Article Text
PDF
Abstract
Background A negative socioeconomic gradient is established for coronary heart disease (CHD) mortality and survival, while socioeconomic patterning of disease incidence is less well investigated. To study socioeconomic inequalities in the incidence of acute myocardial infarction (AMI), the major component of CHD, a meta-analysis was undertaken to summarise existing evidence on the issue.
Methods A systematic search was performed in PubMed and EMBASE databases for observational studies on AMI incidence and socioeconomic position (SEP), published in English to April 2009. A random-effects model was used to pool the risks estimates from the individual studies.
Results Among 1181 references, 70 studies fulfilled the inclusion criteria. An overall increased risk of AMI among the lowest SEP was found for all three indicators: income (pooled RR 1.71, 95% CI 1.43 to 2.05), occupation (pooled RR 1.35, 95% CI 1.19 to 1.53) and education (pooled RR 1.34, 95% CI 1.22 to 1.47). The strongest associations were seen in high-income countries such as USA/Canada and Europe, while the results were inconsistent for middle and low-income regions.
Conclusion AMI incidence is associated with low SEP. The nature of social stratification at the level of economic development of a country could be involved in the differences of risk of AMI between social groups.
- Acute myocardial infarction
- epidemiology ME
- ischaemic heart disease
- meta-analysis
- social inequalities
- socioeconomic position
Statistics from Altmetric.com
- Acute myocardial infarction
- epidemiology ME
- ischaemic heart disease
- meta-analysis
- social inequalities
- socioeconomic position
Acute myocardial infarction (AMI) is a major cause of death and disability worldwide.1 According to an estimate from the WHO, approximately 40% of all deaths will be related to cardiovascular diseases (CVD) in 2020, with AMI being the main single cause.2 A projection of global mortality and burden of disease for 2030 predicted that despite the decline in mortality due to coronary heart disease (CHD) in high-income countries during past decades, it will remain the leading cause of mortality in low and middle-income countries.3–5 Due to the high mortality rate, severe damage to physical and mental health, long rehabilitation periods and a high rate of disability, the burden of AMI is a highly significant social issue.6
There is a considerable body of evidence linking socioeconomic position (SEP) with the conventional risk factors for CHD, including AMI as the major component of CHD.7–9 Lower SEP is often associated with health-damaging lifestyle resulting in the development of poor dietary habits as well as influencing behaviours related to smoking and physical activity.10 Individuals with low SEP are prone to be exposed to multiple risk factors and, therefore, seem to have a dramatically excessive burden of disease.11
Associations between SEP and CVD mortality and survival have been well discussed in a number of reviews.12–16 However, quantitative assessment of socioeconomic patterning of CVD and AMI incidence is presented to a lesser extent. The use of incidence data generally avoids problems with post-diagnosis SEP changes as well as with different survival by socioeconomic groups.17 Addressing the issue of social inequality in AMI further, it is important to estimate the individual contribution of each SEP indicator rather than interchangeable SEP measures as they affect health through different pathways and causal mechanisms.18 We undertook a meta-analysis as a quantitative approach to summarise the existing evidence on the issue19 20 to investigate the association between AMI incidence and various SEP measures, including educational attainment, income and occupation categories.
Subjects and methods
Search strategy
To identify eligible studies of associations between socioeconomic determinants and AMI incidence, we conducted a systematic search in PubMed and EMBASE databases for articles published in English-speaking, peer-reviewed journals from 1966 to 1 April 2009. For this search, we used the relevant medical subject heading (MeSH) terms and key words related to socioeconomic determinants combined with specific outcome defined as ‘acute myocardial infarction’. The details of the electronic search are reported in supplementary appendix 1, available online only. The reference lists were scrutinised to identify additional studies.
Study selection
To be included in this meta-analysis, studies had to use original data, be designed as case–control or cohort studies, consider AMI incidence as an outcome, present risk estimates with 95% CI on the association between incident cases of AMI and at least one measure of SEP, or report sufficient information to compute these for men, women or both.
The articles were selected for inclusion if the study event was originally defined as AMI either non-fatal or in combination with fatal or reported as a composite outcome of AMI and death attributed to CHD/ischaemic heart disease, including sudden death. We also considered studies in which the outcome definition comprised AMI with congestive heart failure21 22 or unstable angina.23 In order to avoid the inclusion of chronic ischaemic disease, we did not include studies if the outcome of interest was presented as a CHD event corresponding to codes 410–414 in the International Classification of Disease, version 9 or I21–I25 in International Classification of Disease, version 10 with no further specification. Neither did we include studies if AMI was combined with angina pectoris, coronary atherosclerosis or stroke with no possibility to extract data on AMI alone. We sought data on the first AMI event to assess the impact of social inequality on the development of disease in people apparently free from acute ischaemic disease. Studies focussing exclusively on mortality and survival were not included.
SEP indicators were included only if they were based on income, educational attainment or occupational categories. Studies utilising a social indicator constructed as a combination of two or more standard socioeconomic indicators were not included. Neither did we include studies in which the SEP measure was based on ownership of car/houses/health insurance or was presented as categories of deprivation. Only studies with an adult individual-level measure of SEP were included. No restrictions were made by the type of SEP, personal or household.
When overlap was identified from various studies, the original data were included only once, prioritising datasets providing maximally adjusted risk estimates. If not, we used the most up-to-date information or studies with greater numbers of participants. Two co-authors (EM-G, AS) independently extracted relevant studies following the inclusion criteria. In the case of missing data we contacted the corresponding authors. All authors24–29 from whom the additional information was requested provided us with the data we asked for. Discrepancies were resolved by consensus in a panel meeting (TM, EM-G, AS).
Data extraction
The following information was extracted from each publication: the first author's last name, the year of publication, the country where the study was performed, study design, years of data collection and type of controls (population-based/hospital-based) in case–control studies, duration of follow-up in cohort studies, the sample size, indicators of SEP, source and type of SEP data, definition and status of the outcome, status of event, sex and age along with the risk estimates for AMI associated with SEP with corresponding 95% CI, and the variables controlled for. The information on country where the study was performed was then classified both according to the geographical area (USA/Canada, Europe, Asia, Latin America, Middle East) and country's income level (high, middle, or low-income countries).5 From each study, we extracted the risk estimate both minimally and maximally adjusted for the potential confounders. Unless otherwise stated, we included in the analysis the maximally adjusted estimates in order to overcome inconsistency in handling confounding and mediating variables. We considered the study risk estimates to be minimally adjusted if unadjusted or adjusted for age, sex and residence, either one or all, by matching, restriction, stratification or statistical adjustment, and to be maximally adjusted if in addition they were adjusted for any of the classic well-recognised AMI-specific risk factors. If the original study reported risk estimates in association with more than one measure of SEP, each estimate was extracted and then analysed as its own association with the specific SEP.
Statistical analysis
RR was used as the measure for the summary statistic of associations between SEP and AMI incidence. To simplify the procedure, RR represented all reported study-specific results derived from cohort studies and OR from case–control studies.30 Due to the initial assumption of between-study heterogeneity, a random-effects model of DerSimonian–Laird,31 which incorporates both within and between-study variability, was applied to combine log RR across the studies. To augment comparability between the studies using different SEP categories, we compared the lowest with the highest SEP category. If the original study reported the risk estimates not in this order, we back-calculated the point estimate and 95% CI. For the articles that did not report estimates in the form of RR or OR, the risk estimates and 95% CI were recalculated from the presented raw data by using standard equations. If the original study reported separate RR for different sexes32–37 or different races34 or if RR were reported separately for two subcohorts with rheumatoid arthritis and non-inflammatory rheumatoid disorders within same cohort,38 the risk estimates were pooled (weighted by the inverse of their variance) to obtain a single estimate per SEP from each study.
To evaluate the statistical heterogeneity among studies we used Cochran's Q test.39 This test examines the null hypothesis that difference between the study estimates of RR is due to chance by using a χ2 test with degrees of freedom equal to the number of studies minus one. For the Q statistic we considered p<0.1 as representative of statistically significant heterogeneity. To describe the proportion of total variation in study estimates explained by heterogeneity rather than chance, we calculated the I2 statistic.40
Random-effect meta-regression analyses were performed to identify study-level factors contributing to heterogeneity between studies. The explanatory variables included study design, geographical area, country's income level, publication year (before 2000/2000 and after), type of adjustment with respect to AMI-specific risk factors, type of controls in case–control studies, personal or household SEP, and status of the event and the natural logarithm of RR was the dependent variable.41–43 A univariate meta-regression was performed for each study-level factor for studies on income, education and occupation. A backward stepwise approach was used to select the significant variables to be included in a multivariate analysis. In addition, a series of subgroup analyses was conducted by stratifying the original studies by sex, country's income level and geographical region, study design, adjustment strategy, publication period, personal or household SEP and type of study event. The stability of the results was also evaluated in leave-one-out sensitivity analysis in which the influence of the individual study on the overall pooled RR was estimated by omitting one study at a time.44
We assessed publication bias by constructing funnel plots and using the Egger's regression asymmetry test and the Begg–Mazumdar adjusted rank correlation test.45 46 All statistical analyses were performed using STATA version 10. p Values that were less than 0.05 were considered statistically significant. All statistical tests were two-sided.
Results
Study characteristics
A total search in the electronic databases revealed 2539 references and among those 1358 were overlapping between different search categories. The search strategy for the 1181 unique references is presented in figure 1 as the QUOROM statement47 flowchart in which the detailed procedure of the reference identification along with the information on exclusion criteria applied on different stages of the selection is described. Briefly, 855 articles did not address the issue of interest and were excluded after screening the abstracts, leaving us with 326 full-text articles for further examination. Of these, only 64 articles fulfilled the predefined inclusion criteria and were selected to be included in the analysis. The reference lists of the selected articles were scrutinised and 19 articles were additionally identified fulfilling the inclusion criteria. To mitigate overlapping in study populations, several studies being initially considered relevant for the analysis24 26 48–63 were excluded and substituted by more recent studies, providing maximally adjusted risk estimates or with greater numbers of participants.18 23 33 35 64–71
The QUOROMa statement47 flowchart for study selection. aQuality of reports of meta-analyses of randomised controlled trials. AMI, acute myocardial infarction; CHD, coronary heart disease; CVD, cardiovascular disease; SEP, socioeconomic position.
Two risk estimates for educational attainment and AMI incidence presented by Chang et al66 for women from eastern Europe and whose from non-European countries (Latin America, Asia and Africa) were independently included in the analysis, referred to as Chang C (A) and Chang C (B), respectively. Similar to that, two substudies on occupational SEP and AMI described in the article by Mattila et al27 (series I and series II) were also analysed separately. The same strategy was applied to data derived from the articles by Eaker et al72 and Qureshi et al,73 in which occupational and educational SEP were presented among women at the age of 45–54 years (Eaker E (A)) and 55–64 years (Eaker E (B)) and patients younger than 50 years (Qureshi A (A)) and older than 50 years (Qureshi A (B)), respectively. Substudies from the articles by Bosma et al74 originated from Lithuania (Bosma H (A)) and The Netherlands (Bosma H(B)) were included independently. Therefore, 70 original studies from 65 articles, extending back to the year 1968, were finally included in the analysis.
In total, there were 37 case–control studies in 35 articles23 25 27 28 32 33 35 65–67 75–99 reporting associations with different SEP among 74 056 AMI cases and 619 652 controls and 33 cohort studies, including two studies nested in cohort, in 30 articles18 21 29 34 36–38 64 68–74 100–114 in which association with SEP was studied for 28 629 incident cases among 3 869 270 participants. Supplementary table S1 (available online only) presents data on detailed study characteristics of the included studies.
Overall result
The overall results of this meta-analysis provided evidence of a significant increase in the risk of AMI among the lowest socioeconomic categories for all three socioeconomic indicators (figures 2–4). Heterogeneity was observed for all three SEP indicators (p<0.001) (table 1).
RR and 95% CI for acute myocardial infarction incidence and income categories (the lowest vs the highest socioeconomic position category) in individual studies and for all studies combined. RR from the individual studies are indicated by squares and the size of the squares represents the statistical weight that each study contributed to the random-effect summary estimate. Horizontal lines indicate the study-specific 95% CI. Diamond represents the overall summary RR and its 95% CI.
RR and 95% CI for acute myocardial infarction incidence and educational attainment (the lowest vs the highest socioeconomic position category) in individual studies and for all studies combined. RR from the individual studies are indicated by squares and the size of the squares represents the statistical weight that each study contributed to the random-effect summary estimate. Horizontal lines indicate the study-specific 95% CI. Diamond represents the overall summary RR and its 95% CI.
RR and 95% CI for acute myocardial infarction incidence and occupational categories (the lowest vs the highest socioeconomic position category) in individual studies and for all studies combined. RR from the individual studies are indicated by squares and the size of the squares represents the statistical weight that each study contributed to the random-effect summary estimate. Horizontal lines indicate the study-specific 95% CI. Diamond represents the overall summary RR and its 95% CI.
Pooled estimates for the lowest versus the highest socioeconomic category and incidence of AMI in series of subgroup analyses
The strongest pattern was seen for the lowest income group in which the incidence of AMI increased by 71% compared with the high income group (RR 1.71, 95% CI 1.43 to 2.05; table 1). Further stratification by sex, adjustment strategy, study design, status of event, personal or household type of SEP or publication years did not alter the overall pooled results. No association was seen for the case–control studies utilising hospital controls, while studies with population controls revealed a statistically significant association between the lowest income group and AMI (table 1).
We observed a 34% increased risk of AMI for the lowest educational group (RR 1.34, 95% CI 1.22 to 1.47; table 1). The increase was apparent in subanalyses after stratifying by the main study characteristics, ie, sex, adjustment strategy, study design, publication period and status of event, apart from for the case–control studies with hospital controls (table 1).
An increased incidence of AMI was observed when studies on occupational categories were pooled (RR 1.35, 95% CI 1.19 to 1.53; table 1). The increase persisted for the lowest occupational SEP when pooling studies within subgroups with different sex, status of event and publication years. A non-significant increase was seen in subanalysis for household SEP measure (husband's occupation for women). The results were less consistent among case–control studies (table 1).
The increased incidence of AMI was evident for the lowest income-based, educational and occupational SEP in high-income countries and in regional areas such us the USA/Canada and Europe (table 1). No significant associations were, however, apparent between any of the SEP determinants and AMI incidence in studies carried out in the middle or low-income countries, particularly in the Asian region. In contrary, for middle or low-income countries an inverse association was observed for the results combined across the studies on occupational SEP.
There was a substantial heterogeneity in overall results among the studies on all SEP determinants (p<0.001) that remained significant for the results from most of the subanalyses. However, no or low heterogeneity was present when analyses were restricted to studies originated from the USA/Canada. There was a low heterogeneity among studies on occupation when results were combined for middle or low-income countries. A similar pattern was seen when studies on income published before the year 2000 were analysed separately.
In random-effect meta-regression analyses the relation between SEP and AMI incidence persisted irrespective of the design of original studies, although in case–control studies multivariate regression indicated the association between the type of control (hospital vs population controls) and RR of AMI for studies on all SEP indicators (for income β coefficient (β)=−2.76, 95% CI −4.77 to −0.76, p=0.007; for education β=−0.40, 95% CI −0.75 to −0.05, p=0.02; for occupation β=−0.51, 95% CI −1.00 to −0.02, p=0.04). Pooling together the cohort studies and case–control studies with population controls only had no effect on the overall results (pooled RR for income 1.80, 95% CI 1.50 to 2.16; pooled RR for education 1.43, 95% CI 1.30 to 1.58; pooled RR for occupation 1.41, 95% CI 1.24 to 1.60). Type of AMI event (potentially first vs clearly first event ever) was significantly associated with outcome in multivariate regression analyses in studies on income (β=0.24, 95% CI 0.04 to 0.46, p=0.02) and occupation (β=0.54, 95% CI 0.27 to 0.82, p<0.001). Among other study-level factors only two revealed associations with the outcome. For studies on income SEP the publication period (published in year 2000 and after vs published before year 2000) reached statistical significance in univariate, but not in multivariate meta-regression analysis (data not shown). For occupational SEP in multivariate analysis country's income group (middle or low-income countries vs high-income countries) was associated with RR of AMI (β=−1.24, 95% CI −1.74 to −0.74, p<0.001). No other study-level characteristics for any of the studied SEP indicators were found significant in meta-regression analyses.
No publication bias was observed for educational SEP (Egger's test p=0.49) with borderline significant results for occupational SEP (p=0.05), while selection of the studies focussing on income introduced publication bias (p=0.03). The publication bias funnel plots are presented as supplementary material (figure S1), available online only. No individual studies significantly altered the overall estimates based on the results of the sensitivity analysis.
Discussion
Our results indicated an increased incidence of myocardial infarction among the lowest socioeconomic categories in income (71%), education (34%) and occupation (35%) compared with the highest category of the corresponding SEP. The associations were significant for both men and women and were consistent for most of the results from the subgroup analyses. The increased risk of AMI for the lowest categories of all SEP indicators was most evident in high-income countries, while middle or low-income countries revealed less consistent associations probably due to a limited number of the studies included in the latter strata.
SEP is a surrogate measure for numerous factors that may affect health. There is an increasing awareness that SEP indicators should not be used interchangeably because they may represent different risk factors115 and relate to different causal pathways.18 116 Recent reviews discussed the new approach for health researchers to study the socioeconomic inequalities in health by measuring the respective impact of separate socioeconomic indicators on health outcomes along with the mediating mechanisms and adjustment suggested by each indicator rather than studying an effect of a single composite SEP variable.18 117 118 In our analysis an attempt was made to identify the independent contribution of each SEP indicator to AMI outcome across regions and over time. The results must be interpreted with caution because of the observational nature of the data.19 119 120
Several methodological issues need to be taken into consideration. One of the limitations in our study was the possible differences in the definition and classification of SEP across studies. It was of particular importance for educational and income SEP categories because differences in countries' general economy, educational systems and cultural issues could cause variability in the scales used to classify the exposure. The methodological limitations for quantitative comparison of studies on SEP due to the substantial dissimilarity in exposure measures used in different studies over time and geographical regions have been discussed previously.115 A lack of uniformity in reporting SEP data in original studies has limited our ability to investigate a social gradient in AMI incidence. To reduce the inconsistency in SEP stratification across the original studies and to obtain a meaningful indicator of socioeconomic inequality in health, it has been suggested that the difference between the extreme categories of SEP should be measured, ie, to compare the highest and the lowest SEP strata.117 118 121 Studies selected for our analysis varied significantly in presenting SEP categories that prevented us from collapsing the middle-level SEP categories into one. To overcome the problem and to increase comparability across the studies, we applied the suggested approach and compared the extreme categories, which, however, impaired the possibility of studying the social gradient and could be considered as a limitation of the analysis.
As a result of significant heterogeneity observed across the selected studies, the pooled results should be interpreted with caution. The nature of observational studies introduces design-related heterogeneity that basically reflects the disparity in the study characteristics.122 In our case the heterogeneity was probably boosted by the initial unevenness in study bases as well as in outcome definitions and status of events used in the original studies, along with inconsistency in measures of exposure and variability in handling confounding and mediating factors. The incidental manner of reporting SEP data in descriptive tables in numerous studies resulted in using estimates adjusted neither by matching nor by statistical adjustment, and probably influenced the completeness of the search. The result-related heterogeneity122 may be less pronounced in our meta-analysis due to non-substantial variability between the original point estimates and considerable overlaps of the CI. Meta-regression analyses indicated the type of study controls in case–control studies as potential sources of heterogeneity between studies for all SEP indicators. Several meta-analyses discussed the limitations introduced by pooling the results of studies different in study design or type of control groups, as in hospital-based case–control studies the choice of controls may affect the representativeness of exposure.115 123 In addition, characteristics of the hospital catchment area and the presence of Berkson's bias, if study exposure is related to the risk of being hospitalised for the control diseases, may jeopardise the comparability of studies selected for the analysis even further.124 Socioeconomic inequalities may influence all above-mentioned conditions and, thus, be one of the reasons for the type of controls to be a source of heterogeneity. For income-based SEP overall results for studies published before 2000 yielded highly a significant association with the outcome along with no heterogeneity between studies, although in multivariable meta-regression the publication year was not recognised as a potential source of heterogeneity. We must, however, acknowledge that in meta-analysis of observational studies the results of multivariable meta-regression can also be confounded by unidentified study-level factors.41
Publication bias is a concern in meta-analysis because it might lead to overestimation of the RR. As our meta-analysis was restricted to English-speaking peer-reviewed publications, our estimates may have been affected by missing data from the studies performed in the low or middle-income countries where SEP may be differently associated with AMI. It may explain the presence of publication bias detected in our analysis for combined estimates for studies on income-based SEP. In our study, publication bias may be partly explained by variability in outcome definitions and SEP measures or if studies with relatively smaller sample sizes and inconsistent results were lacking in the analysis representing significance bias, size bias and suppression publication bias.119
Our selection criteria, particularly with respect to study definition, were rather tight, which resulted in a limited number of studies. We acknowledge the fact that numerous studies potentially relevant for the analysis were, however, excluded if AMI was reported as a study endpoint in combination with chronic heart diseases. In order to control highly heterogeneous outcomes and being mostly interested in social patterning of the first acute coronary event, we deliberately narrowed the analysis to AMI only.
Another restraint deals with the variety of confounders and/or mediators across the original studies. It might be crucial for the pooled results if the adjustment strategy differs between the studies with respect to risk factors for myocardial infarction. The complicated interplay between socioeconomic factors and risk factors for CVD must be acknowledged while interpreting the results of meta-analysis. It is well known that standard risk factors explain more than half of the association between SEP and CHD mortality and morbidity, and there is a pronounced socioeconomic gradient in CHD-related behavioural and lifestyle factors.125–128 Therefore, the choice of variables in the original study to adjust for can influence the pooled results. Adjustment for risk factors acting as intermediate steps in the causal pathways will result in underestimation of the relation between SEP and the disease, while relation will be overstated if genuine confounders are left unadjusted. Evidence from recent studies and reviews indicate that different SEP measures can simultaneously be included in the multivariable models as they do not act as proxy for each other,117 118 129–131 while biological, behavioural and psychological factors can mediate an association between SEP and AMI.126–128 131 To reduce the influence of various adjustments we, therefore, performed separate subanalyses for studies minimally and maximally adjusted that in our analysis yielded similar results.
Our findings present the overall increase in the risk of AMI incidence among the lowest SEP that has previously been reported for AMI mortality and CHD morbidity.13 16 The aforementioned results corroborated the strong evidence of the relation between socioeconomic deprivation and the incidence of acute ischaemic events. Further research providing validated information is required to address public health strategies to reduce the risk of AMI among the most vulnerable groups in different countries and among different societies. It is imperative to emphasise the importance of such studies, particularly for the regions with a lower level of economic development, where the epidemic of CHD is becoming a public health issue.
What is already known on this subject
Adverse SEP is related to CHD mortality and survival.
What this study adds
This meta-analysis reveals an association between low SEP and an increased incidence of AMI. The associations were consistent for both men and women. People from the lowest strata of income-based SEP have a 71% greater risk of developing myocardial infarction compared with those in the highest strata. An increased risk of AMI incidence of 34% and 35% was found for the lowest compared with the highest categories of educational and occupational SEP, respectively. The socioeconomic patterns in myocardial infarction incidence were seen to be most pronounced for high-income countries. The nature of social stratification at the level of economic development of a country could be involved in the differences in the risk of AMI between social groups.
Policy implications
Public health policies aimed at reducing the risk of AMI should address SEP both in the promotion and the evaluation of preventive measures. The potential for variation in the strength of AMI inequalities between different societies should be acknowledged by national and international policy makers.
References
Supplementary materials
Web Only Data jech.2009.104075
Files in this Data Supplement:
Footnotes
EM-G and AS contributed equally.
Funding This work was supported by a grant from the Swedish Council for Working Life and Social Research (FAS 2006-0230 to TM), Stockholm, Sweden.
Competing of interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.