A Full-Likelihood Derivation of Conditional Methods for Matched Case-Control Studies, with Application to Misclassification of Exposure

• Activity Number: 177
• Type: Contributed
• Date/Time: Monday, August 4, 2003 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Kenneth M. Rice*+
• Companies: Medical Research Council
• Keywords: conditional likelihood ; matched case-control study ; Rasch model ; misclassification ; measurement error ; random-effects model

Abstract

Conditional likelihood is standard for the estimation of a common odds ratio in a matched case-control study, but lacks the flexibility of full-likelihood methods. We give a new analysis where the strata-specific parameters come from a mixing distribution. We derive conditions on this mixing distribution such that the integrated likelihood is always exactly equivalent to the well-known conditional likelihood, hence deriving it as a full likelihood. Examples of such 'invariant' mixing distributions with attractive symmetric properties are given for case-control studies and the general Rasch model. Potential applications include measures of fit for the conditional model, justification of the use of prior knowledge in conditional analyses, analysis of complex functions of Rasch parameters, and use of MCMC for exact analysis in conditional logistic regression. Our primary application is to allow for misclassification of exposure in a case-control study. The usual conditional analysis fails when this occurs, but the equivalent mixed model allows extension to include the effects of measurement error. We illustrate this in a study of a mismeasured genetic risk factor for cancer.