in education institutions.

On the component indices of MINDEX, the impact is differentiated. It is not clear why the standardized mortality rate (SMR) is not lower in regions with increased competitiveness (the corresponding regression coefficient is not statistically significant). It is also unclear why increased regional competitiveness tends to lead to a higher overall mortality rate. Both deviations from theoretical expectations seem to confirm the idea that measuring by multiple indicators (such as MINDEX) is better than measuring health by single indicators. Of the three mortality indices, the one that comes closest to the synthetic index pattern (MINDEX) is the standardized mortality ratio. The other two (Under5R and GMR) are more affected by the age structure of the population, uncontrolled in the regression equation in Table 1 to avoid collinearity effects. But what if we replace, in the regression model, synthetic indices related to competitiveness and urbanization with the main component indicators of the factorial score? In other words, to test the sensitivity of the analysis model (Treiman 2014) to methodological changes in work, we replace the factorial score of competitiveness with the gross domestic product per capita, the level of education and net migration, and the factorial score of urbanization with the degree of urbanization of the NUTS 3 region (1 urban, 2 intermediate, 3 rural). We have made this change and present the results of the analysis in Table 1 of the Annex. The results appear of identical significance when MINDEX is considered as a dependent variable. Even if we replace the factorial scores in the predictors with the component indicators, the meaning of the relationships remains the same. MINDEX tends to be lower for NUTS 3 where economic development, education stock and migratory attraction are higher and urbanization higher. For SMR, the only statistically significant predictor in the new Annex 1 variance is education level. The higher the education level of the population in the NUTS 3 region, the lower the MRS. The mortality of children under 5 tends to be lower for more economically and socially developed regions.

What is the country's effect on the health of the population if we keep under control, relatively constant, the values of competitiveness and urbanization? This question is answered by all the data in Table 1 and Annex 1 of the article.

With the analysis in Table 1, country groupings, as a type of effect on health, appear clearer but consistent with the predictions in the bivariate analysis (Figure 1). The new EU Member States appear as two groupings with significant mortality bias effects. Bulgaria, Romania, Hungary and Croatia are the countries with the maximum effect of poor health, followed by Slovakia, Poland and the former German Democratic Republic. As in the bivariate analysis (Figure 1), in the multivariate analysis appears a precisely defined grouping of countries with maximum positive impact on health, consisting of Nordic countries or close to them, namely Sweden, Finland, Ireland and the Netherlands. Also, with a positive impact on health are grouped separately the southern countries, Italy and Spain, but also those close to them, also from the Old EU Member States, respectively France and Austria. Belgium, Portugal, Greece and the Czech Republic appear as countries without a specific significant effect on health. Replacing the two predictors related to competitiveness and urbanization with key component indicators (compare Table 1 in the article with Table 1 in the Annex) does not lead to major changes in the configuration of country effects on mortality indicators. In extreme groupings of countries that contribute to increased mortality (BG, RO, HU, HR, GDR, PL) or to its reduction (SE, FI, NL, IE) no major changes occur in the series of regression coefficients for MINDEX prediction. The exception is the case of Slovakia, where we no longer record, in Annex 1, a significant increase in mortality. However, there are cases of changes in coefficients in some intermediate cases. Denmark, for example, in the new Annex 1 run no longer appears as a significant predictor for mortality reduction.

Annex 1: An alternative model to predicting health status at NUTS 3 level.

Data Source: EUROSTAT. Own computations. OLS regressions in STATA 16. Robust standard errors. Colinearity effects, identified by VIF command in STATA, were avoided by eliminating dummies of counties generating VIF that are higher than 4. Very small countries like MT, EE, LV, and CY were eliminated from country predictors. West Germany was also eliminated from the list of predictors to avoid collinearity. Shadow marks significant coefficients for p less than 0.05.

The health status of the population is strongly structured at the level of NUTS 3 European regions. The essential reference frameworks in structuring it appear to be strongly associated with urbanization, competitiveness and the large geographical-historical groupings of nation-states. If health status is no longer measured cumulatively, by several mortality indicators, regularity is no longer maintained. This finding follows the recommendation to use the composite mortality index (MINDEX) as the preferred measurement for regional health status, compared to single, non-cumulative indices, such as SMR, the overall mortality rate has mortality rates of children under five years of age. The urbanization and increased competitiveness of regions induce, caeteris paribus, better health. The country's effect on the health status of the population remains significant and strongly differentiated. What was behind this regularity is likely the state of health services associated with the type of health culture, both aspects were not measured directly in the analysis. These country effects on the health status of regional populations allow five types of countries to be distinguished. At one extreme are Nordic countries or close to them with very good regional health, such as Sweden, Finland, the Netherlands and Ireland. At the opposite pole, with poor health and, implicitly, high mortality, are regions from the former communist countries – Bulgaria, Romania, Hungary, Croatia, Slovakia, the former German Democratic Republic and Poland. Between the two extremes already mentioned, we have identified a grouping of states with regions that do not have a significant impact on health (Belgium, Greece, Portugal and the Czech Republic). A grouping of southern (Italy, Spain) and western (France and Austria) countries stands out for regions with health status significantly above the European average. In the analysis, we also identified a grouping of three small countries with good health status (Denmark, Slovenia and Latvia).

The mortality index (MINDEX), factorial aggregation of three indicators (SMR, GMR, Under5R) proves to be valid and consistent (faithful). External or predictive validity is given by the significant and interpretable impact of competitiveness, urbanization and membership of regions to EU national frameworks in determining the MINDEX composite index. The Economic and Social Competitiveness Index (used in the regressions in Table 1) is calculated as a factorial score from variables measured at the NUTS 2 level. Country average values of MINDEX are calculated by aggregating regional values. It follows that MINDEX validation also involved multilevel analyses that proved useful. An additional argument in favour of MINDEX as a synthetic health index resulted from the stability of the prediction model of MINDEX chir even under conditions of minor modification of the list of predictors.

Of course, problems remain to be solved. In particular, the degree of specification of the multiple regression model can be increased by adding predictors such as the quality of health services at the national level, the type of medical culture of the population in a national or regional context, etc.

Conflicts of Interest

No conflict of interest.

Data Availability

All the data that are used for this article are public, being downloaded from the EUROSTAT site.

Funding

No funding was received for this paper.