Multivariate imputation in cross-sectional analysis of health effects associated with air pollution

We demonstrate a recently developed spatial interpolation methodology in a study of the chronic effects of air pollution on respiratory morbidity. Our study uses data from the Ontario Health Study, a large survey of households in Ontario conducted for the province by Statistics Canada. The interpolation procedure imputes unobserved vectors of air pollution concentrations for individual Public Health Units, from those observed at a few fixed air pollution monitoring sites. We use logistic regression methods to assess the significance of air pollution levels based on the imputed values after modelling the relationship between binary health responses and assorted covariates such as measures of life style. Our findings prove negative; no significant relationship between chronic respiratory morbidity and air pollution is found. The imputation methodology is seen to be promising and might well be used in other such analyses.