Bayesian Methods and Ethics in a Clinical Trial DesignBMJ 1997; 314 doi: https://doi.org/10.1136/bmj.314.7088.1209a (Published 19 April 1997) Cite this as: BMJ 1997;314:1209
- Robert A Crouch, doctoral candidate
Ed Joseph B Kadane / John Wiley, £65, pp 318 ISBN 0 471 84680 5
When confronted with the ubiquitous P value, the statistical novice usually concludes (wrongly) that it represents the answer to the clinically important question “What is the probability that treatment A is superior to treatment B?” Yet classical (frequentist) statistical methods–the standard used in clinical research–do not address this question; Bayesian methods do. Bayesian researchers can formally quantify prior beliefs about relative efficacies of treatments (from previously published literature and from “informal” clinical experience), and they can then update these “priors” in light of new information emerging from the clinical trial. The resulting posterior distribution yields clinically useful information that is revised as each patient's outcome is known and gives researchers, as Kadane notes, a more ethical clinical trial, one which minimises the probability that patients will be given inferior treatments.
The book–which includes contributions by statisticians, physicians, and lawyers from the United States–focuses on a clinical trial designed using Bayesian methods that was carried out by some of the authors at the Johns Hopkins Medical Institution. The trial, comparing verapamil and nitroprusside in the control of postoperative hypertension, represents an attempt by the authors to implement their novel ideas in the clinical context, and, to their credit, they do a wonderful job in describing the mechanics of the trial from inception to completion. This is the book's strength: the detailed discussion of how admissibility of treatment is determined by eliciting Bayesian priors from expert clinicians; the proffering of a Bayesian interpretation of the trial results; and the frank disclosure of the mistakes that were made during the trial while also offering helpful correctives to these problems.
However, the problem with the book as a whole is that it is about 10 years out of date. This is no small complaint: the decade since most of this book was written has witnessed advances in the study of research ethics and, with the development of new statistical sampling methods in the mid-1980s, a dramatic increase in the literature dealing with Bayesian methods in the context of clinical trials. Those familiar with the literature on research ethics are thus bound to conclude that the book's treatment of ethics (and law) is lightweight, and those familiar with the literature on biostatistics will be interested in how the statisticians think more recent work bears on the ideas they present in this volume. Thus, while the authors are interested in their own particular Bayesian method, it would have been a valuable addition if they had updated their own views in light of recent evidence, both ethical and statistical.