Article Text

Original research
Socioeconomic position and cardiovascular mortality in 63 million adults from Brazil
  1. Poppy Alice Carson Mallinson1,
  2. Shammi Luhar1,2,
  3. Elizabeth Williamson1,
  4. Mauricio L Barreto3,4,
  5. Sanjay Kinra1
  1. 1 Faculty of Epidemiology and Population Health, London School of Hygiene & Tropical Medicine, London, UK
  2. 2 Department of Public Health and Primary Care, University of Cambridge, Cambridge, Cambridgeshire, UK
  3. 3 Centre for Data and Knowledge Integration for Health (CIDACS), Fiocruz Bahia, Salvador, Bahia, Brazil
  4. 4 Institute of Collective Health, Federal University of Bahia, Salvador, Bahia, Brazil
  1. Correspondence to Poppy Alice Carson Mallinson, Faculty of Epidemiology and Population Health, London School of Hygiene & Tropical Medicine, London WC1E 7HT, UK; poppy.mallinson1{at}lshtm.ac.uk

Abstract

Background It has been suggested that cardiovascular disease exhibits a ‘social cross-over’, from greater risk in higher socioeconomic groups to lower socioeconomic groups, on economic development, but robust evidence is lacking. We used standardised data to compare the social inequalities in cardiovascular mortality across states at varying levels of economic development in Brazil.

Methods We used national census and mortality data from 2010. We used age-adjusted multilevel Poisson regression to estimate the association between educational status and cardiovascular mortality by state-level economic development (assessed by quintiles of Human Development Index).

Results In 2010, there were 185 383 cardiovascular deaths among 62.5 million adults whose data were analysed. The age-adjusted cardiovascular mortality rate ratio for women with <8 years of education (compared with 8+ years) was 3.75 (95% CI 3.29 to 4.28) in the least developed one-fifth of states and 2.84 (95% CI 2.75 to 2.92) in the most developed one-fifth of states (p value for linear trend=0.002). Among men, corresponding rate ratios were 2.53 (95% CI 2.32 to 2.77) and 2.26 (95% CI 2.20 to 2.31), respectively (p value=0.258). Associations were similar across subtypes of cardiovascular disease (ischaemic heart disease and stroke) and robust to the size of geographical unit used for analysis.

Conclusions Our results do not support a ‘social crossover’ in cardiovascular mortality on economic development. Our analyses, based on a large standardised dataset from a country that is currently experiencing economic transition, provide strong evidence that low socioeconomic groups experience the highest risk of cardiovascular disease, irrespective of the stage of national economic development.

  • coronary artery disease
  • epidemiology
  • global health
  • stroke
  • cardiac risk factors and prevention

Data availability statement

Data used in this study are made available online by the Brazilian Government. Mortality data were obtained from Ministry of Health’s DATASUS website (http://www2.datasus.gov.br/DATASUS/), census data for denominators were obtained from the Instituto Brasileiro de Geografia e Estatística website (https://www.ibge.gov.br/) and Human Development Index data were obtained from the Atlas do Desenvolvimento Humano no Brasil website (http://atlasbrasil.org.br/).

Statistics from Altmetric.com

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Introduction

Research suggests that cardiovascular disease (CVD) exhibits a social crossover, from greater risk in higher socioeconomic groups to lower socioeconomic groups, on economic development.1–4 However, historical evidence supporting this hypothesis from high-income countries (HICs) has been called into question,5 6 while comparisons of countries at different levels of economic development have reported inconsistent findings, potentially due to systematic differences in data sources and quality.7–10 Linked temporal data on social position and cause-specific mortality are lacking for most low-income and middle-income countries (LMICs). Many highly populated LMICs that are currently experiencing economic transition exhibit considerable subnational variation in levels of economic development, which could be used as counterfactuals to address this hypothesis using cross-sectional data.

Brazil is middle-income country of 200 million people, with levels of economic development ranging from that of some lower middle-income to upper middle-income countries across its 26 states. CVD is the leading cause of death in Brazil, with trends comparable with other LMICs.11 Brazil’s standardised national data and universal healthcare provides a unique opportunity to simultaneously examine the association between socioeconomic position and cardiovascular mortality across different levels of economic development, while minimising potential bias from other factors that vary between countries. We used national census and mortality data to assess the association between education and cardiovascular mortality between the economically diverse states of Brazil, hypothesising a positive association between socioeconomic position and cardiovascular mortality in less developed states and the opposite association in more developed states, consistent with a change in the association on economic development.

Methods

We conducted a cross-sectional study using routinely collected mortality and census data, which are available online on Brazilian government websites.12 13 Data on age, sex, education and municipality of residence were available from both data sources. Deaths are notifiable in Brazil, with coverage of mortality registration estimated to be over 95% in 2010.14 For both data sources, we used data from 2010, as this was the most recent year for which accurate census data were available.15

We identified cardiovascular deaths using International Classification of Diseases 10th revision (ICD-10) codes I00-99. We also examined CVD subtypes ischaemic heart disease (IHD) and stroke (using ICD-10 codes I20–25 and I60–69, respectively), as literature suggests that the social crossover between socioeconomic position and cardiovascular mortality is driven largely by IHD.16 We examined overall, as well as premature, cardiovascular mortality (defined as deaths under age 70 years), as recommended in the WHO’s Non-Communicable Disease Global Monitoring Framework.

Socioeconomic position was defined by number of completed years of education. Education captures multiple dimensions of socioeconomic position relevant to CVD, including relative social status, earning potential and health-related knowledge.17 Education in Brazil follows the same grading system across all states as it is delivered federally. In 2010, the education categories on death certificates were: 0, 1–3, 4–7, 8–11 and 12 or more years of education completed, and on census were: none, creche/preschool, elementary school, high school (and subcategories of these), graduate, masters and doctorate. To minimise potential misclassification of education between the two data sources, we categorised education into two broad groups (<8 vs 8+ years of education completed, corresponding to did not vs did complete elementary school).

We classified state-level economic development using Human Development Index (HDI), a composite measure combining education, health and economic productivity as it captures a broader range of economic indicators than solely income-based metrics.

Statistical analyses

The final aggregated dataset contained, for every unique combination of variables, data on the person-years at risk in 2010 (approximated by midyear population size) and the number of CVD deaths in 2010.

We restricted the study population to adults aged over 20 years, as CVD rarely affects children and adolescents. As the quality of mortality reporting varies across Brazil,14 we derived an indicator for high coverage of mortality registration for each municipality by comparing the number of deaths reported in the 2010 census18 to the number of deaths in the mortality registry over the same period. We included municipalities for which agreement between deaths in the mortality registry and census was of a comparable level to states known to have near-complete mortality registration (ratio of registry to census deaths of ≥1.08). To assess the sensitivity of results to this choice of indicator, we also repeated the analyses using data from all municipalities. As three states (Roraima, Acre and Amapá) were too small to be analysed individually, they were merged with their most similar neighbour, resulting in 23 geographical units (hereafter referred to as states). We also repeated the analyses stratifying by the five regions of Brazil (North, Northeast, Centre-West, South and Southeast).

We modelled cardiovascular mortality count using age-adjusted multilevel Poisson regressions to compare mortality between people with <8 versus 8+ years of education. To capture the non-linear relationship between age and mortality, in addition to age, we adjusted for age-squared and age-cubed, both of which significantly improved the model fit. Person-years at risk was included as an offset term. We included state-level and municipality-level random intercepts to account for geographical clustering of cardiovascular mortality and additionally allowed the association with education to vary randomly across states. To assess for systematic variation in the association between education and cardiovascular mortality by state-level HDI, we included an interaction term between these variables, with HDI considered as a categorical (splitting the 23 states into quintiles according to their HDI) and then a continuous measure. Interaction terms for gender were also included in all models.

We used multiple imputation with chained equations to minimise the potential for bias due to missing data on education (~20%) and ill-defined causes of death (~8% of deaths, considering ICD-10 garbage codes R00–R99 to be ‘missing’) in the mortality registry (see online supplemental methods S1).

Ethical approval was not obtained as this study used only deidentified secondary data available in the public domain.

Results

In 2010, there were 185 383 reported cardiovascular deaths among 62 568 055 adults included in the study (49% of all Brazilian adults). Our primary analyses were restricted to municipalities with high mortality registration (2004 out of 5565 municipalities). The excluded population were more likely to live in the less economically developed regions of Brazil and were less likely to be educated. Cardiovascular mortality was lower among the excluded population, as expected due to the lower mortality registration in excluded municipalities. Sizes of the included and excluded population, and differences between them are given in the online supplemental tables S1 and S2.

Supplemental material

The crude cardiovascular mortality rate in Brazil was 296 per 100 000. After adjusting for age, cardiovascular mortality was over twice as high in men with <8 years’ education compared with 8+ years’ education and nearly three times higher in women with <8 years’ education compared with 8+ years’ education (table 1).

Table 1

Description of the included sample and age-adjusted associations between education and cardiovascular mortality in Brazil, 2010

HDI of the states ranged from 0.63 to 0.78. We observed an inverse association between education and cardiovascular mortality in states of Brazil across all levels of economic development but with considerable subnational variation (figure 1). Among women, higher state-level HDI was associated with a lower rate ratio for education (p value for log-linear trend=0.002), whereas this trend was less pronounced among men (p=0.258) (table 2). For example, the age-adjusted cardiovascular mortality rate ratio for women with <8 years of education (compared with 8+ years) was 3.75 (95% CI: 3.29 to 4.28) in the least developed quintile of states and 2.84 (95% CI 2.75 to 2.92) in the most developed quintile. The age-adjusted cardiovascular mortality rate ratio for men with <8 years of education (compared with 8+ years) was 2.53 (95% CI 2.32 to 2.77) in the least developed quintile and 2.26 (95% CI 2.20 to 2.31) in the most developed quintile (table 2). For IHD and stroke mortality, we observed an inverse association with educational status in all states, and among women, higher state-level HDI was associated with lower rate ratios (although CIs suggested no linear trend by HDI). For IHD among men, there was some suggestion of a non-linear association, with the highest rate ratios for education observed in the lowest and highest HDI states. Restricting the analyses to ‘premature’ cardiovascular deaths (age <70 years) did not substantively alter the results.

Figure 1

Rate ratios for association between education (<8 vs 8+ years) and cardiovascular mortality by quintile of state Human Development Index (HDI) in Brazil, 2010. CVD, cardiovascular disease; IHD, ischaemic heart disease.

Table 2

Associations between education and cardiovascular mortality stratified by quintile of state-level Human Development Index (HDI) in Brazil, 2010

As a sensitivity analysis, we repeated the analyses using all municipalities (n=5565, population=127 826 740) and found similar results (online supplemental table S3). Our results were also robust to use of region (n=5) instead of state as the geographical unit of analysis (online supplemental table S4).

Discussion

We found that lower educational status was associated with higher cardiovascular mortality in both more and less economically developed states of Brazil. Among women, these social inequalities in cardiovascular mortality were larger in the less economically developed states. Associations were robust to use of different geographical units of analysis and different inclusion criteria for data quality. Thus, we find no evidence to support a ‘social crossover’ in CVD mortality, from greater risk in higher socioeconomic groups to greater risk in lower socioeconomic groups, on economic development in a large middle-income country.

The phenomenon of a ‘social crossover’ in CVD on economic development is frequently cited in the global health literature.1–4 Historical data from the UK demonstrated a change in the association between occupational social class and non-valvular heart disease mortality from positive to inverse between 1951 and 1971.5 However, subsequent reanalysis of these data suggested the apparent ‘crossover’ may have been an artefact of bias in mortality coding practices (deaths among low socioeconomic groups were less likely to be attributed to IHD).6 A review of studies from the UK and USA between 1930 and 1960, which measured IHD objectively, found a null or inverse association in 11 out of 12 studies.6 An ecological study from Hong Kong reported a change in the association between neighbourhood median income and cardiovascular mortality from positive to inverse between 1976 and 1995, but this ‘crossover’ is unlikely to be attributable to economic development as Hong Kong was relatively economically advanced throughout this period.19 Long-term data from a demographic surveillance site in Bangladesh reported that household wealth was positively associated with IHD mortality in 1983, and not associated in 2005, although the 10-fold increase in IHD mortality between surveys (presumably due to the high proportion of unassigned deaths in earlier surveys) suggests that these data could not be reliably compared over time.20

By comparing areas at varying levels of economic development at one point in time to investigate a ‘social crossover’ in cardiovascular mortality on economic development, we avoided potential biases related to systematic changes in cardiovascular mortality coding practices and data quality over time. A few other studies have compared social inequalities in CVD across countries at different levels of economic development and generally found an inverse association in HICs as well as LMICs.7–10 However, all of these studies except one10 combined LMICs into a single category and did not examine variation in the association by stage of economic development, which could have provided more relevant evidence for the ‘social crossover’ hypothesis given that the inverse association in HICs is well established.16 A recent prospective study of 20 low-, middle- and high-income countries found that inverse associations between socioeconomic position and incident CVD were stronger in low-income than middle-income countries, consistent with our findings and inconsistent with the ‘social crossover’ hypothesis.10

Explanations for a ‘social crossover’ in CVD generally refer to changes in the socioeconomic patterning of cardiovascular risk behaviours. An important factor in HICs may have been the shift in smoking from high to low socioeconomic groups around mid-20th century,5 16 although there is no evidence to support a similar shift in LMICs.21 More relevant to LMICs may be the replacement of occupational physical activity among low socioeconomic groups in preindustrialised economies by leisure-based physical activity among high socioeconomic groups in industrialised economies, and the increasing availability of low-quality, calorie-dense diets.5 Accordingly, many middle-income countries are experiencing dramatic increases in obesity prevalence, especially among low socioeconomic groups.22 While it seems plausible that a ‘social crossover’ in diet, activity and obesity might explain a similar ‘crossover’ in cardiovascular mortality, data from LMICs do not support this. For example, in India, socioeconomic position is associated positively with overweight/obesity23 but inversely with cardiovascular mortality10; if a ‘social crossover’ in cardiovascular mortality had already occurred in India, it would have preceded a ‘social crossover’ in obesity and thus could not have been caused by it. However, it is possible that an earlier ‘social crossover’ in obesity among women compared with men, observed in many countries including Brazil, contributes to the gender differences in inequalities in cardiovascular mortality that we observed.22 24

The lack of plausible mechanisms to account for a ‘social crossover’ in cardiovascular mortality is consistent with our findings from Brazil of greater cardiovascular mortality in low socioeconomic groups, irrespective of state-level economic development. Reasons for greater cardiovascular mortality in low socioeconomic groups in LMICs may be similar to those in HICs, for example, reduced access to health services,10 exposure to harmful environmental conditions (eg, air pollution and occupational hazards), increased smoking rates21 and diet lower in fruits and vegetables.10 Novel and unestablished risk factors, such as early life undernutrition, frequent infections and psychosocial adversity, could play an important role in LMICs, warranting further research in these settings.

We used an innovative study design to circumvent the lack of high-quality temporal data on social inequalities in cardiovascular mortality from LMICs. The large population sizes (ranging from 1.5 to 45 million) and diversity in economic development (encompassing levels of HDI comparable to lower middle-income (eg, Bangladesh and Honduras) and upper middle-income (eg, Mexico and Albania) countries) in Brazilian states allowed us to use them as counterfactuals to investigate the change in social inequalities on economic development, while controlling for bias from other factors that vary between countries or within countries over time (eg, data quality and health system factors). Furthermore, our use of national routine data allowed us to precisely estimate inequalities across a large number of geographical units.

However, the study also has some limitations. We used a single transitioning country to test a generalisable hypothesis, but each country is likely to have its own setting-specific factors that modify trends in social inequalities in CVD.25 For example, in Brazil, there was an expansion of primary care services and other progressive social reforms from the mid-1990s onwards, which may have resulted in an underestimation of the inverse associations noted in even the least developed states.26 The non-linear trends in social inequalities we observed among men for some outcomes warrant further exploration, as they suggest a potential role for setting-specific differences between the states. Furthermore, the Brazilian states included in our analyses were comparable with lower and upper middle income countries, but the results may not be generalisable to countries outside this range of economic development. Supportive data from countries at extremely low levels of economic development are needed.

The coverage and quality of mortality reporting was variable across states, which could have introduced some potential for bias. First, there was more under-registration of deaths in the lower HDI states.27 To minimise the potential bias arising from this, we restricted our analyses to municipalities with high mortality registration using a robust indicator based on an independent data source.18 The sociodemographic characteristics of the included and excluded populations were similar, and the results hardly changed when we included the whole population in sensitivity analyses (online supplemental tables S2 and S3). Second, a greater proportion of causes of deaths were ill-defined in the lower HDI states. Although we used multiple imputation to account for this, multiple imputation assumes that the data were missing at random, which may not have been the case. In Brazil, ill-defined causes of death are more common in populations with reduced access to medical facilities (eg, rural populations), while CVDs are over-represented among the ill-defined causes of death.27 28 However, we expect that any residual bias due to mortality under-reporting or ill-defined causes of death would underestimate the inequalities in cardiovascular mortality in lower HDI states, suggesting that, if anything, our conclusions are likely to be conservative.

It is possible that there was some misclassification of educational status in the census or, more likely, on the death certificates (which are usually completed by the attending physician). This could potentially introduce numerator–denominator bias, a bias that arises when covariates used to link numerators to denominators are misclassified. The extent or direction of numerator–denominator bias is difficult to predict; however, the social inequalities for IHD mortality in our study were comparable with those for prevalent disease reported in another nationally representative survey, providing face validity to our results.29 Our use of a binary categorisation for education will have helped to reduce misclassification and achieve a more valid match between education categories on the death certificate and census. Furthermore, education has been shown to be less prone to numerator–denominator bias compared with other socioeconomic indicators such as occupation and income and does not change with age among older adults, making it particularly suitable for adult mortality studies.17 30

Another potential limitation is that we used HDI to classify the level of development of the states, which incorporates life expectancy and thus may be influenced by the outcome of cardiovascular mortality. Lower HDI states will tend to have higher mortality rates (including cardiovascular), leading to smaller relative inequalities in mortality in these states.11 Similar to the other potential biases discussed above, this may have led to an underestimation of social inequalities in cardiovascular mortality in lower HDI states, suggesting our conclusions could be conservative.

The belief in a ‘social crossover’ in CVD mortality on economic development, from greater risk in high socioeconomic groups to greater risk in low socioeconomic groups, is not supported by evidence from LMICs. We urge caution when referring to this phenomenon, as doing so risks propagating a misrepresentation of social inequalities in CVD in LMICs. Our analysis provides strong evidence that low socioeconomic groups experience the highest cardiovascular mortality irrespective of stage of economic development. There is a need for high-quality prospective data from LMICs to describe and elucidate mechanisms of these social inequalities in CVD. Policies for the control of CVD and its risk factors in LMICs must be inclusive of low socioeconomic groups.

Key messages

What is already known on this subject?

  • It is often assumed that socioeconomic inequalities in cardiovascular mortality exhibit a ‘crossover’, from greater risk in high socioeconomic groups, to greater risk in low socioeconomic groups, on economic development; however, there is limited evidence to support this claim.

What might this study add?

  • Our data comparing states of Brazil at different levels of economic development suggest that lower socioeconomic groups may experience the highest cardiovascular mortality irrespective of a country’s stage of economic development.

How might this impact on clinical practice?

  • The belief that low socioeconomic groups are initially at less risk of cardiovascular mortality compared with high socioeconomic groups is not substantiated and risks preventing appropriate targeting of policies in low-income and middle-income countries, which could further exacerbate inequalities.

Data availability statement

Data used in this study are made available online by the Brazilian Government. Mortality data were obtained from Ministry of Health’s DATASUS website (http://www2.datasus.gov.br/DATASUS/), census data for denominators were obtained from the Instituto Brasileiro de Geografia e Estatística website (https://www.ibge.gov.br/) and Human Development Index data were obtained from the Atlas do Desenvolvimento Humano no Brasil website (http://atlasbrasil.org.br/).

References

Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

Footnotes

  • Contributors PACM and SK conceived the study. PACM conducted data management and analyses, and wrote the first draft. EW provided statistical support. SL, EW, MLB and SK reviewed and commented on all drafts of the manuscript.

  • Funding This work was funded by Medical Research Council UK (grant number: MR/N013638/1) through a studentship to PACM.

  • Competing interests None declared.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.

Linked Articles