You do this with a proper understanding of statistical power, and you report "no conclusion" when you don't have studies of adequate power. This is not actually a terribly difficult problem with a proper understanding of Bayesian statistics, it just doesn't lead to good headlines, because a lot of the time the answer is "the study couldn't really provide an answer."
I'm not saying you're wrong but doesn't that inherently favour the side claiming "there is no problem because you can't prove there's a problem." Until by the time someone can prove there's a problem we already have a generation with lead poisoning or a collapsed ecosystem or whatever it may be.
