Adjusted Cumulative Distribution Function and Effect Size Methods for Stratified Two-Sample Comparisons

• Activity Number: 256
• Type: Contributed
• Date/Time: Monday, August 2, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Kelly H. Zou and Martin O. Carlsson*+ and Ching-Ray Yu
• Companies: Pfizer Inc. and Pfizer Inc. and Pfizer Inc.
• Keywords: Effect Size ; Receiver Operating Characteristic analysis ; Odds ; Wilcoxon's Rank-Sum Test ; Van Elteren's Test ; Youden's Index

Abstract

Nonparametric and parametric methods for either continuous or ordinal-rating data are developed using receiver operating characteristic (ROC) analysis. Cohen's or Glass's effect size (ES) is examined. The area under the curve (AUC) is computed. The odds are derived via a logit transformation of the AUC. Various weighting schemes to adjust for the effect of strata are compared, based on van Elteren's test to generalize Wilcoxon's rank-sum test and inverse-variance weighting in ROC analysis. Several parametric mixture-distributions between two populations are considered. Stratum-specific, adjusted and pooled ES are derived along with the optimal thresholds via Youden's index. Post-hoc analysis of clinical trial data is illustrated.