A critical appraisal of meta-analysis of randomized controlled trials - a cardiovascular perspective

Article type
Authors
Farkough ME, Flather MD, Yusuf S
Abstract
Introduction: Many meta-analyses have been published in the cardiovascular field over the past decade. There are now examples where the results of meta-analysis combining data from a number of small trials addressing a given clinical problem have been later confirmed (the early fibrinolytic meta-analysis and ISIS-2) and questioned (magnesium and ISIS-4) by large simple trials.

Objective: We sought to evaluate the strengths and limitations of meta-analysis as a tool for clinical decision-making by reviewing the important meta-analyses in cardiovascular medicine.

Discussion: The rationale for conducting meta-analyses include: (1) to answer questions that cannot be addressed by a single trial; (2) to provide a more reliable estimate of treatment effect particularly in important subgroups; (3) to resolve debate where trial results are in conflict; and (4) to prepare for large simple trials by generating hypotheses for a given clinical question. Recently, there have been examples of where the results of meta-analyses and a large simple trial have led to an evaluation of the limitations of meta-analyses. These limitations include: bias - both publication bias where only the most promising studies are published and the "meta-analyst" bias where the researcher decides what question and outcomes should be evaluated; heterogeneity between the smaller RCT's that are included in the overview; variations in the validity of the source evidence, and the possible unreliability of estimates of treatment effects because of relatively few endpoints. A critical appraisal of meta-analysis in cardiovascular disease leads to the following recommendations: (1) meta-analyses should be used to generate hypotheses for important clinical questions that could later be addressed by a large simple trial; (2) every effort should be made to conduct meta-analyses as a prospective effort to limit the degree of bias and (3) while the meta-analysis should not be over-interpreted., it could improve clinical practice by facilitating earlier recognition of the efficacy and safety parameters of investigational interventions.