A Review of Statistical Methods Used in Time-Series Epidemiologic Studies of Ambient Particulate Matter and Acute Health Effects Cited by the June 2003 EPA Draft PM Criteria Document

The fourth draft of the Particulate Matter [PM] Criteria Document [DCD4] was issued by EPA in June 2003. DCD4 examines a number of statistical issues relevant to connections between ambient PM variations and corresponding variations in health and mortality indicators. The discussion centers on relating temporal variations through time series modeling, and on relating geographic variations using cohort studies. Although DCD4 comments critically on a number of issues, its conclusions seem to reflect an unwarranted leap over many of the criticisms that it, itself, has raised. In addition, there are several issues not adequately addressed in DCD4, and DCD4 is sometimes completely uncritical of the literature that it cites in support of its conclusions. In this review I address several statistical issues in DCD4 and indicate how they affect conclusions drawn about the health effects of PM. The issues addressed here are grouped into the following categories. 1. Confounding of weather and PM effects 2. Confounding of time trends and PM effects 3. Heterogeneity of PM effects and effect modification 4. Heterogeneity of exposure 5. The relation between exposure and response 6. Lag selection and distributed-lag models 7. Mortality displacement 8. Long-term PM-mortality studie