Attenuation of Statistical Relationships from PM Community Time-Series Epidemiology Due to Use of Combined, Rather Than Separate, Indicators of Exposure and Mortality
Archived files are provided for reference purposes only. The file was current when produced, but is no longer maintained and may now be outdated. Persons with disabilities having difficulty accessing archived files may contact the NCEA Webmaster for assistance. Please visit http://epa.gov/ncea to access current information.
Attenuation of the statistical relationships between PM and health outcomes may arise from 1) combining exposure indicators, e.g., PM10 instead of PM2.5 and PM10-2.5 or 2) from combining different types of mortality. The Phoenix, AZ data base on air quality offers an opportunity for examining these issues. The initial data reported for Phoenix has been expanded by adding a PM1 time series and using imputation techniques to estimate everyday values for the PM mass indicators. Several modifications have been made in the computational approach that have improved the statistical significance of the results. This paper will examine the differences in excess risk obtained by 1) using PM10 alone, 2) using individual values of PM2.5 and PM10-2.5 as individual explanatory variables, and 3) using PM2.5 and PM10-2.5 together in a multiple regression. In the original study, the PM10-2.5 effect peaked on the day of exposure (lag day 0) and the PM2.5 effect peaked on the day after exposure (lag day 1). Therefore, the paper will include individual and multiple regressions using lag days 0 and 1. Analysis will be reported for total (non-accidental), and cardiovascular mortality. The results show that, in most cases, going from combined parameters (combining two or more exposure variables or causes of mortality) to more specific indicators leads to greater statistical significance and greater excess riskss.