This seems like one of those areas where "double-blind randomized trial" sounds much cleaner on paper than it actually is. If nearly everyone can tell whether they got the drug, expectancy becomes a pretty nasty confound