Efficient Nonparametric Estimation from Kin–Cohort Data

Abstract

In a Kin–Cohort design, genotype data are first obtained from a sample composed mostly of individuals who have experienced onset of a disease. Disease history data are then obtained from their relatives. The design is useful when examining the conditional distribution, given genotype, of phenotypes such as disease severity scores when some genotypes are rare. Here, the problem of combining the genotype data in the probands with the phenotype information in their relatives is considered. A class of unbiased estimators is described, the optimal member which reaches the semiparametric efficiency bound is identified, and results from simulation experiments are discussed.