Clinical Trials

❘❙❚■ Chapter 32 | Overview of ReportingProblem areas in trial reporting: examplesRandomizationThe idea of correct randomization to ensure an unbiased comparison is very old.However, its importance was not acknowledged in early trial reports. Accordingto a review in 1990, a third of published trials provided no clear evidence that groupswere randomized [2]. Indeed, there were indications of post-hoc assemblies ofgroups, with some trials using simple randomization schemes, often with toosimilar sample sizes in both groups than would be expected by chance.Furthermore, in about 40% of published trials from this time the baselinecomparisons were inadequately handled [2].Intention-to-treat analysisA survey in 1999 showed that only 50% of trial reports mentioned ‘intention-totreat’analysis [3]. This can have serious implications on the interpretation of theresults. For example, in a trial comparing medical and surgical treatment for stableangina pectoris, some patients allocated to surgical intervention died before beingoperated on [4]. If these deaths are not attributed to surgical intervention using anintention-to-treat analysis then surgery will appear to have falsely low mortality.Publication biasUntil recently, medical literature was dominated by a bias towards reporting verysignificant positive effects or very large point estimates, but avoiding trials withneutral or negative results. This led to distorted views – an effect known aspublication bias [5]. The scientific community is changing and now encouragesthe publication of all trials. This is of particular relevance in meta-analyses,which seek to support external validity against other studies.Effect sizesCare should be taken in examining effect sizes. Effect sizes are sometimes reportedin such a way that the treatment effect seems massive. For example, a trial ofcholestyramine (a lipid-lowering drug) in men with high cholesterol reported a17% relative reduction in risk for both fatal and nonfatal coronary events, whereasthe absolute risk difference was only 1.7%. Because this study incorrectly usedonly one-sided tests, there remains uncertainty about the real effect ofcholestyramine [6].Changes to regulatory definitions/guidelinesAs mentioned earlier, regulatory definitions or guidelines might change duringthe course of a trial. For example, the RITA (Randomized Intervention Trial ofAngina) 3 trial used a different definition of myocardial infarction from thatpublished some years later by the European and US cardiology societies [7].372