Bootstrap Confidence Intervals for Probability of Better Outcome

• Activity Number: 192
• Type: Contributed
• Date/Time: Monday, August 8, 2005 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Haitao Gao*+
• Companies: Eli Lilly and Company
• Keywords: Bootstrap confidence intervals ; area under ROC curve ; simulations ; probability of better outcome

Abstract

In a randomized placebo-controlled clinical trial setting with continuous efficacy measures, the treatment effect often is assessed by comparing mean change from baseline to endpoint between treatment groups. Clinicians often are more interested in the proportion of patients responding than the mean score for a group of patients. Faraone et al. (2000) proposed a drug-placebo response curve approach to address clinical significance. The approach is a generalization of the receiver operating characteristic (ROC) curve. The area under the response curve estimates the probability that a randomly selected patient given the drug will respond better than a randomly selected patient given placebo. This may have a special appeal to clinicians. Under normality assumptions, the probability can be converted to the effect size (standardized treatment difference) and vice versa. This paper examines bootstrap methods for constructing confidence intervals for the probability of better outcome. Simulations were conducted to examine both the power and Type-I error of the approach under different distributional assumptions. This method was applied to ADHD clinical trial data for illustration.