A program running under MS-DOS for the analysis of epidemiologic stratified or matched data

Abstract

Breslow and Day's method for computing the conditional likelihood of the logistic model in case of stratified samples requires, at each iteration the generation, for each stratum, of all the permutations of nc elements drawn from a total of nc + nt elements, where nc and nt are respectively the numbers of the cases and of the controls. This procedure rapidly proves costly in time as nc and nt increase, and also when nc and nt differ from one stratum to the next. The use of this technique on microcomputers is therefore limited. Presented here is an algorithm written in FORTRAN, based on a binary representation of the possible permutations, which readily permits the implementation of Breslow and Day's model on a microcomputer.