Background
Social inequalities in health, particularly in regard to mortality, are a relevant public health problem [
1]. It is thus important to quantify them, determine their geographical patterns, and monitor them over time. Doing so makes it possible to provide evidence to governments and a diverse range of decision-makers and public administrators for the implementation of policies and interventions intended to reduce or erradicate these inequalities. Over the last decade there has been a proliferation of studies analysing socioeconomic inequalities in mortality taking geographical areas as the unit of analysis. There are at least three reasons for this proliferation. The first is the consideration that certain characteristics or attributes of an area of residence may be health determinants for the people living in them [
2,
3]. The second is that studies of this type permit the identification of those geographical areas with unfavourable socioeconomic and health indicators [
4]. The third is that using ecological data makes it more feasible to perform routine surveillance of health inequalities [
5].
However, the inequalities in mortality may differ depending on whether one studies rural or urban areas; indeed inequalities tend to be greater in urban areas since often they include deprived and poor populations being concentrated in marginalized neighbourhoods and urban slums [
5‐
7]. Also, the fact that the majority of the world’s population lives in urban areas means that the study of the processes acting in these areas is key to understanding the economic, cultural, political and health-related transformations which occur [
4,
8,
9]. In Spain, numerous studies have reported the existence of socioeconomic inequalities in all-causes mortality in urban areas, and that areas with greater deprivation present higher mortality risk [
10‐
16].
Various ecological studies have analysed time trends in socioeconomic inequalities in all-causes mortality in urban areas [
17‐
23]. However, the findings of these studies are rather heterogeneous, not only between countries but also between urban areas of the same country. In the studies conducted outside Europe, an increase in inequalities over time was observed in Sydney [
17], a slight reduction in New York [
18] and that they remain stable in the Montreal metropolitan area [
19]. Some studies have been carried out in the European context, specifically in Rome it was observed that inequalities were increasing with time, but stabilised near the end of the study period [
20]. In Spain, trends in socioeconomic inequalities in mortality have been analysed in cities such as Sevilla [
21], Cádiz [
22], Barcelona and Madrid [
23,
24], and in general it has been observed that inequalities either decline or remain stable over time.
This heterogeneity in the trends in inequalities in mortality at urban level may be due to, among other things, the different methodologies employed in the studies. Firstly, the cities analysed have different socioeconomic and epidemiological contexts which may mean that inequalites can either increase or decrease. Secondly, the different studies analyse geographical areas with different sizes and population densities. In this sense the effect of the so-called “Modifiable Areal Unit Problem” (MAUP) could make it difficult to compare findings between cities [
25]. Thirdly, periods of time analysed differ between the different cities. Fourthly, some studies analyse premature mortality, while others analyse mortality for all ages. Finally, between the studies, different indicators and methods are used to determine levels of socioeconomic deprivation in small areas [
26]. All this means that through these ecological studies of small areas it is difficult to obtain a clear view of the time trends of inequalities in mortality in urban areas.
With the aim of contributing more evidence on how these inequalities are evolving in Spain, the objectives of this study are: (1) to identify trends in geographical inequalities in all-cause mortality in the census tracts of 33 Spanish cities between the two periods 1996–1998 and 2005–2007; (2) to analyse trends in the relationship between these geographical inequalities and socioeconomic deprivation; and (3) to obtain an overall measure which summarises the relationship found in each one of the cities and to analyse its variation over time. This study forms part of the multicentric MEDEA project (
http://www.proyectomedea.org) [
23,
27,
28], which has made it possible to obtain data for the 33 cities in a relatively homogeneous manner, thus solving most of the problems described above. Moreover, in this study, apart from analysing each of the cities independently (the usual procedure), we have conducted a joint analysis by means of a multilevel random effects model, which allowed us to take account of the spatial structure of the data and obtain an overall result which summarises the behaviour of the cities as a whole (as if a metaanalysis had been performed).
Results
Additional file
2: Table S1 shows the numbers of census tracts in each city. The number varies from 57 in Pontevedra to 2358 in Madrid. This table also includes the numbers of deaths, population at risk, and the ASMR per 100,000 inhabitants by period and sex. These rates tend to decline in all the cities and in both sexes, with the exception of Zaragoza where rise in both sexes.
For the majority of cities and in both sexes, the ASMR fall, both in census tracts with less deprivation (first tertile of deprivation), and in those with more deprivation (third deprivation tertile) (Tables
1 and
2). In general, the variations over time of the absolute and relative differences in mortality (between census tracts with less and with more deprivation) go in the same direction. Thus, when the absolute differences increase, so do the relative differences, and similarly, when absolute differences decrease so do the relative differences. In both men and women, we observe that in 20 of the 33 cities the absolute differences decrease over time, while the relative differences decrease in 17 of them. These decreases are due, in the majority of cases, to a greater reduction in mortality in the census tracts with more deprivation, in comparison to those with less deprivation.
In men, all-cause mortality presents a similar geographical pattern to socioeconomic deprivation for the majority of cities and both periods (Additional file
3). In women in the first period we also observe this similarity in the majority of the cities. However, in the second period these spatial patterns become different in most of the cities studied. For example, Fig.
1 presents the geographical distributions of the deprivation index and of the sSMR for the city of Barcelona in the two study periods (1996–1998 and 2005–2007) and by sex. In this figure we may observe how, in men, the geographical distribution of the deprivation index (Fig.
1a) is similar to the geographical distribution of the sSMR in both periods (Fig.
1b and
c). The women follow the same pattern, except in the second period where it can clearly be seen how the sSMR have a rather heterogeneous spatial pattern, which is different from that of deprivation (Fig.
1e).
Figure
2 presents the association between the deprivation index and all-cause mortality in men, by city and period, obtained using Model 2. The RR are over 1 in the two periods in all cities, and these RR are significant (95 % CI does not include 1) in 24 cities in the first period and in 25 in the second. Looking at the time trend in the RR (i.e. reflected by the probability Pr(
ΔRR)), represented using coloured dots in the figure, we see that in 29 cities there is no significant change over time in the RR (Pr(
ΔRR) ∈ [0.025, 0.975)). However, in 3 cities (Madrid, Málaga and Zaragoza) the RR decrease with time and in 1 (Coruña) the RR increase. In regard to women (Fig.
3), the RR are over 1 in the two periods in the majority of the cities, and significant in 11 cities in the first period and in 10 in the second. In contrast, only in the city of Córdoba do we find a RR significantly under 1 (RR = 0.91, 95 % CI = (0.85–0.98). Turning to the trends over time in the RR represented by Pr(
ΔRR), we see that in 29 cities there is no significant evolution in the RR (Pr(
ΔRR) ∈ [0.025, 0.975)). However, in 4 cities (Barcelona, Córdoba, Madrid and Zaragoza) the RR decrease with time. In general the RR for women are lower than those for men in both periods.
Additional file
4: Figure S2, like Fig.
2, presents the association between the deprivation index and all-cause mortality among men, by city and period, obtained using Model 3. In general this model provides much more consistent associations between the different cities than Model 2, since in all the cities and in both periods we obtained RRs significantly greater than 1. Also, it may be seen that in 4 cities (Cádiz, Madrid, Málaga and Zaragoza) the RR decrease with time and the rest of the cities remain stable (Pr(
ΔRR) ∈ [0.025, 0.975)). If we look at the RR obtained using Model 3 in women (Additional file
5: Figure S3), we see that in the first period they are significantly greater than 1 in all the cities except Vitoria. In the second period, the majority of cities have an RR greater than 1, being significantly so in 19 of them. Also, we may observe that in 4 cities (Barcelona, Córdoba, Madrid and Zaragoza) the RR decrease with time and the rest of the cities remain stable (Pr(
ΔRR) ∈ [0.025, 0.975)). Via Model 3 we may also observe that, in general, RR for women are lower than those for men in both periods.
Finally, through Model 3 we calculated a global RR (RRg) which summarises the association of all the cities. This RRg was calculated for each period and sex (Additional file
4: Figure S2 and Additional file
5: Figure S3). In men, the RRg is significantly greater than 1 both in the first period (RRg = 1.13, 95 % CI = 1.12–1.15), and in the second period (RRg = 1.11, 95 % CI = 1.09–1.13). Moreover, the probability that RRg in the second period be greater than that of the first period is very low (under 0.025) which implies that the RRg have fallen with time. Among women the behaviour of the RRg is similar to that of men, although the associations found are weaker. Specifically, the RRg for the first period is 1.07 (95 % CI = 1.05–1.08) and in the second period is 1.04 (95 % CI = 1.02–1.06), with a probability that the second be greater than the first of under 0.025.
Abbreviations
AIDS, acquired immune deficiency syndrome; ASMR, age standardised mortality rate; BYM, Besag, York and Mollié; CI, credible interval; ICAR, Intrinsic Conditional Autoregressive; MAUP, Modifiable Areal Unit Problem; RR, relative risk; RRg, global RR; sSMR, smoothed Standardized Mortality Ratio; T1, 1st tertile; T3, 3rd tertile
Acknowledgments
We would like to thank the members of the MEDEA Project for providing the data in order to perform this study.