Thursday, April 22, 11:00-12:00

Many epidemiologists have suggested that exposure to particulate air pollution results in an immediate 

increase in the risk of acute health events such as mortality. The risk is transient, since it is thought to

subside upon cessation of exposure.  Several methods have been proposed to estimate the magnitude 

of this risk; these include Poisson regression and the case-crossover design.  The case-crossover 

design is somewhat more generally applicable than Poisson regression, and when the assumptions of

Poisson regression are satisfied, both methods yield the same estimate of relative risk.  Since health 

outcomes are affected by many events that are not quantified, these methods must be adapted to 

control for unmeasured confounding.  This can be done by stratification in the case-crossover design,

or equivalently, by the inclusion of dummy variables in Poisson regression.  The method of risk-set 

sampling, first developed for use in occupational studies, can be applied to the case-crossover design 

to provide this control.  This method provides an improvement over earlier stratification methods.