Comparing the methods used to compare antidepressants.

A wide array of antidepressants are now available to treat depression. However, the conventional wisdom that these medications are equally effective may not be true. It is argued that most randomized controlled trials (RCTs) do not have the power to discriminate between an effective and an even more effective medication. Until RCTs that enroll 300 or more patients per arm are conducted, or more sensitive research designs are developed, it will be difficult to determine if a "not statistically different" finding reflects true therapeutic equivalence or if it is a false-negative result (or type II error). Statistical methods that evaluate data from a number of studies are used increasingly to compare treatment strategies. The relative merits and limitations of quantitative meta-analysis and pooled analysis of original data are discussed. The former method is preferred when a large number of relevant RCTs can be analyzed. Beyond the number of studies, publication bias ("file drawer" effect) and study selection criteria may influence outcomes of meta-analysis. Pooled analysis, which combines all original patient data, is preferred when there are only a handful of related RCTs. However, the integrity of a pooled analysis can be ruined by selective inclusion of studies ("cherry picking"). The validity of combining the data from different studies also must be demonstrated. Results of studies using these complementary methods suggest that there may be clinically meaningful differences in efficacy between several classes of antidepressants.