Nonparametric Maximum Likelihood Estimators of Time-Dependent Accuracy Measures for Survival Outcome Under Two-Stage Sampling Designs

• Activity Number: 545
• Type: Topic Contributed
• Date/Time: Wednesday, August 6, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Dandan Liu*+ and Tianxi Cai and Lok Anna and Yingye Zheng
• Companies: Vanderbilt University and Harvard and University of Michigan and Fred Hutchinson Cancer Research Center
• Keywords: biomarker evaluation ; survival outcome ; two-stage sampling ; time-dependent accuracy ; case-cohort ; nested-case-control

Abstract

Two-phase study design becomes increasing popular for large prospective cohorts studies of rare chronic diseases such as cancer due to consideration of cost-efficiency. To account for the selection probabilities reflecting the sampling scheme of a specific design, the inverse probability-weighted (IPW) estimator are usually adopted. Although it is simple to implement, the most efficient approach is based on nonparametric maximum likelihood estimation (NPMLE). When time-to-event outcome is of interest, existing statistical methods focus on making efficient inference on relative hazard parameters from the Cox regression model. Recently, time-dependent accuracy measures for survival outcome evaluated at a pre-specified time given a threshold of the biomarker were developed. The corresponding IPW estimators were also proposed under the two-phase study design. In this paper, we study NPMLEs of time-dependent accuracy measures under the two-phase study design. The asymptotic results were derived and the proposed estimators were used in evaluation of a novel serum biomarker in predicting the diagnosis of hepatocellular carcinoma (HCC) under nested-case-control design.