When experimenting with new treatments, a control group is very important. Otherwise the study could be biased in favor or against the treatment. Even when a control group is present, a poorly-designed experiment could later turn out to be problematic, such as the Diethylstilbestrol (DES) medical tragedy (https://en.wikipedia.org/wiki/Diethylstilbestrol). In that case, studies that were in favor of the treatment were poorly-designed and used historical controls instead of contemporaneous ones.

It's about the fraction of that medical knowledge is likely to be revised when considering new controlled studies.

Ad hoc trials? Aka testing in production.