Individual education, area income, and mortality and recurrence of myocardial infarction in a Medicare cohort: the National Longitudinal Mortality Study

The NLMS is a prospective study of mortality occurring in combined samples of the non-institutionalized United States population drawn from the Census Bureau Annual Social and Economic Supplement (ASEC) ascertained in 1973 and annually from 1978-1998 [15]. Each survey sample is a complex, national probability sample of households surveyed to obtain demographic, economic, and social information about the U.S. population. During the household interview, information was collected on each occupant regarding age, sex, race, Hispanic status, educational attainment, and other socio-demographic characteristics.

Mortality follow-up information was collected through a computer match to the National Death Index (NDI) through 2002. The NDI, a national file containing information collected from death certificates, is maintained by the National Center for Health Statistics. The matching of records to the NDI has been shown to be an effective and accurate means of ascertaining mortality information using personal identifiers including: social security number, name, date of birth, sex, race, marital status, state of birth, and state of residence [15, 16]. Each record match to the NDI is classified as either a true match, false match, or questionable match. All questionable matches are adjudicated by hand by comparing information on the death certificate to the matched ASEC record.

Morbidity and health care encounter data was obtained through record-linkage to the Centers for Medicare & Medicaid Services (CMS) Medicare database. All NLMS persons with suitable identifiers were linked to Medicare data from 1991-2001. Of 458,772 NLMS persons that were age and survival eligible for Medicare and for which ASEC sample design information was available, 391,594 (85%) were successfully linked. Those that were not linked to Medicare were similar to those linked to Medicare with respect to sex (56.0% female vs 56.7%), but were slightly younger (61.6 years of age on 1/1/1991 compared with 62.9 years of age), more likely to have an education level less than high school (41.2% vs 36.4%), more likely to have a family income of less $10 K in 1990 dollars (26.6% vs 16.6%), less likely to be married at time of survey (65.6% vs 71.1%), more likely to be black (12.4% vs 8.2%) and more likely to be Hispanic (6.1% vs 3.9%). Since linkage to Medicare was related to SES, a sensitivity analysis was conducted in which the Medicare population was re-weighted adjusting for the probability of Medicare linkage. Medicare linkage probabilities were adjusted by sex, race, ethnicity, family income and marital status. Since the re-weighting was dependent on factors associated with Individual education, original weights will be the focus of the analysis. Enrollment information, hospital admissions, outpatient and physicians visits were extracted for each linked individual. Eligibility for the current study consisted of first hospitalizations with a primary or secondary discharge diagnosis code of 410 according to the ninth revision of the International Classification of Diseases Clinical Modification (ICD-CM 9), and continuous Medicare type A and B coverage for one year prior to the index MI. A primary or secondary discharge diagnosis code of 410 has been shown to have 94% positive predictive value (PPV) for identifying MIs in a Medicare population [17] and a PPV of approximately 75% in the Atherosclerosis Risk in Communities study [18]. Since the year prior to the index MI was utilized for health related covariates, the minimum age for an MI was 66 years and the earliest date for an index MI was January 1, 1992. Exclusions included admissions from skilled nursing facilities, transfers from other health care facilities, and hospitalizations lasting more than 30 days. To isolate incident cases, ICD-CM9 discharge codes 410.x2 were excluded as an indication of a repeat encounter, and persons with any mention of ICD9 codes 410 or 412 (old myocardial infarction) among the physician encounters, outpatient and hospital records for one year prior to the index MI were excluded.

Procedure codes at the index hospitalization were searched for mention of coronary revascularization (Percutaneous Coronary intervention (PCI) 36.00-36.07, Coronary Artery Bypass Grafting (CABG) 36.10 – 3.6.19). Hospitalization, outpatient, and physician visit records for the year proceeding the MI were searched for any mention of a history of CHD (411, 413-414, 429.2), hypertension (401-404), hyperlipidemia (272.0 – 272.4), diabetes (250), or stroke (430 – 438). Sex and race were taken from the ASEC survey as these are self-reported, age in years was calculated from the date of MI and date of birth, and education was categorized into three groups based on the number of years of attained education: less than high school education, high school diploma but no college degree, college degree or more. Area SES was based on the home zip code at time of the index MI as extracted from the hospitalization record. Median income for all US zip codes was extracted from Census Bureau Zip Code Tabulation Areas Summary Tape Files for 1990 and 2000 and divided into quintiles. Zip Code Tabulation Areas are a geographic representation of postal service ZIP code service areas. Infarctions occurring in 1992-1995 were linked to median income quintiles in 1990 and all other MIs were linked to median income quintiles in 2000.

Since the ASEC survey may have occurred 20 years or more prior to enrollment in Medicare, individual measures of SES were limited to education. Due to low numbers, only those self-identifying as non-Hispanic white or black were included in the analysis.

Mortality follow-up was based on the NDI match and recurrence was defined as a subsequent hospitalization with a primary or secondary ICD9 discharge code of 410 or a CHD death defined as an ICD9 underlying cause of death code of 410-414 or ICD10 code of I20-I22, or I24-I25. Any recorded MI within 28 days of the first MI was considered an extension of the first MI and not a recurrent event. Survival follow-up extended through December 31, 2002 while follow-up for recurrence was censored on December 31, 2001. Median follow-up time for survival was 6 years and median follow-up time for recurrence was 5 years.

All analyses were sex specific. To examine potential differential impacts according to follow-up time, survival analyses were conducted in 3 discrete time periods: in-hospital, discharge-1 year, and 1-5 years. Similarly, follow-up for recurrence was separated into 28 day-1 year and 1-5 years. Since all persons had at least one year of follow-up for mortality, log-linear models were used to estimate relative risk associations with in-hospital and discharge-1 year mortality, and proportional hazards regression was used for long term survival and both 28 day-1 year and 1-5 years recurrence. Earlier reports suggested that there may be effect modification by age with respect to SES associations [13]; therefore, analyses were further stratified into age specific groups: 66-79 years and 80+ years. SUDAAN release 10.0.1 was used for all analyses to account for the complex sample design of the ASEC survey.

The ASEC surveys are undertaken to characterize the US civilian, non-institutionalized population in terms of demographics and employment and therefore a traditional informed consent is not utilized. Participation is voluntary and subjects are informed that their data may be linked with other Census data or external data and are provided with the option to opt-out of any data linkage. The NLMS-Medicare study is a data linkage study without any direct interaction with survey respondents, thus the principles of the Helsinki Declaration are not applicable. The Medicare data linkage protocol was reviewed and approved by the Census Data Stewardship Executive Policy group and linkage to NDI further reviewed and approved by the National Association for Public Health Statistics and Information System (NAPHSIS) committee. As part of their review, these committees specifically evaluate the data security practices and participant privacy protections of studies conducting record linkage.

All models were estimated with adjustment for age, race, history of hypertension, hyperlipidemia, diabetes, CHD, Stroke, PCI-CABG at hospitalization, and year of MI. Models were first run including only individual level education, a second run included only zip code level median income quintiles, and finally models were run including both individual level education and zip code median income quintiles. Adjustment for both individual level education and zip code median income quintile only marginally attenuated the risk ratios and hazard rates; therefore, only the fully adjusted results are presented.

As can be seen in Table 2 zip code level median income was not associated with survival in women. The exception, in-hospital mortality in women 66-79 years of age (RR Q1 v Q5 = 1.40; P = 0.037), was not significant in the sensitivity analysis (Additional file 1: Table S1). Individual level education was associated with 1-5 years mortality (HRR: < HS v College Degree + =1.61, P = .035), but not with in-hospital or discharge-1 year mortality. Education and area SES were not associated with survival in women 80+ years.

Similar to women, individual level education was associated with long term survival in men 66-79 years of age (Table 3, HRR: < HS v College Degree + =1.37, P = .017). The lowest quintile of zip code level median income was marginally associated with long term mortality in younger men (HRR Q1 v Q5 = 1.30; P = 0.052) in both the primary analysis and in the sensitivity analysis (Additional file 1: Table S2). Although zip code level median income was associated with discharge-1 year mortality among men 80+ years of age, these associations were not significant in the sensitivity analysis. No association was found between individual level education and short or long-term mortality in men 80+ years of age.

Neither individual level education or zip code level median income were associated with recurrence in women (Table 4); however, education approached statistical significance with long-term recurrence in women 80+ years of age (HRR: < HS v College Degree + =1.52, P = .064). Contrary to expectations, zip code level median income in quintile 4 was associated with lower long-term recurrence (Q4 v Q5 = 0.65; P = 0.009), a potential false positive since all other associations were non-significant or above 1.0.

As shown in Table 5, individual level education was associated with long term recurrence (1-5 years) in men 66-79 years of age (HRR: < HS v College Degree + =1.68, P = .004). No association was found for zip code level median income with short or long-term recurrence in men 66-79 years of age; however, in men 80+ years of age, the lowest quintile of zip code level median income was associated with increased short term recurrence (Q1 v Q5 = 1.72, P = 0.009).