Problems in using age-stratum-specific reference rates for indirect standardization.

Disease risk in a study cohort can be compared with that in a reference population using the method of indirect standardization, adjusting for a difference in age distribution between the cohort and reference population. The common epidemiological practice is to use categorical age, typically in the form of 5-year age strata, in the standardization. This article discusses problems that arise owing to the categorization of age, including biased estimation and incorrect statistical inference on relative risk parameters. The same problems further extend to more general analyses using Poisson regression. We illustrate the problems using a hypothetical example and propose a simple remedy using linear splines. A slightly more elaborate method and its computer program are given in the appendix.