Systematic reviews of randomized controlled trials: the need for complete data

If the relative effectiveness of different treatments that might be used in clinical practice is to be evaluated reliably, it is very important that the evaluation is carried out in an appropriate manner. This is especially true where the differences between treatments are expected to be moderate, and so easily obscured by the play of chance or systematic bias. Although such differences are often of considerable clinical importance, they can be difficult to assess and require a large amount of randomized evidence. This evidence can be obtained through prospective randomized controlled trials, meta-analysis of results from past randomized trials, or ideally a combination of the two, with prospective trials contributing to future meta-analyses. Whichever technique is adopted, all possible biases must be minimized through the collection of as much randomized evidence as possible. In meta-analyses, this is best achieved by ensuring that all relevant trials, and all randomized participants in these trials, are included in the analysis. The gold standard for this might be a meta-analysis of individual patient data, in which details for each participant in every trial are collected and analysed centrally. This approach requires considerable time and effort. However, it will add to the analyses that can be performed and will remove many of the problems associated with a reliance on published data alone and some of the problems that can arise from the use of aggregate data. This paper sets out some of the reasons for this and some of the techniques used for individual patient data-based meta-analysis.