For example, a good basis would be statistics: how many past studies have been reproduced and confirmed that they proposed the correct conclusion? If you take 1000 past studies that were later reproduced by an independent team and count the number of them that were disproved, if this number is <500, then you know that the probability that the study is crap is lower than the probability that it is not.

There are other basis too. For example, the fact that not everyone will like to have deaths on their conscience. Or the fact that only a fraction of situations are situations suitable to commit fraud ("normal" studies often involve several independent institutes) and that fraud is very risky, especially in a sector like medicine (for example, if someone comes with an alternative to beta blockers, they will want to try to see if it performs better, and will notice that beta blockers don't perform as expected).