Educational inequalities in premature mortality by region in the Belgian population in the 2000s

The data used in the current study were obtained by linking a) the 2001 Belgian population census, b) the National Population Register, and c) the causes of death database for the period 2001–2011 [5, 20]. Although a more recent census has been held in 2011, it has not been used as the databases linkages have not been performed yet.

The study population comprised all persons aged 25–64 at census, officially residing in Belgium, and having the Belgian nationality at birth (N = 4,556,830 persons). First generation migrants (operationally defined as not having the Belgian nationality at birth) made up 17.5% of the census 2001 in the age group 25–64, with substantial differences by region (i.e., 9%, 50% and 22% in Flanders, Brussels and Wallonia, respectively). First generation migrants (henceforth referred to as “Migrants”) experience lower mortality rates than people with Belgian nationality at birth (henceforth referred to as “Belgians”) [21, 22], for each educational level and each region (Additional file 1: Table S1). In this study, we chose to focus on the Belgian population, which has been exposed since birth to the life conditions and health policies prevailing in Belgium. Furthermore, migrants represent a highly inhomogeneous population with various ethnic and socio-economic backgrounds and would therefore deserve a careful study by origin and ethnicity. The follow-up consisted of a 10 years period after the census (Table 1).

This study included people aged 25–64 at census who were followed up for 10 years, except for the age group 60–64 for which the follow up time was censored at 70 years. Mortality before the age of 70 was defined as “premature mortality”, for the sake of simplicity also referred to as “mortality” in the manuscript.

To assess the contribution of the individual socioeconomic (SE) status to regional mortality differences, we used a set of three SE variables: educational level (EL), employment status and housing status. EL was categorized according to the highest obtained degree using the International Standard Classification of Education (ISCED), version 1997 [23]. Three categories were created: lower secondary education or less (ISCED 0–2; “low”), higher secondary education (ISCED 3–4; “mid”) and tertiary education (ISCED 5–6; “high”). The employment status was classified into 4 classes: “working, including students”, “unemployed” (designating people getting unemployment allocations), “retired”, and “not working, other”. This last group, although quite heterogeneous with respect to the reasons for non-working, contains, in men, a large proportion of people with health problems, which probably reflect a health selection in the labor market (people in good health are more prone to work). The housing status was based on information regarding tenure status and housing quality. This variable consisted of six categories (low-, mid- and high-comfort tenants and low-, mid- and high comfort owners) and was measured at the household level [24, 25]. It can be considered as a good proxy for wealth.

To compare SE inequalities by region, we focused on educational level only. Educational attainment is a relatively stable measure of SE position, as usually achieved early in adulthood, and is usually of rather good quality [26].

Causes of death were classified according to the International Classification of Diseases (ICD), version 10 [27]. All-cause premature mortality was divided into two categories (avoidable and non-avoidable mortality) according to the recent UK Office of Statistics definition of avoidable mortality [28] also adopted by Eurostat [29]. In addition, we divided the total premature mortality in four broad groups of causes of deaths (circulatory diseases, cancers, other natural causes of deaths and external causes), and further analyzed 12 avoidable causes of death with a high burden in Belgian society (lung cancer, lip, oral cavity and pharynx cancer, colorectal cancer, liver cancer, ischemic heart diseases (IHD), cerebrovascular diseases that were grouped with hypertension (HTA) as usually recommended [30, 31], alcohol related deaths, diabetes, chronic obstructive pulmonary diseases (COPD), suicide, and transport accidents). In women, breast cancer was analyzed as well. The corresponding ICD10 codes are shown in Additional file 1: Table S2. Analyses were performed separately for men and women as they have very different mortality levels.

Table 1 reveals clear regional differences with respect to individual SE features in our study population. Brussels is characterized by a higher proportion of higher educated individuals compared with the two other regions, whereas Flanders shows a higher housing score than Wallonia and Brussels (with a smaller proportion of owners in Brussels) and a lower rate of unemployment. It is also noteworthy that the proportion of missing values for the SE variables is lowest in Flanders.

Figure 1 shows the total and the EL-specific ASMRs by region for the total Belgian population (detailed numbers in Additional file 1: Table S1, middle part). In the Walloon region, the ASMR was 54% higher than in Flanders among males and 40% higher among females, while in Brussels the ASMR was respectively 52% and 48% higher compared with Flanders. Similarly, all EL-specific ASMRs were higher in Wallonia and Brussels than in Flanders both among men and women, particularly in the lowest EL. In men, ASMRs were higher in Brussels than in Wallonia for the low and mid ELs, but lower than in the Walloon Region for the highest EL. In women, the ASMRs were higher in Brussels than in Wallonia for all ELs. Cause-specific ASMRs by region and regional differences are displayed in Additional file 1: Table S3. The respective effects of three individual SE variables on the regional effect were explored through several Poisson models and summarized in Table 2. Compared to the first basic model, containing region and age only, models 2a, 2b and 2c revealed that all three SE variables had an important and statistically significant effect on the mortality rates (RR between 1.5 and 3.2). In addition, the SE variables had a significant impact on regional mortality differences. This confounding effect varied by SE dimension and gender: 1) after adjusting for education (model 2a), the RR for Brussels versus Flanders increased slightly (which was expected given the higher proportion of the higher educated in Brussels); 2) adjusting for employment status (model 2b) reduced regional differences in males both in Brussels and the Walloon region compared with Flanders (the RR decreased from 1.54 to 1.40 and from 1.53 to 1.40 respectively), but slightly increased the regional RRs in women; 3) introducing the housing score (model 2c) substantially decreased (up to 50% in Brussels) the regional RRs both among men and women. Model 3 showed the combined effect of all three SE variables on regional differences, that all remained significant.

The ranking of the specific contribution of each detailed cause, by sex and region, is shown in Fig. 2. In men, the ranking of the main contributors was quite similar between regions, with lung cancers and ischemic heart diseases at the top. Alcohol-related deaths and mental/neurological diseases contributed more to the PAFs in Wallonia and Brussels than in Flanders, while transport accidents contributed much less to the PAF in Brussels than in the other regions.

In women, ischemic heart diseases ranked first in Wallonia, followed by lung cancer. In Flanders, they ranked equally. In Brussels, lung cancer ranked as first contributor to the PAF. Colorectal cancer ranked lower in Wallonia and Brussels than in Flanders.

Previous research has documented important mortality differences at the regional or district level in Belgium [10‐17] with consistently higher rates in Wallonia (especially in the poorest districts of the Hainaut province) and Brussels as compared to Flanders (with the lowest rates in the eastern districts of the province of Limburg) since World War 2. Until the turn of the century, the link between individual SE characteristics and mortality could only be investigated through ecological studies because of the lack of appropriate data at the individual level. The constitution, in the early 2000s, of a “National Mortality Database” [6, 20], aiming to perform a population-based mortality follow-up, finally allowed for the study of SE mortality differentials in Belgium. Several studies, first based on the 1991 census and later on the 2001 census, assessed the magnitude of the SE health gap (and its change over time) in terms of differences in life expectancy [5], health expectancy [7, 8], all-cause and cause-specific mortality [6, 9] in Belgium.

Results from individual studies about inequalities are difficult to compare with each other because they generally present variations in the design of the follow up, the age limits and the standard population. Moreover, by focusing on the Belgian population, our results are not comparable with studies that also included migrant populations. However, a recent European study included Belgium in cross-country comparisons of inequalities in mortality [1, 2]. This study used 2 years of follow up and focused on people aged 30–74 at entry, which is 10 years older than in our study. Major inequalities were revealed in the Eastern European countries (with RDs in men exceeding 1500 per 100,000 PY and RRs situated between 2.6 and 3.3), rather large inequalities in Northern Europe (with RDs in men between 500 and 700 per 100,000 PY and RRs around 2), an inhomogeneous pattern in Western Europe (RDs in men between 300 and 600 per 100,000 PY, RRs between 1.6 and 2.4), and lower inequalities in Southern Europe (RDs in men between 250 and 400 per 100,000 PY and RRs around 1.6). As compared to the other European countries, inequalities in Belgium could be qualified as moderately high, with a RD and RR in men respectively equal to 385 per 100,000 PY and 1.86, while the mean RD for all the countries was 636 (range: 234–1696) and the mean RR was 2.1 (range: 1.51–3.26).

Only few studies jointly analyzed the effects of place of residence (region, province, district) and SE characteristics. Deboosere et al. [17] examined the district-level patterns of all-cause mortality in the 1990s with and without adjustment for individual SE variables and concluded that individual SE characteristics accounted for half of the mortality risk excess in the poorer districts of the old industrial belt of the Walloon region. Van Hemelrijck et al. [19] found a weak effect of area-level unemployment and percentage of laborers on the sub-district mortality RR, in addition to the effect of individual SE characteristics.

The analysis of inequalities by region has however not yet been performed by specific cause of death.

Large premature mortality excesses were observed in Brussels and Wallonia as compared to Flanders, which is in line with previous geographical mortality studies in Belgium. Our focus on people of Belgian origin increased the magnitude of those mortality excesses even further compared to studies including all residents, which is expected given both the mortality advantage in first generation migrants [21, 22, 42] and their higher representation in Wallonia and particularly in Brussels. In line with previous findings of Deboosere et al. in the 1990s [18], our analysis showed significant regional differences for premature mortality at the global level and within each level of education. The results of the Poisson regression also revealed that the educational distributions do not explain at all the regional differences in mortality. The level of employment in men accounts for a small part of the regional differences, while up to half of the differences can be explained by the housing score, which is a proxy for wealth.

With respect to inequalities in all-cause mortality, the largest inequalities were observed in Brussels and the lowest in Flanders, independently of the three inequality indices used. Important regional differences in absolute inequalities (RDs) were observed, but relative differences (RRs and PAFs) were weak. This pattern was observed for all-cause as well as for most specific causes of premature mortality.

The ranking of the cause-specific PAFs revealed a higher health impact of inequalities in causes combining a high mortality rate and a high relative inequality, with lung cancer and ischemic heart disease on top for all regions and both sexes. Suicide, COPD and cerebrovascular diseases also ranked high. The ranking of the contribution of the specific causes of death showed few differences between the regions.

Our results clearly show that the mortality excess in Brussels and Wallonia is still persisting. While this health gap appears to be difficult to eliminate in the context of unequal economical background between the different regions, this study, as well as two previous ones, showed that individual SE variables accounted only for half of the regional differences. The distribution of poverty in the regions (approached by the housing score), and to a lower extent, the employment status (in men only) are the main individual SE factors that explaining half of the regional difference, while educational level does not account. Up to now, the residual regional effect on mortality has not yet been elucidated, and could involve several factors such as: other macroeconomic variables that could not be captured by the existing data, cultural habits leading to less healthy lifestyle in some regions, effects of indoor and outdoor pollution and/or differences in health policies or health care management. In a multilevel analysis, Van Hemelrijck et al. [19] showed small additional effects of two aggregated variables, the level of unemployment and the percentage of laborers in the district. This means that an important unexplained residual regional effect remains after adjusting for individual and some macro SE variables. Further studies should try to disentangle the respective roles of those other risk factors in order to support policies oriented to reducing the health gap.

Larger inequalities were observed in the regions that also had higher mortality rates, and particularly in Brussels. The magnitude of the regional differences in inequality differs when it is expressed in absolute (RDs) or relative (RR and PAFs) terms, which is quite expected since the ASMR varies greatly between the regions. Indeed, when a region has high mortality rates, larger RDs will be observed for a same RR between ELs. In this study, however, we also observed higher relative inequalities in the regions with higher mortality rates, which is less expected. Indeed, since high RRs are more easily observed when the denominator is small (low mortality rates), observing a higher RR in a region that also have high rates, as compared to a region with lower rates, can only be observed when inequalities are strong.

The more unfavorable situation of Belgians living in Brussels, with respect to both mortality rates and inequality as compared to Flanders and even to Wallonia, should be interpreted in the light of the particular situation of this region. Indeed, in contrast to the other regions, the Brussels region is actually a big city without rural/suburban areas. There is a well-known ‘town attraction’, where people with social problems tend to move towards big cities to search for solutions, leading to some poverty concentration.

The PAF decomposition in specific causes is of major interest for policy-making since it can help set priorities by addressing inequalities in the causes with the highest population level impact. Causes of death combining high mortality rates and high inequalities rank highest. It is important to note that the causes ranking highest for the population-level inequality impact are related to major risk factors and in some measure amenable to health promotion measures. The ranking varies little by region, with the exception of a higher ranking of the PAF for alcohol-related deaths in Brussels, together with a lower PAF for transport accidents, which is expected given its lower mortality rate. Policies addressing inequalities in smoking, obesity and cholesterol level, as well as the medical management of ischemic and cerebrovascular diseases can be recommended in all three regions; ideally, a comprehensive “national and regional” policy should be implemented. Particular attention should be paid to the health of the low educated people in Brussels.

For the first time, inequalities were comprehensively measured and compared at the regional level. In the Belgian situation, where many public health responsibilities are transferred to the regional level, the calculation of inequality indices for each region is highly relevant to support regional policies. Inequalities were analyzed both for all-cause and for cause-specific premature mortality, using various inequality indices, namely the absolute and relative RDs and the PAF. By taking into account the EL distribution, the PAF is probably the best population-based measure to estimate the impact of inequalities on the total population health. It is also the first time that the cause-specific mortality inequalities were expressed by region in terms of a contribution to the total PAF.

By combining national census data with mortality data, the study covered practically the complete Belgian population. The large size of the cohort ensures a good statistical power.

It is important to note that the conclusions only apply to the Belgian population. Migrants registered with another nationality at birth were not included. As they represented a substantial fraction of the population, especially in Brussels, and they also experience lower mortality rates than the Belgians, our findings should not be extrapolated to the whole population. This choice on focusing on Belgian people only was made in first instance to keep a certain homogeneity in the population under study, since migrants have various backgrounds and socio-economic levels. Secondly, as the proportion of migrants varies between regions, nationality may represent a confounding factor in regional comparisons. Health inequalities should also be specifically studied in migrants, taking into account their ethnic specificities.

By definition, people living in Belgium without being registered were also not part of the studied population.

The percentage of missing values for the EL indicator was rather low, with some disparities between regions. We did not impute the missing values as we cannot expect them to be missing at random. Instead, we just treated them as a separate category. Since the study population included people of Belgian origin only, the language was probably not expected to be a barrier to answer the census questionnaire (as it could be in migrants). People hospitalized or severely sick at the time of the census were very likely not to have answered the questionnaire. Also part of the homeless people could not receive the post-mailed questionnaire; this led to missing values for all SE variables for the more sick or the more deprived people. In the Poisson models estimating RRs for all-cause mortality, the cases with missing SES were treated as a separate category for all three SE variables; as expected, those cases appeared to experience higher mortality rates than all other categories. This poorer health status in non-respondents is in line with previous findings related to the 1991 census [5]. In the subsequent analysis, we used as pairwise measures the low-versus-high and mid-versus-high RR and RD, and did not make use of the missing-value group; this is likely to lead to an underestimation of the inequalities (conservative bias).

Some inaccuracy in cause of deaths codification exists, resulting in 8% of ill-defined codes in the considered age group [17], and in an underestimation of cause specific rates. This could possibly be more the case in lower SE classes (if less efforts are made to get an accurate diagnosis in less than in more educated people), resulting in underestimated cause-specific inequalities (conservative effect).

Age limits for ‘premature mortality’ vary in literature. Our study covered people aged 25–64 at baseline, allowing them to reach at maximum 69 years at the end of the follow-up period. Our findings cannot be generalized to other age groups, and any comparison with other studies has to take the age range in consideration.

We chose the old European population as a standard population- even if a new one is now proposed by Eurostat - since this had been used in 2014 and 2016 European comparisons [43, 44]. Of course, the standardized rates are sensitive to the standard population, but most probably the impact on the inequalities will not be very important.