A new semiparametric procedure for matched case-control studies with missing covariates

Abstract

In this paper, we propose an easy-to-use semiparametric method for analysing matched case-control data when one of the covariates of interest is partially missing. Missing covariate information in matched case-control studies may create bias and reduce efficiency of the parameter estimates. In order to cope with this situation we consider a robust approach which is comprised of estimating some functionals of the distribution of the partially missing covariate using a kernel regression technique in a conditional likelihood framework. The large sample theory of the proposed estimator is investigated and the asymptotic normality is obtained. A simulation study is conducted to assess the performance of the proposed method in terms of robustness and efficiency. The proposed method is also applied to a real dataset which motivates this work.

Handling Missing Data in Matched Case-Control Studies Using Multiple Imputation

“…His method assumes the partially observed covariates are all observed or all missing on each individual. Sinha and Wang (2009) take a similar approach, but instead of a parametric covariate model, kernel density estimation is used for those functions in the estimating equations that depend on the distribution of the partially observed covariate. They find their OR estimator is less biased than that of Paik and Sacco (2000) when the latter's covariate model is misspecified.…”

An estimated‐score approach for dealing with missing covariate data in matched case–control studies

“…The Los Angeles Endometrial data (Breslow & Day, 1980) have been analyzed in many articles, such as, Satten & Carroll (2000), Sinha & Maiti (2008), and Sinha & Wang (2009), among others. In order to study the effect of several risk factors on endometrial cancer, a study was conducted among post‐menopausal women in an affluent retirement community of Los Angeles.…”

“…Recently Sinha & Wang (2009) proposed a method for handling missing covariate data where they used a kernel density approach instead of using a parametric model for the distribution of the partially missing covariate. Although the article made an important advancement in this field, there are certain difficulties in the application of the method.…”

“…In this article we propose an easily implemented nonparametric method which is robust to the distribution of the partially missing covariate. The theoretical justification and practical implementation of this method is much more straightforward than Sinha & Wang (2009). Unlike kernel‐based nonparametric approach, we do not need to deal with the issue of bandwidth selection, boundary effects, or occasional negative estimate of $\hbox{pr}(Y=1|{\bf Z}, {\bf S})/\hbox{pr}(Y=0|{\bf Z}, {\bf S})$ .…”