Adaptive randomization for balancing over covariates. (13th June 2014)

• Record Type: Journal Article
• Authors: Hu, Feifang; Hu, Yanqing; Ma, Zhenjun; Rosenberger, William F.
• Is Part Of: Wiley interdisciplinary reviews. Volume 6:Number 4(2014)
• Page Start: 288
• Page End: 303
• Publication Date: 2014-06-13
• Subjects: Mathematical statistics -- Data processing -- Periodicals; Science -- Data processing -- Periodicals; Social sciences -- Data processing -- Periodicals; Mathematical statistics -- Periodicals; 519.50285
• DOI: 10.1002/wics.1309
• Languages: English
• ISSNs: 1939-5108

Abstract

<abstract abstract-type="main" id="wics1309-abs-0001"> <title> <x xml:space="preserve">Abstract</x> </title> <p id="wics1309-para-0001">In controlled clinical trials, balanced allocation over covariates is often viewed as an essential component in ensuring valid treatment comparisons. Minimization, sometimes called 'dynamic allocation', or 'covariate‐adaptive randomization' has an advantage over stratified randomization, in that it is able to achieve balance over a large number of covariates when the sample size is small to medium. Despite its effectiveness, minimization has been questioned by regulatory agencies, mainly because of its increased complexity in practice and its potential impact on subsequent analysis. In recent years, however, with developments in clinical trials information technology, as well as advances in statistical theory, the attitudes toward minimization have evolved. In its 2013 draft guidelines, the European Medicines Agency (EMA) provided instructive guidelines for the implementation of minimization. In this paper we review the broad class of methods that belong to minimization, including its original forms for balancing over covariate margins and its generalization to balancing over other subgroups of interest or over continuous covariates. Moreover, we review the theoretical development in recent years, including the large‐sample properties of balance under minimization, the impact of minimization on inference for different data types, and on suitable randomization tests. <italic>WIREs Comput Stat</italic> 2014, 6:288–303. doi: 10.1002/wics.1309</p> <p>For further resources related to this article, please visit the <ext-link ext-link-type="uri" xlink:href="http://wires.wiley.com/remdoi.cgi?doi=10.1002/wics.1309" xlink:type="simple" xmlns:xlink="http://www.w3.org/1999/xlink">WIREs website</ext-link>.</p> <p id="wics1309-para-0002">Conflict of interest: The authors have declared no conflicts of interest for this article.</p> </abstract>