COVARIATE-ADJUSTED RESPONSE-ADAPTIVE DESIGNS FOR BINARY RESPONSE

Abstract

An adaptive allocation design for phase III clinical trials that incorporates covariates is described. The allocation scheme maps the covariate-adjusted odds ratio from a logistic regression model onto [0, 1]. Simulations assume that both staggered entry and time to response are random and follow a known probability distribution that can depend on the treatment assigned, the patient's response, a covariate, or a time trend. Confidence intervals on the covariate-adjusted odds ratio is slightly anticonservative for the adaptive design under the null hypothesis, but power is similar to equal allocation under various alternatives for n = 200. For similar power, the net savings in terms of expected number of treatment failures is modest, but enough to make this design attractive for certain studies where known covariates are expected to be important and stratification is not desired, and treatment failures have a high ethical cost.

Keywords:

• Clinical trials
• Ethics
• Logistic regression
• Odds ratio
• Power

ACKNOWLEDGMENTS

The research of William F. Rosenberger and Deepak K. Agarwal was supported by grant R29-51017-05 from the National Institute of Diabetes and Digestive and Kidney Diseases. The authors thank Roy N. Tamura for providing covariate data from the fluoxetine trial.