An Investigation of Minimization Criteria

• Activity Number: 407
• Type: Contributed
• Date/Time: Thursday, August 12, 2004 : 8:30 AM to 10:20 AM
• Sponsor: General Methodology
• Author(s): Angie Wade*+
• Companies: University College London
• Keywords: minimization ; patient allocation ; bias ; randomized control trials ; confounder

Abstract

The power of a RCT is maximized if patients are divided equally between treatments with respect to confounding factors. Minimization is a dynamic allocation procedure that has been used increasingly in recent years. Allocation of the next patient is biased, according to his or her characteristics, in favor of the treatment arm that maximizes similarity between the resultant treatment groups with respect to selected potential confounders (known as minimization criteria). Despite becoming more widely used, details of precisely how minimzation has been applied are rarely given and many questions arise for a researcher wishing to use it. For instance: How many and which confounders should be balanced? Should these confounders be prioritized in some way? How biased should the randomization be? This presentation investigates, via simulation of datasets, the relationships between the number, type and weighting of factors, degree of randomization bias, total sample size and the expected size of discrepancies obtained between treatment groups. Results are presented together with practical implications for clinical usage of minimization as a means of patient allocation.