Impact of Ignoring Correlations in Vision Clinical Trials

• Activity Number: 547
• Type: Contributed
• Date/Time: Wednesday, August 7, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Ratna Revankar*+ and Gary Koch and Atsushi Kawaguchi
• Companies: Johnson & Johnson and The University of North Carolina-Chapel Hill and The University of North Carolina
• Keywords: vision clinical trials

Abstract

In Vision Clinical trials, the unit of drug treatment is one of the two eyes. To make the maximum use of the experimental units eye is considered as unit of analysis. Only the case of two treatments, active and vehicle placebo is considered here for simplicity. Clinical parameters of interest as an example would be itching and redness measured on the ordinal scale. For the nature of the data (generally, non-normal and skew distributed), a rank test such as two-sample Wilcoxon Rank Sum Test comes out as the test of choice. However the factor of non-independence of the sampling units (eyes) is generally ignored. This paper investigates the impact of ignoring the non-independence of the sampling units (eyes) on type I and II error rate. In addition, the performance of other methods such as GEE and Mixed effects models which incorporate correlation structure are investigated.