Referent selection in case-crossover analyses of acute health effects of air pollution

The case-crossover design was proposed for the study of a transient effect of an intermittent exposure on the subsequent occurrence of a rare acute on set disease. This design can be an alterative to Poisson time series regres sion for studying the health effects of fine particulate matter air polluti on. Characteristics of time-series of particulate matter, including longter m time trends, seasonal trends, and short-term autocorrelations, require th at referent selection in the case-crossover design he considered carefully and adapted to minimize bias. We performed simulations to evaluate the bias associated with various referent selection strategies for a proposed case- cross-over study of associations between particulate matter and primary car diac arrest. Some a priori reasonable strategies were associated with a rel ative bias as large as 10%, but for most strategies the relative bias was l ess than 2% with confidence interval coverage within 10% of the nominal lev el. We show that referent selection for case-crossover designs raises the s ame issues as selection of smoothing method for time series analyses. In ad dition, conditional logistic regression analysis is not strictly valid for some case-crossover designs, introducing further bias.