|
Abstract:
|
In environmental health studies of time series data where the daily count of a particular disease outcome, and daily averages of air pollutants are available, the estimate of excess risk per unit of pollutant is obtained via the use of a generalized additive model with Poisson link, and smoothing functions for temperature, humidity, and temporal trend. This type of model can be fitted using a back-fitting algorithm (BF), or penalized iteratively reweighted least squares (PIRLS). Using a data set where the outcome is the daily count of children in Ho Chi Minh City who were admitted to the hospitals for acute lower tract respiratory infection, and daily average of PM10, ozone, nitrogen dioxide, sulfur dioxide, we used both methods of estimation. We used SAS/STAT software GAM for BF, and R function GAM from the package MGCV. We found noticeable differences between two estimation methods.
|
