Analysis of the Effect of Pollutants on Respiratory Infection Using Case-Crossover Design and Generalized Additive Models

• Activity Number: 291
• Type: Contributed
• Date/Time: Tuesday, July 31, 2007 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Long H. Ngo*+ and Dzung V. Do and Thuan Q. Thach and Sumi Mehta and Aaron J. Cohen
• Companies: BIDMC & Harvard Medical School and University of Medicine and Pharmacy and University of Hong Kong and Health Effects Institute and Health Effects Institute
• Keywords: pollutants ; ALRI ; case-crossover ; Poisson regression ; generalized additive mixed model

Abstract

In a study to evaluate the effect of ambient pollutants (e.g., PM10) on acute lower respiratory infection (ALRI) in the children of Ho Chi Minh City in Vietnam, we used case-crossover and Poisson regression to analyze the data from 2003 to 2005. We evaluated both the symmetric bi-directional design, and time-stratified design for the case-crossover analysis. Generalized additive mixed models were used for the Poisson models. Systematic missing data of the pollutants and measurement error due to misclassification of diagnosis and exposure estimated from the central monitors, and multicollinearity among the pollutants posed problems for inference. We hypothesized that poverty status, using treatment fee, acted as a modifier on the association between pollutants and ALRI. We explored the issues of non-random missingness, small sample size, and misclassification in both methods of analysis.