Despite important medical advances in the prevention and treatment of coronary heart diseases (CHD), these diseases remain the leading cause of mortality in most industrialized countries. Their reduction represents a major public health challenge.1,2 Several individual factors (eg, smoking habits, leisure or physical activity, diabetes, hypertension, cholesterol level, obesity, etc.) have been advanced to explain CHD risk.2,3 More recently, neighborhood characteristics have also been shown to be related to the incidence of CHD.4 In a recent paper, Fiscella and Tancredi3 addressed the importance of socioeconomic status (SES) in CHD risk assessment. Disparities in the risk of CHD have been observed across a variety of SES indicators, including income, education, occupation, and deprivation indices estimated at ecologic levels.3,5,6 Usually a higher risk has been reported among groups with a low SES.5,6 These results have been found using individual or contextual/ecologic SES data, and have been confirmed by multilevel analysis using both levels of SES data.4–6
It is well-known that the incidence of CHD is higher in men than in women at all ages, although this difference tends to decrease with age. Behavioral, psychosocial, and biologic factors may explain part of this sex contrast.1,7 Although socioeconomic disparities have been observed separately in men and women, gradients in CHD risk have been reported to be stronger among women in several studies, on both relative risk or odds ratio scales,8–12 even after adjusting for individual characteristics.4,13,14 These sex differences may be in part a direct consequence of unequal incidence rates,15–17 but other contributions remain unclear and require further investigation.
In this context, we used myocardial infarction (MI) data collected from a regional CHD registry to explore the association between neighborhood deprivation level and the risk of MI at a small-area scale in the Strasbourg area (northeastern France), using an ecologic study design. To derive relevant epidemiologic results from such area-based aggregated data, it is essential to account for spatial dependence among geographical areas.18 Neighboring areas are more likely than distant areas to share similar health, socioeconomic, or environmental characteristics, and this spatial phenomenon can strongly influence results. Failure to consider this in the analysis may lead to biased results.18 To take into account the spatial dependence of the data and the variability of MI rates due to the small number of events per geographic unit, we used a hierarchical Bayesian modeling approach.
As recently recommended by Kaufman,19 we analyzed the socioeconomic gradient in the MI risk both in absolute and relative scales, separately for men and women. By doing so, we attempted to assess effect modification of SES by sex in a way that allowed the study of deviation from additive or multiplicative joint effects.
Study Setting and Small-area Level
Our study setting is the Strasbourg metropolitan area, an urban area of 28 municipalities (316 km2) located in the Bas-Rhin district in northeastern France, with a population of about 450,000. The small-area level used was the French census block, a submunicipal division designated by the National Census Bureau. This unit corresponds to a residential neighborhood with an average of 2000 inhabitants, and is the smallest administrative geographic unit in France for which socioeconomic and demographic information is available from the national census. The division of neighborhoods into census blocks takes into account the physical obstacles that may break up urban landscapes (important traffic arteries, bodies of water, green spaces, etc.) and aims to maximize homogeneity of population size, socioeconomic characteristics, and land use and zoning.
The Strasbourg metropolitan area is subdivided into 190 blocks with a mean population of 2382 inhabitants (range: 2–4885 inhabitants) and a mean area of 1.65 km2 (range: 0.05–19.60 km2). To comply with confidentiality guidelines, 16 blocks with very small populations (<250 inhabitants) had to be excluded before health data collection; these accounted for only 0.8% of the total population.
The Bas-Rhin CHD registry continuously monitors all MIs, coronary deaths and possible coronary deaths (sudden deaths, and deaths of unknown causes) that occur among men and women aged 35–74 years in the Bas-Rhin district. From 1984 to 1993 this registry participated in the World Health Organization's MONICA Project (Multinational Monitoring of Trends and Determinants in Cardiovascular disease).20 Since 1997, a national standardized registration procedure, simpler than the original protocol, was adopted for case monitoring.21 The main sources for MI case-finding were public and private hospitals, emergency departments, and cardiologists' and general practitioners' private practices.
Our study population is the people aged 35–74 years living in the census blocks of the Strasbourg metropolitan-area between 1 January 2000 and 31 December 2003. Our events are all MI events (International Classification of Diseases, 9th revision (ICD-9) code 410) that arise from our population. According to the registry's monitoring criteria, a MI event is defined as an event (nonfatal or fatal) for which the MI diagnosis has definitively been established by a trained physician.21 An event is considered as new if it occurs more than 27 days after a previous MI event, and is considered as fatal if death occurs within 28 days. Because comparable associations have been observed for recurrent and incidents events with deprivation level16,22,23 and with sex,8,23 we combined incident and recurrent events in our study. Incident cases were defined as all first coronary events without evidence of a clinically recognized previous MI event in the patient's history, whereas recurrent events were to those for which the subject had a history of one or more previous MI. Among all events registered during the 4-year period, 87% were first and 12% were recurrent (missing information for 1%).
Each event was assigned to the subject's census block of residence after geocoding his or her postal address using ArcGIS version 9.1 (ESRI, Redlands, CA). If the postal address was incomplete or erroneous, the general practitioners, hospitals, or other healthcare services who had treated the subject were contacted again to try to obtain correct address information (less of 1% of all events). Only 6 events (0.5%) could not be geocoded, and they were excluded from the analysis. Ultimately, our study population included 1193 events. This research protocol was approved by the French data protection authority.
Neighborhood Deprivation Level
To characterize accurately the neighborhood deprivation level, we used a socioeconomic deprivation index developed for the Strasbourg metropolitan area at the census block level. A detailed description of the methodologic development and statistical validity of this index can be found elsewhere.24 Briefly, this index was constructed by a principal component analysis from a selection of 19 socioeconomic and demographic variables from the 1999 national census, reflecting the multiple dimensions of socioeconomic deprivation: income, education attainment, job, housing characteristics, family structure, and immigration status. We have previously demonstrated the ability of this index to capture socio-spatial inequalities in health and environmental outcomes such as asthma attacks25 and traffic-related air pollution exposure.26
Census blocks were divided into 5 deprivation categories according to their index value; the first category comprised the most privileged blocks, and the fifth the most deprived.24 This classification ensured the best discrimination of neighborhoods according to their SES, and highlighted a strong socioeconomic gradient within the Strasbourg metropolitan area from the most privileged blocks (located on the outskirts) to the most deprived ones (concentrated in the urban center and its inner suburbs).24
Relative Rate Estimation
We used a hierarchical Bayesian modeling to estimate the census-block-specific age-adjusted relative rates (RRs) of MI event by neighborhood deprivation level and by sex, using the intrinsic Gaussian conditional autoregressive model proposed by Besag et al27 This model controls both the heterogeneity effect and the spatial autocorrelation component that had been found in our data. This model is structured into 3 hierarchical levels described later in the text. The first level of the model is defined following the classic methodologic process.
Let Oijg and Nijg represent the number of MI events and the number of persons at risk in census block i, i = 1,...., n (where i ranges from 1 to 174, the total number of census blocks), in age-group j, j = 1,.... J (with 2 age-groups, 35–54 years and 55–74 years), and in sex-group g (where g equals 1 for men and 2 for women).
A Poisson model for the count data typically assumes that Oijg∼ Poisson(Nijgπijg) where πijg is the rate in age-group j of the census block i and sex-group g. The data will be too sparse to obtain robust estimates of age-specific relative rate. Instead, the data are indirectly age-standardized using the age distribution (35–54 years and 55–74 years) of the Strasbourg metropolitan-area population separately for men and women obtained from the 1999 national census. Thus, the model is now described by the following equation:
Oig ∼ Poisson(Eigθig)
where Oig=∑jOijg represents the total number of MI events in the census block i and sex-group g, Eig=∑j Nijgπjg denotes the expected cases in the census block i and sex-group g based on age-group specific reference rates πjg by sex-group g and,
is the relative rate of the MI event in the census block i and sex-group g relative to that expected. It is equivalent to a standardized morbidity ratio (SMR). The Strasbourg metropolitan area population aged between 35 and 74 years has been used.
The final layout of the first level of the Bayesian model is
The second level assigns to θig a log-normal prior distribution defined by
where αg is the sex-specific intercept in the least deprivation blocks, Xigk are census-block-specific covariates with corresponding parameters βgk, and where ui and νi, correspond to 2 census-block-specific random components that model the effects varying in a spatially structured way (ie, census block clustering) and in an unstructured way (ie, census block heterogeneity), respectively. We introduced the neighborhood deprivation level into the model as 4 covariates (Xigk with k=1,..., 4) corresponding to all the deprivation categories except the first (the most privileged), which is used as the reference.
The second level also assigns a Gaussian conditional autoregressive prior distribution and a normal prior distribution to ui and νi respectively.
where u−i denotes the spatial effects for all blocks other than the block i,
is the mean of the ui for the blocks bordering the block i; adjacency being used as criterion of geographic proximity, mi is the number of neighbors of the block i, and σu2 and σv2 are the variance parameters of the ui and vi distributions, respectively.
The third level assigns hyper prior distributions to different parameters. Without prior expectations about direction and magnitude of the covariate effects, a vague prior distribution is put on the regression coefficients: normal distributions with very large variance to the intercept αg and to the regression coefficients βgk.
Classically used,28 for analytic and numeric considerations, gamma prior distributions have been associated with the inverse of the variance parameters σu2 and σv2 (4).
Parameters for the gamma hyperprior distributions were selected after sensitivity analysis using the Deviance Information Criterion developed by Spiegelhalter et al29 to compare the different models. More details are presented in eAppendix 1 (http://links.lww.com/EDE/A396).
Absolute rates of MI rather than relative rates have thereafter been estimated to test departure from additive joint effects of sex and deprivation on the appropriate metric—to avoid being misled by differences in background risk levels between sexes.30 The methodologic procedure is decomposed into the following steps.
Our Bayesian model produces the posterior distribution of θ, a matrix of relative rates for 174 census block by 2 sex-groups. The relative rates have been converted into indirectly age-standardized event rates (per 100,000 person-years) λig=ωgθig. The ωg represents the crude sex-and-age-adjusted rate for the less deprived census blocks. More precisely, ωg is defined as the total number of observed events in the less deprived census-blocks divided by the corresponding total person-years. The variability of λig incorporates the variability of both ωg and θig.
Weighted regression models were used to assess the relationship between MI rates and the deprivation index while testing for interaction by sex. The weights are equal to the inverse of the sex-specific rate variance estimated in each census block. We first analyzed the fit of a simple linear model for the deprivation effect with different slopes for the 2 sexes (M0)
where λig denotes the absolute rate for sex g in census block i, α is the intercept, Gg represents the sex dummy variable (g takes value 1 or 2, for women and men), and Di the quantitative value of deprivation index for census block i.
The M0 goodness of fit was compared with the M1 model in which an interaction term between deprivation level and sex was added. Finally, we assessed the fit of a nonlinear model (M2) for women, introducing a quadratic term. The final best model of age-adjusted event rates λi was:
Nested models were compared using the likelihood ratio test. The likelihood ratio test statistic approximately follows a χ2 distribution, with the degrees of freedom equal to the number of additional parameters in the more complex model; in our cases, it is always equal to 1.
Relative rates were estimated via Markov Chain Monte Carlo algorithms with WinBUGS software, version 1.4.1 (MRC Biostatistics Unit, Cambridge, UK). Parameter means and 95% credible intervals (CIs) were estimated from 2 independent chains of 30,000 iterations after a burn-in of 30,000 iterations. The Winbugs code is presented in eAppendix 2 (http://links.lww.com/EDE/A396). The convergence of chains was checked with the Gelman and Rubin convergence diagnostic as modified by Brooks and Gelman.31 (eAppendix 3 gives the trace plot for the smallest block and the largest block [http://links.lww.com/EDE/A396]). Rates were estimated using SAS v9.1 software (SAS Institute Inc., Cary, NC). Relative-rate maps were drawn with ArcGIS version 9.1 (ESRI, Redlands, CA).
During the 4-year study period, 1193 MI events (912 men and 281 women) were identified among the Strasbourg metropolitan area. Most were older than 55 years (62% of men and 75% of women). The spatial distribution of the Strasbourg metropolitan area deprivation index (Fig. 1) shows a socioeconomic gradient, from the most deprived census blocks in the urban center and inner suburbs, to the most privileged blocks on the city outskirts. Figure 2A and B presents the spatial distributions of relative rates of MI without adjustment for the deprivation level, for men and women respectively. They show a spatial heterogeneity for both sexes. Comparison with Figure 1 suggests a spatial association between the risk of MI and the neighborhood deprivation level.
Table 1 presents results from 4 Bayesian models, all of which include the deprivation level. Estimates of the RR coefficients are very similar in all models, but models 2, 3, and 4 give a higher standard error resulting in a slightly less precise 95% credibility interval. Table 2 shows the goodness of fit of the model with and without the deprivation covariate. Including the deprivation covariate in the model improves both the deviance and the deviance information criterion. The model that yields the best results is the one that includes both heterogeneity and spatial autocorrelation effects. (eAppendix 4 summarizes all the posterior model parameters [http://links.lww.com/EDE/A396].) Relative rates increase with the deprivation level for both sexes. However, women living in the most deprived neighborhoods exhibited a risk of MI almost 2.5 times greater than those living in the least deprived neighborhoods (RR = 2.49 [95% CI = 1.63–3.76]), whereas the corresponding RR for men was 1.24 (0.95–1.62).
Values of the likelihood ratios statistics for nested models are given in Table 3. The best-fitting model was M2 whose parameter estimates are presented in Table 4. Figure 3 exhibits, for men and women separately, incidence rates derived from the Bayesian model in each census block and the predicted incidence rates from the best regression model according to the deprivation index. The shape of the relation between deprivation levels and MI rates is different across sexes. Although a linear gradient of MI risk according to deprivation fits the data among men, the relationship is nonlinear among women. Our data provide evidence of effect modification, with departure from an additive joint effect of sex and deprivation.
The main findings of our study are, first, that the risk of MI in the Strasbourg metropolitan area is higher among men than among women irrespective of the deprivation level. Second, although MI rates increase with the deprivation level of census blocks both for men and women, the shape of this association differs for men and women. The increase of the risk become steeper among women living in the most deprived areas.
There are important unanswered questions regarding sex differences in the SES gradient in CHD. Few studies have been able to examine formal interactions, as has been recently demonstrated by Knol et al32 Following the warning by Rothman,30 it was recently recalled19,33,34 that assessing effect modification by comparison of relative risks is inappropriate when the baseline risks are uneven. For this reason, we recalculated incidence rates using the parameters of the Bayesian model and compared the shape of the relationship between deprivation level and MI rates across sexes, demonstrating a substantial difference between men and women (linear versus nonlinear relationships, respectively). The deviation from additivity found among women is of interest from a public-heath perspective because it underscores a subgroup of the population (namely women living in the most disadvantaged neighborhoods) who might benefit from dedicated public health interventions. This result is not a consequence of a higher proportion of women in the most disadvantaged category in the Strasbourg metropolitan area, the proportion of women being stable across the deprivation level categories.
Several epidemiologic studies that have used individual11,12 or neighborhood SES measure levels8,9 have demonstrated a sex difference in the effect of neighborhood SES on CHD risk. Similar results were suggested in multilevel studies4,13–17,35 independently of individual SES effect. For example, Thurston et al,11 in a large and a representative study of the US population collected from the National Health and Nutrition Examination Survey (NHANES), revealed that the increased CHD risk associated with low education appeared stronger in women than men. In Rome, using MONICA data as in our study, Picciotto et al9 found that overall incidence had a strong socioeconomic gradient, which was steeper for women than for men. However, in all cases, SES contrasts accounted for more coronary events among men than among women because women's absolute risk was much lower.
Several studies have recognized that the sex difference may be in part a direct consequence of the unequal frequency of these events between men and women.9,15–17 For this reason, we assessed the sex interaction on the absolute scale. Some studies have offered hypotheses to explain this differential effect of SES according to sex. The Thurston et al study11 is the only one, to our knowledge, that clearly explored individual factors explaining sex disparities in the association between SES and CHD risk. The authors argued that body mass index, combined with other metabolic factors (diabetes and cholesterol), may account for most of the sex difference observed in the educational gradient of CHD risk. A stronger association between SES and metabolic factors among women had already been mentioned in earlier works.36–38 A French population-based study using data from 3 regional CHD registries39 supports metabolic factors as a realistic hypothesis to explain the sex interaction observed in our study; it found that poor household income may increase the risk of metabolic syndrome in a sex-specific manner after adjusting for lifestyle variables. Dallongeville et al39 argued that women in the lowest household-income category are more likely than men to be unemployed and have limited economic resources; these conditions are associated with low physical activity and increased stress, 2 factors that promote weight gain and metabolic disorders. Limited economic resources may also engender other unhealthy behaviors, such as consumption of low-cost high-calorie foods and psychotropic drugs, both of which are risk factors for metabolic disorders.39 Thurston et al11 advanced similar hypotheses, emphasizing the greater psychosocial disadvantages of women with low educational levels, who are more likely than men to be depressive and single parents. These characteristics generate stressful conditions at home and at work, promoting the development of obesity and coronary diseases.
However, such hypotheses cannot be verified in our ecologic study, which lacks individual data on risk factors and socioeconomic characteristics. Nonetheless, these hypotheses are plausible explanations for our findings. The use of small and homogeneous areas helped ensure that the associations we observed would converge toward the estimates observed at an individual level. Further, the more disadvantaged neighborhoods of the Strasbourg metropolitan area present socioeconomic characteristics that can explain a greater association between metabolic factors and SES for women than men (higher unemployment, lower educational levels, and more single-parent households among women). In this setting, in absence of data on individual risk factors, our fine spatial analysis might have the virtue of capturing some sources of variability that would follow social and spatial patterns. Conventional risk factors for CHD such as smoking, hypertension, high cholesterol, and diabetes, are such candidates.40,41 Air pollution also exhibits spatial patterns that are related to social characteristics, including in our study area.26 Now particulate air pollution is known to affect the risk of MI.42,43
Several strengths of our study deserve to be highlighted. First, few studies have compared sex differences in the risk of MI associated with deprivation level. An earlier study that also used French registries data showed socioeconomic inequalities in the MI incidence and case-fatality rates according to occupational category, but it included only men.44 Second, this study is based on data from a nationally coordinated CHD registry that monitors coronary events according to a validated standardized protocol. The high monitoring quality ensures the good specificity of the health events recorded. Third, we implemented a complex and powerful method for studying the associations between deprivation level and MI risk at an ecologic level. This approach, by smoothing the noise caused by the strong initial instability of our data due to the small number of events per census block (especially among women), allowed display of a potential spatial structure of the health data and thus made it easier to relate health data to explanatory factors such as the deprivation level. This approach also allows us to take into account the heterogeneity and spatial dependence of the geographic units—essential phenomena to consider in ecologic studies to minimize biases in the estimation.18 The importance of Bayesian methods for epidemiologic questions has recently been underscored,45 and these methods are being used more often in CHD social epidemiology46 and environmental epidemiology.47
Finally, beyond the limitations directly attributable to the ecologic design, the main limitation of this work is the lack of individual information on subjects' SES and on their principal MI risk factors, especially metabolic factors. The availability of these data would have allowed us to conduct a contextual analysis to estimate the specific effect of neighborhood SES on the risk of MI, to test the hypothesis that metabolic factors play a role in the interaction between sex and SES in relation to MI risk, and possibly to identify the role of other factors.
In conclusion, this study confirms that men are at greater risk of MI than women, and adds an important new finding: although neighborhood deprivation raises the level of risk CHD for both men and women, women are substantially more susceptible than men to this neighborhood deprivation. This finding should be confirmed and potential explanatory factors for these sex differences explored in further studies, preferably with a contextual approach.
1.Yusuf S, Hawken S, Ounpuu S, et al. Effect of potentially modifiable risk factors associated with myocardial infarction in 52 countries (the INTERHEART study): case-control study. Lancet
2.Kotseva K, Wood D, De Backer G, et al. Cardiovascular prevention guidelines in daily practice: a comparison of EUROASPIRE I, II, and III surveys in eight European countries. Lancet
3.Fiscella K, Tancredi D. Socioeconomic status and coronary heart disease risk prediction. JAMA
4.Diez-Roux AV, Merkin SS, Arnett D, et al. Neighborhood of residence and incidence of coronary heart disease. N Engl J Med
5.Gonzalez MA, Rodriguez AF, Calero JR. Relationship between socioeconomic status and ischaemic heart disease in cohort and case-control studies: 1960–1993. Int J Epidemiol
6.Kaplan GA, Keil JE. Socioeconomic factors and cardiovascular disease: a review of the literature. Circulation
7.Rosengren A, Hawken S, Ounpuu S, et al. Association of psychosocial risk factors with risk of acute myocardial infarction in 11,119 cases and 13,648 controls from 52 countries (the INTERHEART study): case-control study. Lancet
8.Morrison C, Woodward M, Leslie W, et al. Effect of socioeconomic group on incidence of, management of, and survival after myocardial infarction and coronary death: analysis of community coronary event register. Br Med J
9.Picciotto S, Forastiere F, Stafoggia M, et al. Associations of area based deprivation status and individual educational attainment with incidence, treatment, and prognosis of first coronary event in Rome, Italy. J Epidemiol Community Health
10.Salomaa V, Niemela M, Miettinen H, et al. Relationship of socioeconomic status to the incidence and prehospital, 28-day, and 1-year mortality rates of acute coronary events in the FINMONICA myocardial infarction register study. Circulation
11.Thurston RC, Kubzansky LD, Kawachi I, et al. Is the association between socioeconomic position and coronary heart disease stronger in women than in men? Am J Epidemiol
12.Vogels EA, Lagro-Janssen AL, van Weel C. Sex differences in cardiovascular disease: are women with low socioeconomic status at high risk? Br J Gen Pract.
13.Sundquist K, Malmstrom M, Johansson SE. Neighbourhood deprivation and incidence of coronary heart disease: a multilevel study of 2.6 million women and men in Sweden. J Epidemiol Community Health
14.Winkleby M, Sundquist K, Cubbin C. Inequities in CHD incidence and case fatality by neighborhood deprivation. Am J Prev Med
15.Diez-Roux AV, Nieto FJ, Muntaner C, et al. Neighborhood environments and coronary heart disease: a multilevel analysis. Am J Epidemiol
16.Kolegard SM, Diderichsen F, Reuterwall C, et al. Socioeconomic context in area of living and risk of myocardial infarction: results from Stockholm Heart Epidemiology Program (SHEEP). J Epidemiol Community Health
17.Stjarne MK, Ponce de LA, Hallqvist J. Contextual effects of social fragmentation and material deprivation on risk of myocardial infarction–results from the Stockholm Heart Epidemiology Program (SHEEP). Int J Epidemiol
18.Elliott P, Wakefield J, Best N, et al. Spatial Epidemiology: Methods and Applications.
Oxford: Oxford University Press; 2000.
19.Kaufman JS. Interaction reaction. Epidemiology
20.Tunstall-Pedoe H, Kuulasmaa K, Amouyel P, et al. Myocardial infarction and coronary deaths in the World Health Organization MONICA Project. Registration procedures, event rates, and case-fatality rates in 38 populations from 21 countries in four continents. Circulation
21.Arveiler D, Wagner A, Ducimetiere P, et al. Trends in coronary heart disease in France during the second half of the 1990s. Eur J Cardiovasc Prev Rehabil
22.Barakat K, Stevenson S, Wilkinson P, et al. Socioeconomic differentials in recurrent ischaemia and mortality after acute myocardial infarction. Heart
23.Davies CA, Dundas R, Leyland AH. Increasing socioeconomic inequalities in first acute myocardial infarction in Schotland, 1990–92 and 2000–02. BMC Public Health
24.Havard S, Deguen S, Bodin J, et al. A small-area index of socioeconomic deprivation to capture health inequalities in France. Soc Sci Med
25.Laurent O, Filleul L, Havard S, et al. Asthma attacks and deprivation: gradients in use of mobile emergency medical services. J Epidemiol Community Health
26.Havard S, Deguen S, Zmirou-Navier D, et al. Traffic-related air pollution and socioeconomic status: A spatial autocorrelation study to assess environmental equity on a small-area scale. Epidemiology
27.Besag J, York J, Mollie A. Bayesian image restoration with two applications in spatial statistics. Ann Inst Stat Math
28.Lawson A, Browne W, Vidal Rodeiro C. Disease Mapping With WinBUGS and MLWin.
Chichester, United Kingdom: John Wiley & Sons; 2003.
29.Spiegelhalter D, Best N, Carlin B, et al. Bayesian measures of model complexity and fit (with discussion). J
Royal Stat Soc B
30.Rothman KJ. Interactions between causes. In: Modern Epidemiology.
Boston, MA: Little Brown and Company; 1986:311:–326.
31.Brooks S, Gelman A. Alternative methods for monitoring convergence of iterative simulations. J Comput Graph Stat
32.Knol MJ, Egger M, Scott P, et al. When one depends on the other: reporting of interaction in case-control and cohort studies. Epidemiology
33.Greenland S. Interactions in epidemiology: relevance, identification, and estimation. Epidemiology
34.VanderWeele TJ. Sufficient cause interactions and statistical interactions. Epidemiology
35.Stjarne MK, Fritzell J, De Leon AP, et al. Neighborhood socioeconomic context, individual income and myocardial infarction. Epidemiology
36.Loucks EB, Rehkopf DH, Thurston RC, et al. Socioeconomic disparities in metabolic syndrome differ by gender: evidence from NHANES III. Ann Epidemiol
37.Sobal J, Stunkard AJ. Socioeconomic status and obesity: a review of the literature. Psychol Bull
38.Tang M, Chen Y, Krewski D. Gender-related differences in the association between socioeconomic status and self-reported diabetes. Int J Epidemiol
39.Dallongeville J, Cottel D, Ferrieres J, et al. Household income is associated with the risk of metabolic syndrome in a sex-specific manner. Diabetes Care
40.Singh GK, Kogan MD. Persistent socioeconomic disparities in infant, neonatal, and postneonatal mortality rates in the United States, 1969–2001. Pediatrics
41.Lynch J, Davey SG, Harper S. Explaining the social gradient in coronary heart disease: comparing relative and absolute risk approaches. J Epidemiol Community Health
42.Bhaskaran S, Hajat A, Smeeth L. Effects of air pollution on the incidence of myocardial infarction. Heart
43.Zanobetti A, Schwartz J. The effect of particulate air pollution on emergency admissions for myocardial infarction: A multi-city case-crossover analysis. Environ Health Perspect
44.Lang T, Ducimetiere P, Arveiler D, et al. Incidence, case fatality, risk factors of acute coronary heart disease and occupational categories in men aged 30–59 in France. Int J Epidemiol
45.Graham P. Intelligent smoothing using hierarchical Bayesian models [commentary]. Epidemiology
46.Beard JR, Earnest A, Morgan G, et al. Socioeconomic disadvantage and acute coronary events: a spatiotemporal analysis. Epidemiology
47.Naess O, Piro FN, Nafstad P, et al. Air pollution, social deprivation, and mortality: a multilevel cohort study. Epidemiology