Drug Information Journal - March 2009 - (Page 174) 174 BIOSTATISTICS Ross meta-analytic evidence. For instance, to compare treatments A and B when there are no direct comparison studies of A versus B, if there are a sufficient number of studies of A versus C and B versus C, then it is possible (and valid) to indirectly compare A and B. A case in point is a systematic review and meta-analysis of published RCTs of GPIIb/IIIa inhibitors in PCI patients (25). In this example, the odds of 30-day death or nonfatal MI (D/MI) for abciximab and tirofiban were compared even though no direct comparison studies were available. There were, however, sufficient numbers of placebo-controlled RCTs of each agent alone (6 and 3, respectively), and these were used to develop an indirect comparison of the two active agents. The results of the meta-analysis of the six abciximab studies [OR = 0.52 (0.43, 0.63)] appeared superior to the results of the three tirofiban studies [OR = 0.73 (0.55, 0.96)]. The quantitative results of this indirect comparison, namely that abciximab had roughly a 40% advantage relative to tirofiban for 30-day D/MI, were subsequently validated in an RCT of the two agents, the TARGET trial (26). In this headto-head comparison of tirofiban versus abciximab in PCI patients, the OR for 30-day D/MI was 1.3 (P = .04), in favor of abciximab. Q. Can I compare treatments A, B, C, D, and so on, to each other, all at once? A. Yes, with caution. Network meta-analysis (mixed treatment comparisons or umbrella reviews, per the Cochrane Collaboration) is a more complicated application of indirect comparison methods of metaanalysis (27–30). It also has the potential to be a hugely important advance in the field, whereby statistical techniques for combining large amounts of data allow analysts to generate comparisons of efficacy (or safety) of many drugs in specific patient care settings. An illustrative example can be found in a network meta-analysis of efficacy of first-line antihypertensive drugs (31). A major advantage of network meta-analysis is that by including more patients and more data in the analysis, more precise comparisons are the result. And most important, as noted above, a comprehensive ranking of treatments is possible. A disadvantage of network meta-analysis derives from the need for a careful examination of the underlying assumptions, particularly with regard to pooling data from sufficiently similar studies of similar patient populations. Furthermore, as a relatively new method, there is still a need for user-friendly tools to assess inherent characteristics of such analyses. Incoherence is an example of one of these characteristics of network meta-analysis. Incoherence is a measure that quantifies the variation among estimates, similar to heterogeneity in standard meta-analyses, and it is incorporated into the calculation of the confidence interval (precision) of the overall estimate. There is a need for better—simpler— ways to describe incoherence and to understand its impact upon overall results. In conclusion, meta-analysis techniques are evolving in exciting directions that hold great promise for all who wish to use information better in health care. We are already seeing an increase in the applications of IPD meta-analyses, CMA, indirect comparisons, and network metaanalyses in medicine. The information needs of decision makers will continue to drive the science and applications of meta-analysis in the future. For instance, better ways of relating medication adherence to efficacy and safety, using single risk-benefit metrics, adjusting prospective planned meta-analyses for adaptive trial designs, and simultaneous adjustment for several levels of variability across study sets (ie, hierarchical modeling) are just some of the exciting directions the evolutionary path is pointing. Nevertheless, it must not be forgotten that meta-analysis is only as good as the best available evidence. While it is a better way to explore and understand existing evidence, such as it is, meta-analysis can never be a panacea for invalid or insufficient studies. And therein lies a challenge for all who generate clinical evidence in health care. REFERENCES 1. Cook DJ, Mulrow CD, Haynes RB. Systematic reviews: synthesis of best evidence for clinical decisions. Ann Intern Med. 1997;126:376–380.
For optimal viewing of this digital publication, please enable JavaScript and then refresh the page. If you would like to try to load the digital publication without using Flash Player detection, please click here.