One of the major problems translating "science to advice" is the need to specify a risk/reward model for the person being advised.

If P(TheScience) = 0.01, but P(GeneralInjury|TheScience) = 0.00000000000...1, and P(GeneralBenefit|TheScience) = 0.5 .... then most people should do "what the science says".

Of course there's P(MyBenefit|TheScience) and P(MyInjury|TheScience) which makes rendering general advise often impossible/unethical.

There's a class of people (eg., neurodivergent, autoimmune problems, etc.) for whom P(MyInjury|...) is "anti-correlated" with P(GeneralInjury|...)

...which leads to more pronounced mistrust regarding medical interventions.