Bias and Variance Improvements in Nonparametric Estimation of Time-Dependent Accuracy Measures

• Activity Number: 252
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Chin-Tsang Chiang*+
• Companies: National Taiwan University
• Keywords: bandwidth selection ; conditional survivor ; marker-dependent censoring ; receiver operating characteristic curve ; kernel function ; U-statistic

Abstract

Under the well-established time-dependent cumulative cases and dynamic controls, the retrospective and prospective accuracy measures have been commonly employed to assess the overall discrimination capacity of a continuous baseline marker during the past decade. With censored survival data for the presence of marker-dependent censoring, our major research achievement is to offer practitioners a better nonparametric estimation approach for these time-dependent measures. Specifically, our bandwidth selection procedures are practically feasible and are theoretically valid. Moreover, we provide a thorough understanding for general large sample properties of the proposed estimators. To accommodate stratified survival data and survival data with multiple markers, appropriate modifications are made on the estimation of single-maker accuracy measures. It is evidenced through a class of simulation studies that the introduced estimators are far superior to the existing nonparametric ones in the literature and the developed inference procedures perform quite satisfactorily. Two empirical examples are further illustrated to demonstrate the applicability of our methodology.