It depends what you are looking at. "Do antibiotics work?" is a slightly different question than "Is prescribing antibiotics effective?".

Of course you'd want data on how people actually take them, because to make decisions you'd like to know if the result means "prescribing antibiotics just doesn't work" or "we need to figure out how to make sure people actually take the prescribed antibiotics effectively"

Exactly. A more apt analogy would be the hypothesis "do antibiotics reduce the prevalence of MRSA, if prescribed for everyone?"

We can know that antibiotics work, but still fail spectacularly when used improperly. It's important to test both.