Date Published: January 3, 2008
Publisher: Public Library of Science
Author(s): Gordon H Guyatt, Edward J Mills, Diana Elbourne
Abstract: Background to the DebateSystematic reviews that combine high-quality evidence from several trials are now widely considered to be at the top of the hierarchy of clinical evidence. Given the primacy of systematic reviews—and the fact that individual clinical trials rarely provide definitive answers to a clinical research question—some commentators question whether the sample size calculation for an individual trial still matters. Others point out that small trials can still be potentially misleading.
Partial Text: Funding agencies, ethics review boards, journals, and investigators are often preoccupied with power calculations and sample sizes required in clinical trials. We argue that the current practice of sample size justification for randomized clinical trials (RCTs) represents a willing self-deception. Recognizing and adjusting to current realities of RCT conduct may be necessary.
When I first started working with clinicians on randomised trials a quarter of a century ago, the most important point I felt I needed to stress was the centrality of randomisation (and allocation concealment) for reducing the very real risks of selection bias at trial entry. Only once that idea was firmly embedded would I consider the many other aspects of trials to which I, as a methodologist, felt I might have something to contribute—such as post-randomisation selection biases, assessment biases, and, of course, sample size. While I would still consider randomisation and allocation concealment the most important issue, sample size doesn’t come far behind.
Professor Elbourne makes excellent points. Minimizing bias in clinical trials should be paramount. Systematic reviews should inform clinical trials. Investigators should conduct the largest multicenter trials that feasibility permits.
Readers who are looking for a clash between two diametrically opposed views may be disappointed in this exchange—we agree on so much.