Response-adaptive randomization in clinical trials: from myths to practical considerations

Abstract

Response-adaptive randomization (RAR) is part of a wider class of data-dependent sampling algorithms, for which clinical trials have commonly been used as a motivating application. In that context, patient allocation to treatments is defined using the accrued data on responses to alter randomization probabilities, in order to achieve different experimental goals. RAR has received abundant theoretical attention from the biostatistical literature since the 1930's and has been the subject of heated debates. Recently it has received renewed consideration from the applied community due to successful practical examples and its widespread use in machine learning. Many position papers on the subject present a one-sided view on its use, which is of limited value for the non-expert. This work aims to address this persistent gap by providing a critical, balanced and updated review of methodological and practical issues to consider when debating the use of RAR in clinical trials.

Subgroup analysis and adaptive experiments crave for debiasing

“…A recent article by Robertson et al (2020) provided a comprehensive review of RAR designs from both frequentist and Bayesian perspectives.…”

“…On the other hand, while adaptive experimental designs-that allow randomization probabilities to be adapted during the trial based on sequentially accrued data-have received much attention in the past decade (Hu & Rosenberger, 2006;Robertson et al, 2020;Rosenberger & Lachin, 2015;Thall & Wathen, 2007;Wathen and Thall, 2017), even without subgroup analysis, the statistical validity of classical treatment effect evaluation approaches can be challenged given that the collected data are no longer independently distributed. The first part of this manuscript reviews some related literature on experiment design strategies and potential bias issues of some commonly adopted estimators from a frequentist viewpoint.…”

Minimization in randomized clinical trials

“…Prognostic balance may be sought to avoid accidental bias, but does it also increase the power of the test for a treatment effect? In nonlinear models or generalized linear models commonly used in clinical trials (eg, logistic regression and survival models), balance does not minimize the variance of the estimate of the treatment effect (see section 3.2 in Robertson et al 33 ). In such cases, optimal treatment-allocation procedures have been proposed to maximize statistical efficiency.…”

The Acute Stroke Therapy by Inhibition of Neutrophils study – Key features and impact

“…25 By comparison, real time response-based dose adaptation as deployed in ASTIN was rare at the turn of the century and continues to be rare today. For a full discussion of the pros and cons of response-adaptive randomization the reader is referred to the paper by Robertson et al 26 In our experience, when response-based randomization is used in dose-response studies today, most of the cases pertain to changing dose allocation at fixed points of a trial and the adaptations often involve adding or removing a dose. This makes trial execution more straightforward and lessens the risk of errors.…”