Meta-analyses in Cancer Clinical Trials: Principles and Pitfalls

Excerpt

The term “meta-analysis” was first applied by Glass in 1976 to describe the combination and analysis of data from multiple research projects.1 Other terms such as “overview”2,3 and “combined analysis” were in usage prior to this and results from physical sciences have been pooled to generate meta-analyses or overviews since the 1930s.4 Indeed, the practice is commonplace and not controversial within this discipline where individual trials tend to be less heterogeneous than within the biological field. The utilization of this approach as a research tool in clinical medicine was introduced in the 1970s, initially to investigate therapeutic questions in cardiovascular disorders.5‐7 More recently, this methodology has been championed by Peto (among others) as an effective means of increasing the power of similar studies on cancer treatments where the same or a similar question is addressed by the randomization process.8 Meta-analyses have been rapidly accepted over the past 10 years, and continuing analysis of past studies for a particular type and stage of cancer have provided more definitive guidance on treatment efficacy and the direction of future research strategy. The formulation of public policy should be based on robust randomized clinical data and ideally come from large single randomized trials.9‐11 Though a meta-analysis is not a substitute for the latter, these “studies of studies”12 can yield affirmative conclusions, and when the null hypothesis is not refuted, resources can be channeled into other areas of research. Meta-analysis is an instrument of exploration rather than confirmation,13 and it leads to hypothesis generation rather than testing.14,15 Such analyses can provide powerful statistical evidence to either support or denounce a new treatment policy.16 …