More Balanced Treatment Allocation When Randomization by Center

• Activity Number: 372
• Type: Contributed
• Date/Time: Tuesday, August 11, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Ruji Yao* and Norman Ying Yao
• Companies: and Miller Institute for Basic Research in Science
• Keywords: combinatorics ; randomization by center ; Latin-Squares

Abstract

For a multi-center clinical trial with randomization by center, the final treatment allocation depends on two random factors: 1) the random code generated prior to and used during the trial and 2) variations in the number of enrolled subjects and their order of randomization. For a design based upon equal assignment over all treatment arms, imbalances are accumulated over all centers. To reduce the noise induced by the first factor, we have introduced a Latin-squares method in [1] to generate a random code for each center. To approach the second issue, we have defined an "optimum" situation in [1], which assumes that the predicted enrollment numbers for each center at the start of a trial are the same as the actual enrolment. Then, we can combine centers with similar predicted size into a Latin-square that achieves treatment balance independent of the order of the randomly generated code. Although the optimum situation rarely arises in practice, it suggests that any information about the final enrollment may help in the overall Latin-squares balancing. Here, we introduce a simple modified Latin-squares method, which assumes that the IVRS system can determine when the final block is being assigned to a given center. We use real world data to demonstrate our approach and to quantify the improvement.