Annals of Epidemiology

Transcription

1 Annals of Epidemiology 22 (2012) 541e546 Contents lists available at SciVerse ScienceDirect Annals of Epidemiology journal homepage: Socioeconomic environment and recurrent coronary events after initial myocardial infarction Avshalom Koren MA a,c, David M. Steinberg PhD c, Yaacov Drory MD b, Yariv Gerber PhD a, *, The Israel Study Group on First Acute Myocardial Infarction 1 a Department of Epidemiology and Preventive Medicine, School of Public Health, Sackler Faculty of Medicine, Tel Aviv University, Israel b Department of Rehabilitation, Sackler Faculty of Medicine, Tel Aviv University, Israel c Department of Statistics and Operations Research, School of Mathematical Sciences, Faculty of Exact Sciences, Tel Aviv University, Israel article info abstract Article history: Received 27 December 2011 Accepted 23 April 2012 Available online 1 June 2012 Keywords: Acute coronary syndromes Cohort studies Multiple failure-time data analysis Myocardial infarction Neighborhood socioeconomic status Survival analysis Purpose: Longitudinal data linking area-level socioeconomic status (SES) to repeated acute coronary syndrome (ACS) events are limited. Using multiple failure-time data, we examined the association between neighborhood SES and ACS in a community-based cohort of myocardial infarction (MI) survivors. Methods: Consecutive patients aged 65 years or younger discharged from eight hospitals in central Israel after first MI in 1992e1993 were followed through Recurrent MI and unstable angina pectoris (UAP) leading to hospitalization were recorded. Neighborhood SES was assessed through a composite census-derived index developed by the Israel Central Bureau of Statistics. Different variance-corrected proportional hazards models were used to account for multiple recurrent events: Andersen-Gill, Wei- Lin-Weissfeld (WLW), and Prentice-Williams-Peterson. Results: During follow-up, 531 recurrent MIs and 1584 UAP episodes occurred among 1164 patients. Adjusting for known prognostic factors and individual SES using the Andersen-Gill model, higher estimated hazards were associated with poor neighborhood SES (hazard ratio, 1.55; 95% confidence interval [CI], 1.13e2.14 for recurrent MI; and hazard ratio, 1.48; 95% CI, 1.22e1.79 for UAP; in the 5th vs. 95th percentiles). The WLW and Prentice-Williams-Peterson models yielded similar results. When the two outcomes were combined, the WLW-derived hazard ratio was 1.64 (95% CI, 1.39e1.93). Conclusions: MI survivors living in a deprived neighborhood are at higher risk of repeated hospital admissions because of ACS. Secondary prevention initiatives should incorporate multilevel approaches to increase effectiveness and reduce geographic health disparities. Ó 2012 Elsevier Inc. All rights reserved. Introduction Numerous studies conducted in the last decades linked socioeconomic status (SES) to health outcomes in general [1,2] and to cardiovascular disease (CVD) morbidity and mortality in particular [3,4]. SES is a multidimensional construct that covers a wide range of measures of income, education, housing, employment,living conditions, and many other socioeconomic aspects of life [3,5,6]. The role of neighborhood socioeconomic context in individuals health outcomes has been the focus of several previous reports [7,8]. Indeed, living in a disadvantaged neighborhood was associated with * Corresponding author. Department of Epidemiology and Preventive Medicine, School of Public Health, Sackler Faculty of Medicine, Tel Aviv University, Ramat Aviv, Tel Aviv 69978, Israel. address: yarivg@post.tau.ac.il (Y. Gerber). 1 See the Acknowledgments section for a list of participating medical centers and investigators. higher incidence of coronary heart disease (CHD) [8e10] and increased risk of all-cause death [7], even when individual risk factors were accounted for. Suggested mechanisms include features of both the physical (e.g., environmental exposures, food and recreational resources, built environment, and quality of housing) and social (e.g., safety/violence, social connections/cohesion, local institutions, and norms) environments [11]. Recent studies have also evaluated the association between neighborhood SES and longterm mortality after myocardial infarction (MI); it was found that survival post-mi is lower in more deprived neighborhoods [12e14]. However, data relating neighborhood SES to the risk of recurrent acute coronary syndrome (ACS) events after MI are lacking. Previous studies that examined the relationship between neighborhood SES and CVD risk considered only the first event, ignoring subsequent occurrences during follow-up. This might result in a potential loss of statistical power for detecting exposure effect. Furthermore, the use of multiple failure-time data leads to /$ e see front matter Ó 2012 Elsevier Inc. All rights reserved. doi: /j.annepidem

2 542 A. Koren et al. / Annals of Epidemiology 22 (2012) 541e546 greater efficiency in estimating covariate effects relative to singleevent models [15]. Clinically, the single-event approach might not fully capture long-term prognosis. From a health policy standpoint, neglecting multiple ACS recurrences can cause an underestimation of the health burden associated with poor neighborhood SES. Applying different methods of analyzing multiple failure-time data, the study objective was to examine the association between residing in a disadvantaged neighborhood and repeated ACS hospitalizations among MI survivors, adjusting for established prognostic factors and individual SES measures. Methods The data for this study were drawn from the Israel study of first acute myocardial infarction, which examines long-term outcomes after MI and has been described in detail in previous reports [12,16]. Briefly, the study consists of patients aged 65 years or younger, who were hospitalized for initial MI in all medical centers (n ¼ 8) in central Israel. Recruitment lasted from February 15th,1992 to February 15th, Patients were followed up until death or by December 31st, 2005, the end of the study. Personal sociodemographic and clinical data were obtained via medical records, and structured interviews were performed approximately 1 week after the index hospitalization. Four follow-up interviews were subsequently conducted 3e6 months, 1e2 years, 5 years, and 10e13 years after MI. Among 1626 consecutive patients hospitalized with first MI, 1545 were discharged alive, of whom 1521 consented to participate. The present study consists of 1164 participants, after excluding patients with missing data on either neighborhood SES (n ¼ 111) or multiple ACS recurrences (n ¼ 246). No differences were detected between study participants and nonparticipants in any of the variables assessed at baseline, including demographic, socioeconomic, or clinical characteristics, or in mortality. The primary explanatory variable The main exposure was neighborhood SES that was estimated through a composite index developed and validated by the Israel Central Bureau of Statistics [17]. The index, based on the 1995 National Census of Population and Housing, was constructed from a list of measures that includes demography, education, employment, occupation, and standard of living. The variables used in the construction of the index were chosen based on factor analyses. Neighborhood SES was measured on a 20-point scale (a higher score means better SES). Patient addresses were classified according to neighborhood SES by geographic information system tools; each patient received an SES score based on geocoding of his/her address at the time of the index MI [12,16]. Individual socioeconomic variables Other socioeconomic measures were self-reported by the patients and included family income relative to national average at that year, living with a steady partner, years of education, and employment status before MI [18]. Clinical variables The clinical covariates included biobehavioral risk factors. Diabetes, hypertension, and hypercholesterolemia were defined according to standard criteria based on clinical and laboratory data. Smoking was self-reported. MI severity indices included chronic CHD (4 months or more since diagnosis), Killip class, hospitalization in intensive care unit, unstable angina pectoris (UAP) at baseline, and comorbidity burden (as assessed by the Charlson index). Early coronary intervention included percutaneous transluminal coronary angioplasty and coronary artery bypass grafting performed within 45 days of the index MI. The outcome variables The primary outcome variable was time to hospital admissions because of recurrent ACS, which included the following: recurrent MI, UAP, and sudden cardiac death. Outcome data were ascertained through various sources, including medical records, the Israeli Population Registry, death certificates, hospital charts, family physicians and family members, and verified by a senior cardiologist (Y.D.). Initially, the time to hospital admissions because of recurrent MI/sudden cardiac death and UAP were analyzed separately. These outcomes were subsequently combined ( recurrent ACS events ) and analyzed. The time to first event was measured from the date of the index MI. The time to subsequent hospitalizations was measured either from baseline or from the previous hospital admission because of ACS (depending on the model used) [19,20]. Statistical analyses The following analytical methods for recurrent events data were used. Mean cumulative function A nonparametric estimator, described by Nelson [21], was used to compare the average number of recurrent ACS events between neighborhood SES groups. It estimates the mean number of recurrent events and its curve changes as a function of time. Semiparametric methods for recurrent events data A number of methods were used for modeling survival data. The ordinary Cox proportional hazards model was initially applied using time to first event as the outcome [22]. Additionally, three methods for analysis of multiple events per patient were applied: the Andersen-Gill (AG) model [23], the Wei-Lin-Weissfeld (WLW) marginal model [24], and the Prentice-Williams-Peterson (PWP) conditional models [25]. Each of these methods, which have been described and compared previously [19,20,26e28], estimates the effect of covariates on the time until the events occur ignoring the correlation between the multiple observations (induced by multiple events from a given subject) but applying a robust covariance matrix that corrects for this correlation. The AG model assumes that all events are of the same type and independent. It uses the counting process (CP) formulation of risk intervals, where a subject contributes to the risk set for a specific event time as long as he is under observation at the time the event occurs, regardless of the number of events he had experienced until this time. This model assumes a common baseline hazard that is shared by all K recurrent events and estimates global parameter for the exposure variable. The AG model is typically used when recurrent events are treated as identical, and the event order is not important. Unlike AG, WLW is a marginal model with eventspecific baseline hazard function. This model is an extension of the stratified Cox model. It considers each event as a separate process and uses the total time formulation of risk intervals. The risk set for the kth recurrence includes all the subjects under observation at the time the event occurs and had not yet experienced the kth recurrence (semirestricted risk set). The regression estimates represent the average effect of the covariates on the risk of recurrences, whatever is the nature of the dependence structure

3 A. Koren et al. / Annals of Epidemiology 22 (2012) 541e among related failure times. The main advantage of this approach is that it is able to model multiple-type event data. However, it fails to take into account the ordering of failure times because of the definition of the risk set. PWP offers two conditional models with event-specific baseline hazard function: conditional 1 (PWP-CP) and conditional 2 (PWP-gap time [GT]). Like WLW, these models are an extension of the stratified Cox model with strata defined by the number of recurrent events. In both models, the risk set for the kth event is restricted and includes all subjects who were under observation, have experienced the (k 1)th event and have not yet had their kth event. PWP-CP uses the CP formulation of risk intervals like the AG model, whereas PWP-GT uses the GT formulation of risk intervals. The PWP-CP model can be seen as a stratified AG model with event-specific baseline hazards and a restricted risk set. Like the WLW model, the advantage of PWP models is that they are able to estimate event-specific baseline hazards. However, unlike WLW models, they are not able to model multiple-type event data unless we treat them as if they were of the same type [19,26]. The analyses presented in this study were performed using the SURVIVAL package in R software (R Foundation for Statistical Computing, Vienna, Austria) [29] and SPSS version 18.0 (SPSS Inc., Chicago, IL). Neighborhood SES was treated as a continuous variable, unless when required for presentation purposes, where tertiles were used (lower [10], middle [11e14], and upper [15]). Outcome events were modeled as follows: recurrent MI/sudden cardiac death, UAP, and recurrent ACS. We fitted the ordinary Cox proportional hazards models and multiple event models sequentially adjusting for the following risk factors: age and sex (model 1); personal SES variables (model 2); and CVD risk factors, MI characteristics, and acute management (model 3). We estimated the hazard ratios (HRs) for outcomes and their 95% confidence intervals (CIs) per 13-unit decrease in neighborhood SES score (corresponds to the 5th vs. 95th percentiles). The multiple events models were fitted based on robust standard errors to account for interdependence across recurrent events. Only the first three recurrent MIs and the first seven events of UAP and ACS were considered because of the limited number of patients with further events. Noninformative censoring was assumed in all models. The proportional hazards assumption was tested, and it was not violated for the variables considered. There were no missing values in the covariates used in the regression models except for family income (w16%), which was imputed using multiple imputation [30]. Participants in this study were recruited from eight medical centers, thus differences in baseline hazard between patients from different hospitals ( shared frailty ) might be a concern. In a complementary analysis, random-effects Cox models were examined, accounting for potential intrahospital correlation. These models yielded similar results as those obtained with fixed-effects models, indicating negligible random medical center effects in these data. Results The average age of the 1164 participants was 54 (SD, 8) years; 81% were males. Table 1 depicts the baseline characteristics across neighborhood SES tertiles. On average, patients from the lower tertile had lower income and employment levels, were less educated, and included more current smokers and diabetics than patients in the upper tertile. In addition, patients from poorer areas were less likely to undergo revascularization procedures early after MI. Neighborhood SES and recurrent coronary events During a median follow-up of 13 years, 531 recurrent MIs and 1584 UAP episodes were recorded. The maximal number of Table 1 Pertinent baseline characteristics across neighborhood SES tertiles among participants in the Israel study of first acute myocardial infarction, 1992e2005 Baseline characteristics recurrent MIs was 4 and of UAP 19. Throughout follow-up, 31% (n ¼ 362) of the patients had at least one MI,12% (n ¼ 134) had more than one MI, and only 3% (n ¼ 29) had more than two. Correspondingly, 52% (n ¼ 610) of the patients had at least one UAP episode, 30% (n ¼ 347) had more than one UAP episode, and 19% (n ¼ 215) had more than two episodes. Regarding the recurrent ACS outcome, 65% (n ¼ 752) of the patients had at least one hospitalization, 41% (n ¼ 481) had more than one recurrent ACS, and 26% (n ¼ 301) had more than two. Among all participants, 9% (n ¼ 111) died without experiencing recurrent ACS and another 18% (n ¼ 207) experienced at least one ACS event and died. Among the 73% who were still alive at the end of follow-up, 47% (n ¼ 545) experienced at least one ACS event and 26% (n ¼ 301) had no further ACS events. The number of events by type, person-time incidence rates, and crude rate ratios between neighborhood SES tertiles are shown in Table 2. The incidence rates decreased with increasing SES tertiles, resulting in significantly lower rate ratios associated with better neighborhood SES. Multivariable survival analyses All patients (n ¼ 1164) Neighborhood SES tertiles Lower (n ¼ 360) Middle (n ¼ 434) Upper (n ¼ 370) Sociodemographic variables Age, mean (SD), y 53.9 (8.4) 54.0 (8.4) 53.2 (8.7) 54.5 (7.9) Female, % Education, mean (SD), y 11.1 (4.3) 9.3 (4.5) 11.0 (3.8) 12.9 (3.7) Relative income, % Below average Average Above average Pre-MI employment, % None Part time Full time Living with a steady partner, % Cardiovascular risk factors Hypertension, % Diabetes, % Hypercholesterolemia Smoking, % Never Past Current MI characteristics, comorbidity, and acute management Admission to ICU, % CABG within 45 days, % PTCA within 45 days, % Charlson index, % 0 points e2 points points Chronic coronary disease, % Killip class >1, % UAP, % CABG ¼ coronary artery bypass graft; ICU ¼ intensive care unit; PTCA ¼ percutaneous transluminal coronary angioplasty. The mean cumulative function estimates of recurrent ACS by neighborhood SES tertiles are shown in Figure 1. For all categories, the adjusted event rates increased much more rapidly in the months following the index MI than subsequently. Beyond approximately 2 years of follow-up, the cumulative means increased more or less linearly in all categories, at a much slower

4 544 A. Koren et al. / Annals of Epidemiology 22 (2012) 541e546 Table 2 Number of events, person-time incidence rates, and rate ratios (overall and by type) between neighborhood SES tertiles in the Israel study of first acute myocardial infarction, 1992e2005 Event type Neighborhood SES tertiles Middle vs. lower Lower (n ¼ 360), Middle (n ¼ 434), Upper (n ¼ 370), SES tertiles 3829 person-years 5058 person-years 4487 person-years Upper vs. lower SES tertiles No. of events Rate * No. of events Rate * No. of events Rate * Rate ratio y 95% CI Rate ratio y 95% CI Sudden death * 0.68, *** 0.37, 0.59 Nonfatal MI UAP ** 0.74, *** 0.51, 0.66 Total ACS *** 0.75, *** 0.49, 0.61 * Incidence rate per 1000 person-years. y Crude rate ratio (95% CI) estimated by Poisson regression. rate than previously. Nevertheless, patients in the lower tertile had a higher rate of recurrent ACS than those in the upper tertile throughout the entire follow-up. In addition to nonparametric estimates, ordinary Cox models and several multiple event methods (AG, WLW, PWP-CP, and PWP-GT) were fitted to analyze time to hospitalization because of recurrent MI (Table 3). We applied two models of the AG approach: an ordinary AG model (AG-1) and an AG model that includes a time-dependent covariate that represents the number of previous MI recurrences (AG-2). The results are shown for each of the first three MIs (when applicable) and altogether. Using Cox regressions, higher estimated hazards for first reinfarction were associated with poor neighborhood SES (HR,1.46; 95% CI, 0.99e2.15; for the 5th vs. 95th percentiles in the fully adjusted model). Accounting for multiple events yielded a slightly stronger association with a narrower CI in all the methods. Neighborhood SES was inversely associated not only with the risk of first recurrent MI but also with the risk of subsequent MIs. Cox, AG-1, and AG-2 models were fitted to assess the association between neighborhood SES and UAP (Table 4). In Cox regressions, Number of Ischemic Events Neighborhood SES tertiles: lower middle upper Time (Years) Fig. 1. Estimated mean cumulative function of recurrent acute coronary syndrome events across neighborhood socioeconomic status (SES) tertiles among 1164 participants in the Israel Study of First Acute Myocardial Infarction, 1992e2005. Adjustment was made for age, sex, education, income, pre-mi employment, living with a steady partner, hypertension, diabetes, hypercholesterolemia, smoking, comorbidity index, Killip class, chronic coronary heart disease, unstable angina pectoris, percutaneous transluminal coronary angioplasty, and coronary artery bypass grafting. higher estimated hazards for first UAP were associated with poor neighborhood SES (HR, 1.46; 95% CI, 1.09e1.97; for the 5th vs. 95th percentiles in the fully adjusted model). The corresponding HR using the AG-1 model was similar, although the CI was narrower (HR, 1.48; 95% CI, 1.22e1.79). Further adjustment for the number of previous UAP admissions (AG-2) slightly reduced the association (HR, 1.36; 95% CI, 1.12e1.66). For recurrent ACS, standard Cox and WLW models were fitted (Table 5). Using Cox models, higher estimated hazards for first ACS post-mi were associated with poor neighborhood SES (HR, 1.54; 95% CI, 1.18e2.01, in the 5th vs. 95th percentiles in the fully adjusted model). The corresponding HR using the WLW model was somewhat higher with a narrower CI (HR, 1.64; 95% CI, 1.39e1.93). Discussion In this study, we examined several multiple failure-time survival models to evaluate the long-term association between neighborhood SES and ACS recurrences after initial MI. Recurrent ACS events were highly common in the aftermath of MI. Neighborhood SES was inversely associated not only with the risk of first recurrent ACS but also with the risk of subsequent events. The fact that even after adjusting for a wide range of potential confounders and for multiple recurrences, neighborhood socioeconomic context was still associated with ACS emphasizes its independent prognostic role after MI. These findings expand previous reports of our group that revealed that neighborhood SES is strongly associated with survival after MI [12]; living in a deprived neighborhood is associated with increased risk for ischemic stroke [16]; and incorporating individual and neighborhood SES measures into a clinical model improves long-term mortality risk prediction [18]. Although multiple failure-time methods can produce different results because each addresses different research questions, all methods evaluated in this study yielded similar results. Importantly, the estimated HRs associated with neighborhood SES were stronger and more robust (i.e., with narrower CIs) than those obtained from single-event Cox models. These findings stress the exposure-outcome relationship between residential-level SES and post-mi risk. In general, we recommend using the PWP models for multiple MI recurrences because the risk set definition is suitable for these types of data. Moreover, disease risk may differ between recurrences, and the underlying hazard might increase with each additional MI. For UAP episodes, it may be preferable to use the AG model with adjustment for the number of previous hospitalizations (i.e., AG-2). For ACS recurrent events, it may be suitable to use the WLW model because it allows to model different type of events.

5 A. Koren et al. / Annals of Epidemiology 22 (2012) 541e Table 3 HRs (95% CIs) for recurrent MIs in the 5th versus 95th neighborhood SES percentiles using different analytical methods for recurrent events among 1164 participants in the Israel study of first acute myocardial infarction, 1992e2005 Method HR and 95% CI 1st MI 2nd MI 3rd MI All MIs Model 1 Cox 2.13 *** 1.52, 2.98 d d d d d d AG-1 d d d d d d 2.30 *** 1.74, 3.04 AG-2 d d d d d d 2.02 *** 1.52, 2.68 PWP-CP 2.13 *** 1.52, ** 1.44, , *** 1.56, 2.75 PWP-GT 2.13 *** 1.52, ** 1.31, , *** 1.53, 2.70 WLW 2.13 *** 1.52, *** 2.15, , *** 1.86, 3.24 Model 2 Cox 1.60 * 1.09, 2.35 d d d d d d AG-1 d d d d d d 1.72 ** 1.25, 2.37 AG-2 d d d d d d 1.56 ** 1.14, 2.15 PWP-CP 1.60 * 1.09, ** 1.34, , ** 1.18, 2.24 PWP-GT 1.60 * 1.09, ** 1.24, , ** 1.16, 2.20 WLW 1.60 * 1.09, *** 1.55, , ** 1.31, 2.48 Model 3 Cox , 2.15 d d d d d d AG-1 d d d d d d 1.55 * 1.13, 2.14 AG-2 d d d d d d 1.45 * 1.06, 2.00 PWP-CP , * 1.08, , * 1.09, 2.08 PWP-GT , * 1.05, , * 1.07, 2.04 WLW , * 1.29, , * 1.16, 2.21 Model 1: adjusted for age and sex. Model 2: model 1 plus individual SES measures (education, income, pre-mi employment, and living with a steady partner). Model 3: model 2 plus post-mi prognostic factors (hypertension, diabetes, hypercholesterolemia, smoking, comorbidity index, Killip class, chronic coronary heart disease, UAP, percutaneous transluminal coronary angioplasty, and coronary artery bypass grafting). AG-2 is also adjusted for the number of previous hospitalizations because of MI (modeled as a timedependent covariate). Strengths and limitations To our knowledge, this is the first study to evaluate the longterm association between neighborhood SES and multiple ACS recurrences. Accordingly, it has some innovations. First, it has been based on a well-defined community cohort of first MI survivors. Second, it adjusts for several key prognostic factors and accounts for multiple outcome events rather than focusing only on the first recurrence [31]. This multiple failure-time approach uses longitudinal information that better reflects long-term prognosis. Third, the exposure was based on a validated index that covers a wide range of socioeconomic measures compared with specific determinants assessed previously [9,10]. Several limitations of the present study warrant consideration. Neighborhood SES was assessed at study entry; therefore, we cannot exclude some exposure misclassification because of residential mobility during follow-up. Additionally, 23% of the original cohort were excluded from the analysis because of missing data on either exposure or outcome. However, nonparticipants did not Table 4 HRs (95% CIs) for UAP episodes in the 5th versus 95th neighborhood SES percentiles using Cox and AG models among 1164 participants in the Israel study of first acute myocardial infarction, 1992e2005 Model Cox AG-1 AG-2 HR 95% CI HR 95% CI HR 95% CI *** 1.23, *** 1.60, *** 1.37, ** 1.08, ** 1.21, ** 1.12, * 1.09, ** 1.22, ** 1.12, 1.66 Model 1: adjusted for age and sex. Model 2: model 1 plus individual SES measures (education, income, pre-mi employment, and living with a steady partner). Model 3: model 2 plus post-mi prognostic factors (hypertension, diabetes, hypercholesterolemia, smoking, comorbidity index, Killip class, chronic coronary heart disease, UAP, percutaneous transluminal coronary angioplasty, and coronary artery bypass grafting). AG-2 is also adjusted for the number of previous hospitalizations because of UAP (modeled as a time-dependent covariate). differ significantly from participants in any of the variables measured. All failure-time methods described in this study assume noninformative censoring. Recently, some advanced methods for modeling recurrent events and death when the noninformative assumption does not hold have been proposed in the literature [32]. These methods need to be further examined in future studies. Possible mechanisms Although the mechanisms for the association between neighborhood SES and ACS recurrences are still unclear, various causal pathways can be suggested. Area SES may affect CVD risk through divergent pathways not captured by individual-level indicators, such as accessibility to medical care resources, availability of healthy foods and green spaces, exposure to air pollution and to noise nuisances and social norms affecting health habits, such as smoking or drinking. Neighborhood-level disparities may also be augmented by unequal adherence to secondary prevention medications and cardiac rehabilitation and by lower utilization of coronary revascularization over time [33]. Indeed, although there is Table 5 HRs (95% CIs) for recurrent acute coronary syndrome events in the 5th versus 95th neighborhood SES percentiles using Cox and AG models among 1164 participants in the Israel study of first acute myocardial infarction, 1992e2005 Model Cox WLW HR 95% CI HR 95% CI *** 1.51, *** 1.94, ** 1.17, ** 1.41, ** 1.18, ** 1.39, 1.93 Model 1: adjusted for age and sex. Model 2: model 1 plus individual SES measures (education, income, pre-mi employment, and living with a steady partner). Model 3: model 2 plus post-mi prognostic factors (hypertension, diabetes, hypercholesterolemia, smoking, comorbidity index, Killip class, chronic coronary heart disease, UAP, percutaneous transluminal coronary angioplasty, and coronary artery bypass grafting).

6 546 A. Koren et al. / Annals of Epidemiology 22 (2012) 541e546 universal access to primary medical care services in central Israel, we found that patients residing in poor neighborhoods were less likely to be admitted to intensive care unit and to undergo early revascularization procedures than those living in more affluent neighborhoods. Additionally, the quality of care in secondary health care clinics may differ substantially between neighborhoods as a result of difficulties in recruiting and retaining qualified medical staff in deprived areas. Finally, patients residing in poor neighborhoods may be less aware of the harmful consequences of behavioral and psychological risk factors, such as smoking, sedentary lifestyle, untreated stress, depression, and anxiety [34e37]. Recommendations Typically, health care providers tend to put the patients at the center of the process and apply the most advanced knowledge to optimize their health outcomes. However, the impact of residential environment on health in general and on CVD in particular has often been neglected by public health policy makers. Once the role of residential SES in cardiovascular health is established, secondary prevention initiatives should incorporate multilevel approaches to increase clinical effectiveness and reduce geographic health disparities. Further research is needed to characterize the independent effect of personal- versus area-level SES and to elucidate the precise pathways by which SES may get under the skin and affect the heart [38]. Acknowledgments The following investigators and institutions took part in the Israel Study Group on First Acute Myocardial Infarction: Yaacov Drory, MD, Principal Investigator, Department of Rehabilitation, Sackler Medical School, Tel Aviv University, Tel Aviv; Yeheskiel Kishon, MD, Michael Kriwisky, MD, and Yoseph Rosenman, MD, Wolfson Medical Center, Holon; Uri Goldbourt, PhD, Hanoch Hod, MD, Eliezer Kaplinsky, MD, and Michael Eldar, MD, Sheba Medical Center, Tel Hashomer; Itzhak Shapira, MD, Amos Pines, MD, Margalit Drory, MSW, Arie Roth, MD, Shlomo Laniado, MD, and Gad Keren, MD, Tel-Aviv Sourasky Medical Center, Tel-Aviv; Daniel David, MD, Morton Leibowitz, MD, and Hana Pausner, MD, Meir Medical Center, Kfar Sava; Zvi Schlesinger, MD, and Zvi Vered, MD, Assaf Harofeh Medical Center, Zerifin; Alexander Battler, MD, Alejandro Solodky, MD, and Samuel Sclarovsky, MD, Beilinson Medical Center, Petach Tikvah; Izhar Zehavi, MD, and Rachel Marom-Klibansky, MD, Hasharon Medical Center, Petah Tikvah; and Ron Leor, MD, Laniado Medical Center, Netanya. The authors are also indebted to Zalman Kaufman, MSc for assistance with the Geographic Information System analysis. This work was supported in part by the Israel National Institute for Health Policy and Health Services Research (grant number r/89/2008 to Drs. Drory and Gerber), and the Environment and Health Fund, Jerusalem, Israel (grant award number RGA 0904 to Dr. Gerber). The funding sources played no role in the design, conduct, or reporting of this study. References [1] Marmot M, Smith G, Stansfeld S. Health inequalities among British civil servants: the Whitehall II study. Lancet 1991;337(8754):1387e93. [2] Pappas G, Queen S, Hadden W, Fisher G. The increasing disparity in mortality between socioeconomic groups in the United States, 1960 and N Engl J Med 1993;329(2):103e9. [3] Kaplan GA, Keil JE. Socioeconomic factors and cardiovascular disease: a review of the literature. Circulation 1993;88(4 Pt 1):1973e98. [4] Rose G, Marmot MG. Social class and coronary heart disease. Br Heart J 1981; 45(1):13e9. [5] Braveman PA, Cubbin C, Egerter S, Chideya S, Marchi KS, Metzler M, et al. Socioeconomic status in health research: one size does not fit all. JAMA 2005; 294(22):2879e88. [6] Mueller C, Parcel T. Measures of socioeconomic status: alternatives and recommendations. Child Dev 1981;52(1):13e30. [7] Haan M, Kaplan GA, Camacho T. Poverty and health. Prospective evidence from the Alameda County Study. Am J Epidemiol 1987;125(6):989e98. [8] Diez Roux AV, Merkin SS, Arnett D, Chambless L, Massing M, Nieto FJ, et al. Neighborhood of residence and incidence of coronary heart disease. N Engl J Med 2001;345(2):99e106. [9] Sundquist K, Winkleby M, Ahlen H, Johansson SE. Neighborhood socioeconomic environment and incidence of coronary heart disease: a follow-up study of 25,319 women and men in Sweden. Am J Epidemiol 2004;159(7):655e62. [10] Stjarne MK, Fritzell J, De Leon AP, Hallqvist J. Neighborhood socioeconomic context, individual income and myocardial infarction. Epidemiology 2006; 17(1):14e23. [11] Diez Roux AV, Mair C. Neighborhoods and health. Ann N Y Acad Sci 2010; 1186:125e45. [12] Gerber Y, Benyamini Y, Goldbourt U, Drory Y. Neighborhood socioeconomic context and long-term survival after myocardial infarction. Circulation 2010; 121(3):375e83. [13] Gerber Y, Weston SA, Killian JM, Therneau TM, Jacobsen SJ, Roger VL. Neighborhood income and individual education: effect on survival after myocardial infarction. Mayo Clin Proc 2008;83(6):663e70. [14] Tonne C, Schwartz J, Mittleman M, Melly S, Suh H, Goldberg R. Long-term survival after acute myocardial infarction is lower in more deprived neighborhoods. Circulation 2005;111(23):3063e70. [15] Box-Steffensmeier JM, Jones BS. Event history modeling. Cambridge (UK): Cambridge University Press; [16] Gerber Y, Koton S, Goldbourt U, Myers V, Benyamini Y, Tanne D, et al. Poor neighborhood socioeconomic status and risk of ischemic stroke after myocardial infarction. Epidemiology 2011;22(2):162e9. [17] Burck L, Feinstein Y. Characterization and classification of geographical units by the socio-economic level of the population, series of 1995 census of population and housing publications. Jerusalem (Israel): Central Bureau of Statistics; [18] Molshatzki N, Drory Y, Myers V, Goldbourt U, Benyamini Y, Steinberg DM, et al. Role of socioeconomic status measures in long-term mortality risk prediction after myocardial infarction. Med Care 2011;49(7):673e8. [19] Kelly P, Lim L. Survival analysis for recurrent event data: an application to childhood infectious diseases. Stat Med 2000;19(1):13e33. [20] Therneau TM, Grambsch PM. Modeling survival data extending the Cox model. New York: Springer; [21] Nelson WB. Recurrent events data analysis for product repairs, disease recurrences, and other applications. 1st ed. Philadelphia (PA): ASA-SIAM; [22] Cox DR. Regression analysis and life table. J R Stat Soc B 1972;34:187e222. [23] Andersen PK, Gill RD. Cox s regression model for counting processes: a large sample study. Ann Statist 1982;10(4):1100e20. [24] Wei L, Lin D, Weissfeld L. Regression analysis of multivariate incomplete failure time data by modeling marginal distributions. J Am Stat Assoc 1989; 84(408):1065e73. [25] Prentice RL, Williams BJ, Peterson AV. On the regression analysis of multivariate failure time data. Biometrika 1981;68:373e9. [26] Ezell ME, Land KC, Cohen LE. Modeling multiple failure time data: a survey of varianceecorrected proportional hazards models with empirical applications to arrest data. Sociol Methodol 2003;33(1):111e67. [27] Kleinbaum D, Klein M. Survival analysis. New York: Springer; [28] Lee E, Wang J. Statistical methods for survival data analysis. 3rd ed. Hoboken (NJ): Wiley; [29] R Development Core Team. R: a language and environment for statistical computing. Vienna (Austria): R Foundation for Statistical Computing; [30] Rubin D. Multiple imputation for nonresponse in surveys. New York: John Wiley & Sons, Inc; [31] Winkleby M, Sundquist K, Cubbin C. Inequities in CHD incidence and case fatality by neighborhood deprivation. Am J Prev Med 2008;32(2):97e106. [32] Rondeau V, Mathoulin-Pelissier S, Jacqmin-Gadda H, Brouste V, Soubeyran P. Joint frailty models for recurring events and death using maximum penalized likelihood estimation: application on cancer events. Biostatistics 2007;8(4): 708e21. [33] Cooper R, Cutler J, Desvigne-Nickens P, Fortmann SP, Friedman L, Havlik R, et al. Trends and disparities in coronary heart disease, stroke, and other cardiovascular diseases in the United States: findings of the national conference on cardiovascular disease prevention. Circulation 2000;102(25):3137e47. [34] Hamer M, Molloy GJ, Stamatakis E. Psychological distress as a risk factor for cardiovascular events: pathophysiological and behavioral mechanisms. J Am Coll Cardiol 2008;52(25):2156e62. [35] Shemesh E, Yehuda R, Milo O, Dinur I, Rudnick A, Vered Z, et al. Posttraumatic stress, nonadherence, and adverse outcome in survivors of a myocardial infarction. Psychosom Med 2004;66(4):521e6. [36] Gerber Y, Koren-Morag N, Myers V, Benyamini Y, Goldbourt U, Drory Y. Longterm predictors of smoking cessation in a cohort of myocardial infarction survivors: a longitudinal study. Eur J Cardiovasc Prev Rehabil 2011;18(3):533e41. [37] Gerber Y, Myers V, Goldbourt U, Benyamini Y, Drory Y. Neighborhood socioeconomic status and leisure-time physical activity after myocardial infarction: a longitudinal study. Am J Prev Med 2011;41(3):266e73. [38] Spertus J. Broadening our understanding of survival after myocardial infarction: the association of neighborhood with outcomes. Circulation 2010; 121(3):348e50.