A Weighted Harrell-Davis Distance Test with Applications to Censored Data

• Activity Number: 177
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Dongliang Wang*+ and Alan D. Hutson
• Companies: SUNY Upstate Medical University and University at Buffalo
• Keywords: Quantile function ; Censored data ; Distance test ; log rank test

Abstract

Consider the standard two-sample treatment-control model with time-to-event as the endpoint. Some commonly used test statistics for the two-group comparison include estimating the integrated weighted differences between either the hazard functions or survival functions. In this note we propose a novel test statistic derived from a quantile function point of view, which corresponds to an estimate of the average survival time difference. Utilization of a tuning parameter through the application of a smooth quantile function estimator shows an improvement of efficiency in terms of the MSE when compared to direct application of classic Kaplan-Meier survival function estimator used in other approaches. The large sample consistency of our estimator is proved for fixed bandwidth values theoretically and validated empirically. A Monte Carlo simulation study also shows the relative efficiency of our estimator given small sample sizes. The procedure is finally illustrated via an application to epithelial ovarian cancer data.