IMPROVED ODDS RATIO ESTIMATION BYPOST HOC STRATIFICATION OF CASE-CONTROL DATA

Abstract

We propose a logistic regression analysis of unmatched or frequency matched case-control studies with conditional maximum likelihood estimation through post hoc stratification. In this model fewer parameters have to be estimated. With a simulation study we show that parameter estimates have smaller variance and are less biased. Also, the residual confounding effect was quantified. A more refined post hoc stratification reduces computing time, but to the cost of a larger bias and a loss in efficiency. The model was also applied to data of unmatched case-control studies on laryngeal cancer, oesophageal cancer and lung cancer. © 1997 by John Wiley & Sons, Ltd. Stat. Med., Vol. 16, 993–1004 (1997).