Standard of “label invariance” should not be abandoned

EDITOR,—Stephen J Sharp and colleagues' discussion of the dangers of attempting to compare treatment benefit with underlying risk in meta-analyses of placebo controlled trials of a presumed active treatment is excellent in two respects.1 Firstly, the authors clearly show a point that I have made—that when the difference in mean observed outcome between groups is related to the mean outcome in the placebo group then a spurious correlation is induced.2 Secondly, they make the important point that it …