Bayesian Adaptive Randomization for Point-of-Care Clinical Trials

• Activity Number: 647
• Type: Topic Contributed
• Date/Time: Thursday, August 4, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Section on Government Statistics
• Author(s): Gheorghe Doros*+ and Robert Lew and Sarah Leatherman and Philip Lavori and Louis Fiore
• Companies: Boston University and VA Boston Cooperative Studies Program Coordinating Center and Veterans Administration and Stanford University and Veterans Administration
• Keywords: clinical trials ; adaptive design ; bayesian

Abstract

The point of care (POC) occurs when the patient is assigned a treatment. If electronic records capture all baseline and follow-up clinical data, then, in principle, randomizing treatment at the point-of-care imbeds a low-cost clinical trial into routine medical care. The POC pilot trial at the Veterans Administration compares two low risk insulin administration protocols, A and B. A simple form of an adaptive design begins with equal 1:1 allocation and periodically updates treatment assignment, using odds of p:1-p, where p is the current posterior probability that A is superior to B. A two-sided stopping rule periodically tests the null hypothesis, rejecting it if p = 1 - e or if p = e, where e is a threshold such as 0.01. We can add more parameters: time-dependent thresholds, irregular update and testing periods, stopping for futility, maximum sample size, and a monotone function of p in place of p. Our R-language software simulates Type I error and power. We offer a strategy to modify the design to close in on target values such as 5% and 90%. We use graphics to compare designs and to assess how each parameter affects performance.