Methods for Assessing the Credibility of Clinical Trial Outcomes
- PDF / 669,908 Bytes
- 10 Pages / 504 x 720 pts Page_size
- 49 Downloads / 208 Views
0092-8615ROO1 Copyright 0 2001 Drug Infomation Association Inc.
METHODS FOR ASSESSING THE CREDIBILITY OF CLINICAL TRIAL OUTCOMES ROBERTA. J. MATTHEWS Visiting Research Fellow, Department of Information Engineering, Aston University, Birmingham, England
Credibility-the believability of new findings in the light of current knowledge-is a key issue in the assessment of clinical trial outcomes. Yet, despite the growth of evidencebased medicine, credibility is usually dealt with in a broad-brush and qualitative fashion. This paper describes how Bayesian methods lead to quantitative credibility assessments that take explicit account of prior insights and experience. A simple technique based on the concept of the critical prior interval (CPI) is presented, which allows rapid credibility assessment of trial outcomes reported in the standard format of odds ratios and 95% confidence intervals. The critical prior interval is easily determined via a graph, and provides clinicians with an explicit and objective baseline on which to base their assessment of credibility. The use of the critical prior interval is demonstrated through several working examples. Key Words: Credibility; Critical prior interval; Bayesian methods
INTRODUCTION THE OUTCOME OF MOST clinical research is now stated in quantitative terms, a trend encouraged by the emergence of evidence-based medicine. vpically a single number, such as an odds ratio (OR), is used to capture the essence of the research finding, for example, the OR for mortality after five years. This number is conventionally accompanied by a measure of the probability of the result emerging through the play of chance, such as a 95% confidence interval (CI). If this measure meets a preestablished criterion, for example, the 95% CI excludes an OR of
Based on a presentation from the DIA Workshop “Statistical Methodology in Clinical R&D,” April 2-4, 2001, Vienna, Austria. Reprint address: Robert Matthews, 47 Victoria Road, Oxford, OX2 7QF, UK. E-mail: r.matthews@ physics.org.
1.00, corresponding to no effect-the finding is held to be “statistically significant.” While specific statements of statistical significance are routinely included in reports of new findings, discussion of their credibility-that is, their believability in the light of existing knowledge-usually consists of broad-brush arguments based on the outcome of previous research. Such arguments are, however, all too easy to devise; Egger et al. (1) cite a case where the authors of two mutually contradictory studies were both able to supply apparently reasonable qualitative credibility arguments for their conflicting findings. The desire to include some quantitative measure of credibility has prompted the illegitimate use of statistical significance as a surrogate for such a measure. However, the warnings of statisticians that “significance” and “confidence” have specific technical meanings with no direct connection with
1469
Downloaded from dij.sagepub.com at Universitaet Bonn on April 6, 2015
Robert A. J. Matthews
1470
thei
Data Loading...
