Meta-analysis: Its Role in Medical Research and in Assessment of the Association Between Low Levels of Cholesterol and Excess Mortality

Abstract

The purposes of this editorial are to consider the role of meta-analysis in medical research from a nontechnical per­ spective and to discuss the recent application of meta-analy­ sis for evaluating the possibility of an association between low total cholesterol concentration and increased mortality. What Is Meta-analysis?-Meta-analysis can be viewed as a type of review of the literature. In a meta-analysis, however, additional steps are taken to enhance scientific validity: criteria for deciding which studies will be included are defined in advance so that only studies of sufficient merit will be considered, extensive efforts are undertaken to iden­ tify all pertinent studies (even those that are unpublished), the strengths and weaknesses of each study are carefully evaluated, and the raw data may even be solicited to enable the meta-investigators to perform their own analyses rather than relying solely on summary statistics reported by the original investigators. Sometimes the term "meta-analysis" is used more nar­ rowly to refer to the statistical manipulations whereby the estimates of size of effect and corresponding P values are combined to obtain overall estimates and corresponding overall P values. Unfortunately, such usage draws attention to computational techniques and summary statistics. In medical research, the primary focus should always be on thoughtful interpretations of data analysis and not on the computations themselves. An obvious advantage of a meta-analysis is that it pro­ vides both a more rigorous review of the literature and the additional information from the combined analysis. It also is a method of increasing sample size and thereby enhancing statistical power and the possibilities for subgroup analyses (to identify subgroups that may respond especially well to therapy, for example). Unfortunately, these benefits are often overstated. For instance, combining the results of multiple clinical trials by using meta-analysis is sometimes viewed as equivalent to having conducted a single, large multicenter study in which all centers have followed the same protocol. Obviously, all studies in a meta-analysis will not have used a uniform protocol. For the reasons discussed