How is that possible though? You cant really tweak clinical results because it is auditable by authorities, and at most what is done to favor one drug over another is cherry picking during recruitment of patients (carrying the risk of narrowing your indications in the end) or excluding patients from the study (again that has to be fixed in the study design, you cant tweak it after the fact). The kind of hacking you describe would be just plainly not allowed and easy to discover upon any investigation.
There is quite a bit of leeway when you have real world data. You have to clean it. You can always find some justification. This is not neatly black/white at all. Similar in all other fields from accounting to law. I think it's a serious issue that we managed to make so many people believe way too strongly in absolutes ("science" as a magic word to end all discussions, because hey, "science" and "data" - now we have an objective reality!), when in reality the world is way more messy.
The original parent comment is describing someone hand-picking samples to obtain a predetermined conclusion. I know touching data to clean it before analysis can be a grey area, but the original parent comment -- if it is accurate -- is describing clear-cut fraud.
I refer back to what I wrote. You miss the point. There is no black/white in what is necessary and unnecessary or even nefarious "data cleaning". You just use the same fuzzy undefinable words again, but you did not make it any more a "hard fact" than any of it was before.