Shrinkage Nonparametric Estimation of Median Survival Time from Censored Data with Applications to Multicenter Studies

• Activity Number: 679
• Type: Topic Contributed
• Date/Time: Thursday, August 8, 2013 : 10:30 AM to 12:20 PM
• Sponsor: International Indian Statistical Association
• Author(s): Mohammad Rahbar*+ and Xuan Zhang and Sangchoon Jeon
• Companies: Univ of Texas Health Science Center and Univ. of Texas Health Science Center at Houston and Yale School of Nursing
• Keywords: Nonparametric ; survival time ; shrinkage estimation ; pretest estimation ; simulations ; PROMMTT

Abstract

We present unrestricted, pretest, and shrinkage estimation of the median survival time from several independent samples of censored data. Assuming the hypothesis of homogeneity is tenable the vector of median survival times is estimated from the combined sample, or by combining the estimates from each sample with some prior non-sample information. We present asymptotic properties of seven nonparametric procedures: unrestricted, combined (CE), shrinkage combined, pretest, shrinkage pretest, Stein-type shrinkage, and positive part shrinkage. Through simulations we compute asymptotic risks using squared error loss. We compare the efficiency of these procedures relative to CE using ratio of the asymptotic risks. We demonstrate application of these methods to estimating median time for a trauma patient to receive 6 units of red blood cells in the Prospective Observational Multicenter Major Trauma Transfusion (PROMMTT) study. Our results indicate that performance of these estimation procedures depends on the strength of homogeneity. When homogeneity holds the CE is the most efficient estimator. However, the CE becomes inconsistent when homogeneity fails.