Improved Temporal Smoothing for Estimating Population Health Risk

• Activity Number: 34
• Type: Contributed
• Date/Time: Sunday, August 4, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Wesley Burr*+ and Glen Takahara and Hwashin H. Shin
• Companies: Queen's University, Dept. of Mathematics & Statistics and Queen's University and Health Canada, Population Studies Division
• Keywords: smoothers ; population health risk ; Generalized Additive Models ; filters ; air pollution

Abstract

The Air Health Indicator is a joint Health Canada / Environment Canada initiative that seeks to model the Canadian national population health risk due to short-term effects of air pollution. The commonly accepted model in the field uses cubic spline-based temporal smoothers to account for seasonal and long-term variations in the response. From a spectral point of view, it is natural to think of these smooth, long-term variations as low-frequency components. We apply signal processing techniques to create an improved concentration smoother which "plugs into" already existing models in place of the temporal cubic spline smoother. The risk estimates obtained are comparable with previous work, with several valuable insights gained, especially in interpretation of the residual effective response. Further decomposing the pollutant covariate into low- and high-frequency components using strongly concentrated filters, we observe that this split changes the national posterior mean risk, moving it away from zero. We further attribute this behaviour to the iterative nature of the GAM solver, and contrast the residual effective response before and after the introduction of the split.