Jake Olivier, The University of New South WalesMarch 29, 2012 @ 3:30 - 4:30 pm
Location: Blockley Hall - Room 701
- On The Use of Population Offsets for
Estimating Incidence Rate Ratios
Many epidemiological studies contrast the rate of occurrence of an outcome such as injury mechanism or disease notification between demographic groups through estimation of the incidence rate ratio (IRR). Proper estimation of the IRR relies on the identification of cases and the quantification of exposure. Exposure can take on many forms such as person years, number of kilometers travelled or population size. A lot of effort is put into the identification of cases through hospitalization data and disease specific registries; however, little effort is devoted to quantifying exposure. The use of population size as a proxy measure of exposure is often used in lieu of more appropriate measures when they do not exist or are difficult to ascertain. Within the framework of log-linear regression, exposure is accounted for through the use of an offset. Population size offsets do account for the dynamic nature of population demographics; however, their use does make the implicit assumption that everyone in a demographic group is equally exposed. The effect this assumption has on the IRR, estimated through Poisson regression with a binary variable to indicate groups of varying incidence, is explored through a simulation study. The results demonstrate the IRR can be underestimated when the exposure rate is greater in the group with lower incidence; whereas, the IRR is overestimated when the exposure rate is greater in the group with higher incidence. For demonstrative purposes, model estimates using population size and measures of exposure are contrasted using drowning morbidity data. This work suggests IRR estimates that use population size offsets should be interpreted with caution.