Design, units of analyses and study population

This study is part of a wider European project called EUROHEALTHY (http://www.euro-healthy.eu/), which aims to increase knowledge and resources related to policies promoting health and health equity across European regions, with a focus on metropolitan areas. We performed an ecological study of trends based on three periods, two before the economic crisis (2000-2003 and 2004-2008), and one during the crisis (2009-2014). The units of analysis were small areas in nine European urban areas: Athens metropolitan area, Barcelona city, Berlin-Brandenburg Metropolregion, Brussels-Capital Region, Lisbon Metropolitan Area, Greater London, Prague city, Turin city, and Stockholm metropolitan area. These areas were selected because of their availability within the EURO-HEALTHY project, which covers a broad range of geographic areas and socioeconomic conditions²⁰. The study population consisted of individuals residing in these areas during the three time periods.

Information sources

For most of the urban areas, mortality and population data were available for the three time periods, and socioeconomic indicators for 2001 (Tables 1 to 3). Mortality data were mainly obtained from mortality registers. Population data stratified by age (5-year groups) and sex were obtained from census or population registers. Socioeconomic data were mainly obtained from the census records.

Variables and indicators

In this study we analysed all-cause mortality in each small area of residence during the three time periods. The mortality indicator used for this analysis was the Standardized Mortality Ratio (SMR). For descriptive purposes we used the Indirectly Standardised Rate (ISR), which we calculated by multiplying the SMR for a study population by the crude rate in the standard population (defined as the European Union [EU]-28 in the year 2007).

We used a composite deprivation index in 2001 as a socioeconomic indicator for each small area. This index was the first component of a principal components analysis performed within each urban area.²¹ It includes unemployment in people aged ≥16 years; the percentage of manual workers in people aged ≥16 years; the percentage of people aged 25-64 years with primary education as their highest education (ISCED 0 and 1, except London where it was ISCED 0, 1 and 2); and the percentage of people aged 25-64 years with a university education (ISCED 5 and 6).²² There were no data for the percentage of manual workers in Stockholm, so the index was constructed using the other three available indicators.

Data analysis

The SMR depends on population size because its variance is inversely proportional to the expected values. Therefore, areas with low populations tend to present very variable estimates. To smooth the SMR, we used the hierarchical Bayesian model proposed by Besag, York and Mollié,²³ which takes two types of random effects into account, spatial and heterogeneous. The former considers the spatial structure of the data, while the latter deals with nonstructural (non-spatial) variability. We used the following model to estimate smoothed SMR (sSMR) for each sex, cause of death, and period:

where, for each area i, O_i is the number of observed cases, E_i is the expected number of cases, ϴ_i is the expected sSMR with respect to the European population, S_i is the spatial effect, and H_i is the heterogeneous effect. The expected number of cases were calculated by indirect standardization, taking the mortality rates for 5-year age groups of the EU-28 in 2007 (approximately the year in the middle of the study period) as reference.

The geographical distribution of the sSMR, for each time period, was represented using septile maps. The socioeconomic indicators and the composite deprivation indicator were also represented as septile maps.

To analyse the evolution of socioeconomic inequalities, we fitted an ecological regression model that takes into account the composite deprivation index (D), the period (through two dummy variables P2 and P3) and their interaction:

where, for each area i and period t (t = 1 for the first period, t = 2 for the second period and t = 3 for the third period), O_it is the number of observed cases, E_it is the expected number of cases, ϴ_it is the sSMR with respect to the European population, S_it is the spatial effect, and H_it is the heterogeneous effect. Finally, P_2t and P_3t take the following values: P_jt = 1 if j = t, and 0 otherwise, where j = 2 or 3. The expected number of cases was calculated as in the previous model. Changes between periods in the relationship between the socioeconomic deprivation index and mortality were evaluated by the interaction terms in model 2.

In the two models, the spatial effect was assigned an intrinsic conditional autoregressive prior distribution, which assumes that the expected value of each area coincides with the mean of the spatial effect of the adjacent areas. Its variance σ_s² depends on the number of adjacent areas, since those areas that are more connected will show lower variability.²⁴ The heterogeneous effect was represented using independent normal distributions with mean 0 and variance σ_h². A uniform distribution U(0,∞) was assigned to the standard deviations σs and σh. A normal vague prior distribution, with mean 0 and precision 0.001, was assigned to the parameters α, β1, β2, β3, β4 and β5.

Since the composite deprivation index scale is dimensionless and arbitrarily fixed, we calculated the increase in risk that would correspond to a change in the composite deprivation index from its 5th percentile value (low deprivation) to its 95th percentile value (severe deprivation). Relative risk estimates were obtained based on the mean of their posterior distribution, along with corresponding 95% credible intervals.

Analyses were also performed for the single socioeconomic indicators: unemployment, percentage of manual workers, the percentage of those with primary education only, and the percentage of those with university education (results are shown for unemployment and for primary education while the others can be found in online Appendix).

All analyses were performed using R 3.5.0 ²⁵ and the R-INLA package ²⁶.

Changes in mortality due to the economic crisis

There has been much debate about whether mortality rates increase or decrease during economic crises.^{12,27 28 29 30 31 32-33} There has been little consensus, except for the general observation that the number of suicides increases and that of traffic injuries decreases, at least during recent economic crises.^{5,34 35-36}

We did not find any previous studies that focused on urban areas, so it is difficult to compare our results with others. Nevertheless, we observe that mortality during the crisis seems to follow the downward trend observed before the crisis. However, it is not possible to determine whether the rate was different to that in the periods prior to the crisis.

Changes in socioeconomic inequalities in mortality

Even if mortality does not increase for the entire population during periods of economic crisis, it may increase in certain socioeconomic groups, e.g. in socioeconomically deprived groups, who are more negatively affected by economic crises¹⁴. Some studies have shown an increase in inequalities following previous crises, such as during the Southeast Asian, Japanese, and Soviet Union crises, whereas these patterns were not so clear for the Nordic countries.¹⁴ Following the recent crisis, two studies reported an increase in socioeconomic inequalities in total mortality in two areas of Spain (Barcelona city and Andalucía autonomous community), one of which found increasing inequalities in men only.^16-17 Two other studies found a decrease in inequalities in premature mortality in Spain¹⁸ and in the Lisbon Metropolitan area.¹⁹ The last authors suggest that the decrease in inequalities is due to stability or deterioration of health in the middle and upper-middle social groups, and improvements in the most disadvantaged groups. They argue that this is probably an unexpected effect of austerity policies, which have promoted the emergence of newly-deprived and vulnerable groups.¹⁹

In our study we did not find statistically significant changes between periods in socioeconomic inequalities in men. Two statistically significant changes occurred among women in the crisis period: deprivation inequalities increased in Stockholm, and unemployment inequalities increased in Prague. These countries were not affected especially hard by the economic crisis compared to other countries, and indeed to other crises.^37-38 In addition, they have quite different state welfare policies.³⁹ It is well known that Sweden has one of the most generous welfare policies, so the strong inequalities found are quite counterintuitive.

Differences in inequality due to gender

In our study we found more socioeconomic inequalities among men than among women, which has already been reported in previous studies.³⁸ Surprisingly, however, we found some increases in inequality among women only. Allowing for spurious relationships and factors other than the crisis that could provoke such an increase in inequalities, other studies have indeed found similar results. The economic crisis of 1994 in Sweden was associated with greater inequalities in suicide among women, which was partially related to greater job insecurity and poorer working conditions in the female-dominated public sector after the recession.⁴⁰ In the late 1990s economic crisis in South Korea, following neoliberal restructuring imposed by the IMF, females of working age were more affected than males in terms of life expectancy and mortality.⁴¹ These authors argued that, since the economic crisis, women from disadvantaged classes had to join the labour force to compensate for their husbands' unemployment and wage cuts, and thus suffered both household economic collapse and unstable working conditions.⁴¹

Limitations and strengths

We only had socioeconomic data for 1 year and thus considered them constant during all of the periods. Although socioeconomic indicators may have changed, the ranking of the areas in terms of deprivation did not change much and we were not using the value itself but only the distribution.

The sizes of the small areas differ between the urban areas studied, in fact some of the areas may not be considered as small areas given their size, which may partly explain why the results are not consistent. Smaller areas are more homogeneous and the possibility of observing higher effects is higher, which may be the case in Stockholm, London and Turin. So the comparison of RRs among urban areas is not very suitable in this case.

Finally, we did not consider any lag between the cause (economic recession) and its effects (mortality). It has been reported that the impact of unemployment and other adverse circumstances on health in times of crisis only becomes evident after many years, especially in disadvantaged populations.¹⁴

However, this study has important strengths. It fills the gap in research related to the effect of the crisis on socioeconomic inequalities in mortality in urban areas. Also, it reports mortality and socioeconomic data collected in a comparable way for a number of urban areas from different parts of Europe.