Nonparametric Estimation for Censored Mixture Data with Application to the Cooperative Huntington's Observational Research Trial

• Activity Number: 566
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Tanya Garcia*+ and Yuanjia Wang and Yanyuan Ma
• Companies: Texas A&M University and Columbia University and Texas A&M University
• Keywords: Censored data ; Finite mixture model ; Huntington's disease ; Kin-cohort design ; Quantitative trait locus

Abstract

In kin-cohort studies, interest lies in making inference on genotype-specific distributions using data from a mixture of scientifically meaningful subpopulations. Although the mixing proportions can be obtained, inference is complicated by unobserved genotypes and random right-censoring. Current methods for this situation include two nonparametric estimators which make no assumptions on the associated density function. However, we show one is inefficient and the other inconsistent. Instead, we propose three classes of consistent nonparametric estimators which are robust to misspecification of a density model and easy to implement. They are based on the inverse probability weighting (IPW), nonparametric imputation and augmented IPW. The latter estimator achieves the efficiency bound without additional modeling assumptions. Extensive simulations exhibit that these estimators perform well even when the data are heavily censored. Applying these estimators to a kin-cohort study of Huntington's Disease (HD), our analyses underscores the elevated risk of death in HD carriers compared to non-carriers, and that Huntington gene mutation equally affects survival rates in both genders.