Why we need a broad perspective on meta-analysis

Abstract

In the world of clinical trials and meta-analyses there is an important debate between the “lumpers” and the “splitters.” This relates to whether the overall findings of clinical trials and meta-analyses are the appropriate outcome to apply to individuals (lumping) or whether it is better to try to match the characteristics of particular patients to characteristics of subgroups within trials or meta-analyses (splitting). Although the splitters' view seems intuitively correct, there are usually substantial clinical and methodological advantages to lumping. The generalisability and usefulness of meta-analyses are increased considerably if the individual trials cover different patient populations, settings, and concomitant routine care. For example, when a meta-analysis showed that the use of human albumin increased mortality1 this result applied to all three groups of critically ill patients studied. For patients with hypovolaemia the difference was not conventionally significant (95% confidence interval for the odds ratio 0.99 to 3.15), but it would be wrong to interpret this result as meaning that clinicians should continue to give these patients albumin. Most significant results will disappear because of lack of power if trials in a meta-analysis are split up into a large enough number of subgroups. It is more relevant that the point estimates were similar …