Introduction

Clinical measurements on humans, whether healthy or ill, rarely give exactly the same results from one occasion to the next. It is often quite difficult to appreciate the magnitude of this variability and therefore assess its influence on the conclusions that one may wish to draw from any investigation. Thus, although penicillin was one of the few miracle cures in which little dispute of its benefit remained after use in only a few cases, variable levels of benefit achieved when patients with rheumatism are given gold therapy are much more common in clinical practice and therapeutic research. With such variability present, it follows that differences are almost bound to occur in any comparison between patient groups. These differences may be due either to real effects, for example, the two groups have received treatments that really do differ in efficacy, or to random variation, for example, the different responses in diagnostically similar patients to gold therapy as indicated earlier, or these differences could arise from a combination of both. Such variation is just as likely to be present in patients who require some form of rehabilitation therapy.

The presence of variation means that choosing the correct design for any study is crucial. A well-designed study which is then carefully carried out but poorly analysed, can be rescued by a more appropriate re-analysis. A flawed design, however, may not provide the answer to the question posed, no matter how detailed the analysis. Consequently, such a study is wasteful of resources. It is a sad fact that many clinical studies fail in their objectives, either through poor design or the lack of (patient) numbers.