Nonparametric Regression and the Secondary Analysis of Case-Control Studies

• Activity Number: 380
• Type: Invited
• Date/Time: Tuesday, August 2, 2011 : 2:00 PM to 3:50 PM
• Sponsor: IMS
• Author(s): Raymond James Carroll*+ and Jiawei Wei
• Companies: Texas A & M University and Texas A & M University
• Keywords: Case-control studies ; Nonparametric regression ; Semiparametric regression ; Biased sampling

Abstract

Case-control studies collect information on case-control status and covariates. Increasingly, investigators have wished to exploit case-control data sets to explore relationships among the covariates themselves. Thus, case-control status is D, and covariates are (Y,X), and the goal is to make inference about the relationship of Y and X. One cannot simply use the (Y,X) data directly because of the biased sampling caused by the case-control design, although with rare diseases regression of Y on X among the controls is essentially unbiased. The goal though is not to throw away the case data, but to exploit it for more efficient inference. Efficient semiparametric approaches exist if there is a parametric model relating Y to X, but these approaches can lead to considerable bias if the model is incorrectly specified. We develop parametric and nonparametric methods that are robust to model misspecification and have much greater efficiency than analysis of the control data only.