Correcting for baseline differences in the comparison of rates

Abstract

Statistical adjustment of baseline differences (bias attributable to confounding covariates) in the comparison of rates is often carried out by the stratification approach (e.g. the "direct standardization method"). However, this approach is usually not feasible for simultaneous adjustment of more than one confounding covariates given modest sample size. Also, the stratification approach requires grouping a continuous adjustment variable into broad categories, and consequently, residual confounding of the adjusted rates can still occur. These drawbacks can be overcome by multiple regression. This communication considers a statistical procedure for the comparison of occurrence rate of some event across two or more exposure or treatment groups adjusting for one or more confounding covariates. The procedure is based on the multiple logistic regression model. A detailed numeric example to illustrate the application of the method is presented. A computer program to carry out the statistical procedures is available from the authors. © 1987.