Abstract:
|
Subgroup analyses in randomized controlled trials (RCTs) are troublesome. Informal assessments of consistency or inconsistency are frequently misleading. Formal assessments are usually too underpowered to be useful, and the rare "statistically significant" difference is subject to over- (or mis-) interpretation. To an extent, these problems are endemic to these analyses, but common approaches to presenting subgroup results, particularly graphically, may exacerbate the problem.
In this talk, we will look at common methods of presenting results of subgroup analyses and discuss ways in which they may mislead. We will look at two types of alternative presentations: those that might be used for presentation of final results, and those that might be considered "diagnostic tools" for statisticians to better visualize limitations of these analyses and gain a more intuitive understanding of the potential for exaggeration of the size of differences or overconfidence in the robustness of the results.
|