Regression analysis of the log odds ratio: a method for retrospective studies.

Quantification of the dependence of the odds ratio on concomitant variables associated with each of several 2 X 2 tables, using a regression model proposed by Zelen [1971], is an important tool for retrospective studies in epidemiology, An "exact" analysis may be based on the conditional likelihood obtained by fixing all the marginal totals. A symptotically this approach is equivalent to use of an unconditional log-linear model. The method is used to reanalyze data reported by Kneale [1971] on the relationship between obstetric radiation and childhood cancer.