|
Abstract:
|
The FDA's Guidance for Industry: E9 Statistical Principles for Clinical Trials presents unrestricted, restricted, and dynamic randomization methods as acceptable. However, unrestricted randomization and stratification using only a few key variables has been shown to be highly inefficient and have a much higher risk of serious imbalance in important covariates than optimally stratified randomization (OSR), which incorporates all important covariates. OSR, aka matched randomization, can handle cluster randomized trials, multi-arm trials, and be used in trials with enrollment coming all at once or sequentially. OSR has been shown to dramatically improve covariate balance, yielding greater study power and more persuasive evidence. However, limited knowledge of good practices and implementation methods remains a barrier. This talk presents new research into choosing a balance metric, setting a reservoir size, setting a fixed or dynamic threshold for determining acceptable matches, prioritizing covariates, and handling missing data. We compare and contrast a new dynamic randomization variant of optimally stratified randomization and present example code for implementing the methods in R.
|
