MCAPS is a retrospective study of a cohort of 13.2 million persons ≥ 65 years of age enrolled in the U.S. Medicare system during the 6-year period 2000–2005. To create the cohort, we used the Medicare enrollment file for the study period, which provides a listing of all Medicare enrollees, along with demographic information (age, race, and sex) and ZIP code of residence. New participants enter each year as they enroll in Medicare, making this a “dynamic cohort.”
More specifically, the cohort consists of all those ≥ 65 years of age who enrolled in Medicare between 2000 and 2005 with ZIP code centroids within 6 miles of a U.S. EPA PM2.5 monitoring station. Although the Social Security Administration maintains the addresses of those enrolled in Medicare, the Center for Medicaid and Medicare (CMS) provides an annual report of Medicare enrollees by ZIP code (often referred to as the enrollee file). Medicare enrollees enter the cohort on reaching their 65th birthday or on 1 January 1999 should they be ≥ 65 on that date. A small number of individuals enroll in Medicare the year after their 65th birthday, and those individuals enter the cohort on January 1 of the year of their enrollment. Individuals contribute time to the cohort until they die or are otherwise censored. Censorship occurs when individuals move to a ZIP code > 6 miles from a U.S. EPA PM2.5 monitoring station or are no longer reported in the enrollee file. We calculated age-specific mortality rates as the total number of deaths occurring within an age group and ZIP code divided by the total person-years contributed by that age group and ZIP code.
We obtained the date of death from the CMS. The date of death is provided to CMS by the Social Security Administration, rather than by the National Center for Health Statistics (NCHS), which maintains the national death certificate system. To validate the mortality data from the CMS, we compared annual age- and sex-adjusted mortality rates from the CMS with the corresponding rates calculated from NCHS data for the 250 largest counties for the year 2000. The correlation coefficient was 0.998, indicating a high level of agreement between the two sources of mortality data aggregated to the county level—the finest partition available from the NCHS—for the 1-year period.
For this article, the outcome measure is the 6-year (2000–2005) mortality rate for persons residing within each of 4,568 ZIP codes for each of three age strata: 65–74, 75–84, and ≥ 85 years of age.
We obtained the PM2.5
data from the U.S. EPA’s AirData database (http://www.epa.gov/oar/data/
), which included 1,006 monitors for the period 2000–2005. We calculated mean annual PM2.5
values for the study period for all 4,568 ZIP codes with centroids within 6 miles of a monitor with > 10 months of data per year. If three or more observations were available for a month, we considered this amount of data sufficient because PM concentration was measured every sixth day at many locations. Because the focus of this study was to estimate the effect of long-term exposure to PM2.5
, we used a ZIP code 6-year average of PM2.5
as a measure of the long-term exposure to PM 2.5
for an individual living within a ZIP code both during the 6 years of follow-up and for some time before cohort enrollment. We omitted the 1999 PM2.5
data because this was the initial year of the U.S. EPA monitoring program and coverage was limited.
An advantage of MCAPS is that it comprises persons ≥ 65 years of age from nearly all of the major urban ZIP codes in the United States, and large numbers of deaths are reported within each age stratum and region. We have therefore estimated the age- and region-specific relative risks of chronic PM2.5 exposure for a) the eastern region of the United States, with 2,938 ZIP codes in 421 counties; b) the central region, with 990 ZIP codes within 185 counties located between the Mississippi River and the Sierra Nevada range; and c) the western United States, with 640 ZIP codes within 62 counties extending from Washington State to Southern California. shows the location of the 4,568 ZIP code centroids, the three geographic regions, and the spatially smoothed levels of the 6-year average PM2.5. These spatially smoothed PM2.5 levels should be interpreted with caution because of the sparseness of monitors in some areas.
Map of spatially smoothed averages of PM2.5 during the study period 2000–2005. The map also indicates 4,568 ZIP code centroid locations (black circles) and western, central, and eastern U.S. regions.
We conducted the analyses separately within each of these three geographic regions and for three distinct age strata: 65–74, 75–84, and ≥ 85 years of age. We also stratified initial analyses by sex and by the ZIP codes that were above and below the national median for education and income variables. Because the estimated effects for men and women and for high- and low-SES subgroups were very similar, we did not stratify the analyses reported here by sex or SES. The results of these stratified analyses are available in the Supplemental Material, (online at http://www.ehponline.org/members/2008/11449/suppl.pdf
Numbers of ZIP codes, counties, monitoring sites, Medicare enrollees, person-years of follow-up, deaths, and crude death rates stratified by region and age group for MCAPS.
In estimating the effect on mortality of PM or other air pollutants, previous cohort studies and this new study rely entirely on cross-sectional comparisons of covariate-adjusted mortality rates across geographic locations with different PM levels, because PM is not time varying in the analyses. Previous studies have accounted for potential confounding by a) individual-level lifestyle factors, including age and smoking, and b) area-level characteristics such as county-level SES. The MCAPS provides individual-level age, sex, and race data but not data on lifestyle factors. To account for SES at the ZIP code level, we used age-specific SES variables from the 2000 U.S. Census. After preliminary analysis, we selected five SES variables at the ZIP code level from the U.S. Census Bureau’s Summary File 3. We restricted the analysis to those enrollees who report ZIP codes to CMS that correspond to ZIP code tabulation areas recognized by the U.S. Census Bureau. We selected two education variables, percentage of the population with a high school diploma and the percentage with a higher education degree, along with two household income measures, percentage of households living below the poverty level and median household income, as well as percentage unemployed. To create a univariate measure of SES by which to stratify the analysis, we averaged the ranks of the five SES variables for each county.
Previous cohort studies have found little effect of adjusting for self-reported smoking status (Krewski et al. 2000
). Area-level differences in cigarette smoking, however, could potentially confound the association between PM2.5
and mortality. Because the MCAPS data have neither individual- nor area-level smoking information, we used data from the NCHS to calculate the standardized mortality ratio (SMR) for chronic obstructive pulmonary disease (COPD) for the period 1993–2002, adjusted for age, race, and sex for each county. Because the vast majority of deaths from COPD in the United States are attributable to smoking (U.S. Department of Health and Human Services 2004
), we used the SMR for COPD as a surrogate indicator of the long-term smoking pattern of its residents. We included the county-level COPD SMR in the regression model, assigning the county value to all ZIP codes within a county.
For exposure, reliance on ZIP code–level rather than county-level PM concentration is a strength, but person-level covariate information is unavailable. To assess the potential consequences of imperfect control for confounding variables, we estimated the main models with three levels of adjustment: no control for ZIP code–level confounders, control for ZIP code–level SES variables, and control for ZIP code–level SES and county-level COPD SMRs.
where Yi, Ni, Zi, and Xi are the number of deaths, number of person-years at risk, PM2.5, and SES and COPD SMR for ZIP code i. The parameter βPM denotes the log relative risk of mortality associated with a 1-μg/m3 difference in average PM2.5 comparing ZIP codes that are otherwise similar with respect to SES and COPD SMR.
We report results for each region by age stratum and aggregated over the three age groups. To obtain the aggregated value, we fit a single log-linear regression with a common PM effect across the strata. We use generalized estimating equations (Diggle et al. 2002
) to account for the correlation among age groups from the same ZIP code.
We carried out all analyses with the statistical programs R (R Development Core Team, Vienna, Austria) and SAS (version 9.1; SAS Institute Inc., Cary, NC). Programs are available from the authors.