We evaluated the association between tertiles (low, medium, and high) of census tract-level neighborhood household income (nINC) and lipid-lowering medications received during hospitalization or at discharge among 3,546 (5,335 weighted) MI events in the US Atherosclerosis Risk In Communities (ARIC) surveillance study (1999–2002). The ARIC study’s community-based surveillance of coronary heart disease has been ongoing since 1987 and is designed to capture MI and fatal coronary heart disease events in four US communities [Jackson, Mississippi (MS); Forsyth County, North Carolina (NC); Washington County, Maryland (MD); and Minneapolis, Minnesota (MN)]). While it comprises the same communities from which ARIC cohort members were recruited, ARIC community surveillance does not include in-person physical exams, annual follow-up, or any contact with ARIC cohort participants (unless they happen to be sampled as a surveillance case). ARIC community-surveillance staff ascertained coronary heart disease-related hospital discharges and deaths and abstracted data related to the event of interest. Institutional Review Board (IRB) approvals were obtained by each participating ARIC study center (the Universities of NC, MS, MN, and John Hopkins University) and the coordinating center (University of NC), and the research was conducted in accordance with the principles described in the Declaration of Helsinki. Data for this study were abstracted from medical records and strict data confidentiality was maintained. Further details regarding ARIC’s methods for data collection are provided elsewhere
. For our analyses, we weighted the hospitalized MI cases based on the probability sampling of selected International Classification of Disease codes
 in order to estimate the eligible population of cases that would have been studied had the probability sampling not been employed.
We estimated prevalence ratios and 95% confidence intervals for receipt of lipid-lowering medication post-MI using weighted Poisson regression, and we used generalized estimation equations (PROC GENMOD, SAS Institute) to account for the clustering of MI events within census tracts and within patients, as incident and recurrent MI events were considered together based upon our previous analyses
[7, 8]. Model 1 included nINC, race, gender, age, study community and year of MI. Model 2 included the same parameters as well as hospital type (teaching vs. nonteaching), current or past history of hypertension, diabetes or heart failure, and presence of cardiac pain. Additionally, we examined whether the following parameters modified the nINC – lipid-lowering therapy relationship: race, gender, age, study community and year of MI. Models utilizing tertiles defined by overall nINC cut-points were evaluated, as interpretations of our earlier work in this population did not change based on the delineation (community-specific, race-specific, and overall cut-points) of nINC tertiles