Improved Efficiency in Multiphase Case-Control Studies

• Activity Number: 559
• Type: Invited
• Date/Time: Wednesday, August 4, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Christopher J. Wild*+ and Alastair Scott
• Companies: University of Auckland and University of Auckland

Abstract

In a recent paper (Lee, Scott & Wild, Biometrika, 2010), we developed efficient methods for fitting regression models to multi-phase case-control data in the special case where all covariates, apart from those measured in the final phase, are categorical. The method can be implemented by taking a linear combination of the estimating equations obtained by applying conditional maximum likelihood (see Breslow & Cain, Biometrika, 1988, 11-20) to the completely-observed units and equations similar to those used in calibrating (see Sarndal, Survey Methodology,2007, 99-119) the sampling weights to quantities known for the partially-observed units. We can extend this procedure to get very good procedures for situations where implementable optimal solutions are not available, for example when continuous covariates are measured in earlier phases or when we have appreciable non-response.