Analysis of Non-Negative, Continuously Distributed Data with Excessive Zeros

• Activity Number: 669
• Type: Contributed
• Date/Time: Thursday, August 5, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Huiling Li*+ and Lynn Wei and Hui Quan and Nanshi Sha
• Companies: sanofi-aventis and sanofi-aventis and sanofi-aventis and Columbia University
• Keywords: Skewed distribution ; Two-part model ; Empirical distribution ; Type I error rate control ; Power

Abstract

Excessive zeros often occur in non-negative continuously distributed data in randomized clinical trials. Analysis of covariance (ANCOVA) might not provide appropriate statistical inference for the comparison of two treatment groups due to the large number of zeros and highly skewed distribution. We will explore three other options including RANK-ANCOVA, the two-part model, and the trimmed mean model. Empirical distribution of existing clinical trial data will be used to generate data in simulation to compare the performances of the four methods in terms of type-I error rate control and power. A data example will be used to illustrate the application of these analysis approaches.