Study Population
Using Medicare data for persons aged 65 and older for the years 1985 to 1999, we constructed a cohort of survivors with a specific condition we hypothesized might render subjects at greater risk, defining cases as an emergency admission for a primary or secondary discharge diagnosis of chronic obstructive pulmonary disease (COPD, ICD-9: 490–496, except 493); we excluded subjects who died within three months of their admission.
Medicare data provide also the date of death for those subjects who die, and therefore we could define whether they were still alive as of the end of 1999; the date of death field is derived from the Medicare enrolment database and cross-referenced with the Social Security Administration's Master Beneficiary Record [22].
Medicare provide also information on age, gender, race, number of coronary and medical intensive care days, and on factors that might modify the risk of survival, such as primary or secondary diagnoses of atrial fibrillation (ICD-9: 427.3), myocardial infarction (MI, ICD-9:410), diabetes (ICD-9: 250), congestive heart failure (CHF, ICD-9: 428), and essential hypertension (ICD-9: 401) on previous admissions, or whether these were noted as secondary diagnoses on the index admission. We defined a categorical variable for type of COPD based on ICD codes (ICD-9: 491,492,496).
Subjects alive the first of January of the year following the admission were entered into the cohort, and follow-up periods were calendar years. We excluded subjects whose death or subsequent admission occurred within the first three months of their index admission, and those who were admitted in 1999.
During the years 1985 to 1999 survival may have improved due to changes in therapy, underlying disease state, etc. To control for these changes, we used strata to allow a different underlying hazard for each 5-year interval in the study.
City characteristics as population density, percentage of population with central air conditioning, and percentage of population 65 and older with income > $50,000 were obtained from the 1990 United States census [23]; the average annual mortality rates for emphysema among people = 65 years old were obtained from the National Center for Health Statistics.
Environmental Data
We obtained PM10 (particulate air matter with aerodynamic diameter less than 10 μm) data from US Environmental Protection Agency's Aerometric Information Retrieval System [24] for the years 1985 to 1999.
We selected thirty-four cities with daily monitoring of particulate matter and representing a geographic distribution across the country (Figure 1).
When more than one monitor was available in one county, PM10 was averaged over the county using a method previously described [25, 26]. Briefly, we computed local daily mean concentrations using an algorithm that accounts for the different monitor-specific means and variances. However, before averaging, any monitor that was not well correlated with the others (r<0.8 for 2 or more monitor pairs within a community) was excluded as it likely measured a local pollution source and would not represent the general population exposure over the entire community.
For each subject and follow-up period we created yearly averages (January-December) of pollution for that year and up to the 3 previous years.
PM10 was then treated as a time varying covariate in the survival analysis.
Statistical Methods
To define the cohort we assumed that each subject entered the cohort if he/she survived at least 3 months and was alive on the first January of the year following the admission. For each subject the follow up periods were 1 year periods (January – December) until the year in which they die (failure) or until December 1999 (censoring). This method has been previously described [21].
We analyzed the data with an extended Cox's proportional hazard regression model (Proc PHREG in SAS [27] which allow for time-varying covariates in survival analysis [28], as previously described [21].
We controlled for individual risk factors such as age, gender, race, season of admission, number of days of coronary and medical intensive care, previous diagnoses for atrial fibrillation and MI, and secondary or previous diagnoses for diabetes, CHF, and hypertension, time period, and season.
Season of the index discharge that defined entry into the cohort was defined as: cold (December through February), hot (June through August), and transitional. To allow for possible non-proportionality of the survival rates, time period (3 categories, one for each 5 year increment), age (5 year categories), gender, race (white, black, others), and type of COPD were treated as stratification variables.
To control for tied observations we used the appropriate likelihood function as given by Kalbfleisch and Prentice [29].
In the second stage of the analysis, the city specific results were combined using a random effect meta-regression [30]. To be conservative we report the results incorporating a random effect, whether or not there was significant heterogeneity.
We examined effect modification by city characteristics by entering them as predictor variables in the meta-regression. These included measures of socio-economic condition (percent of persons 65 and older with income > $50,000), exposure related measures (mean PM10 in the city), general social factors (population density), and the emphysema death rate in persons aged 65 and older as a surrogate for the smoking history of the population.
We also examined effect modification by age group (65–75 vs 76 and over), race (white vs other), and sex by including the interaction terms between the yearly PM10 and each variable in the model.
For each subject in each follow-up period, we considered the following possible exposure indexes: the average PM10 in their city in that follow-up period, and a model containing simultaneously the exposure during the follow-up period and each of the three previous years (distributed lag), to see if we could determine how the PM effect dropped off over time. We also computed the sum of the PM10 effect from lag 0 to the three previous years. The results are expressed as Hazard ratio (HR) for a 10 μg/m3 increment of PM10.