Lately, most of the argumentation said, "Look, Mendelian randomization studies, which are objective, which have less biases, they found that there is no beneficial effect." That's how the benef- how the Mendelian randomization came in.

Yeah, I think on the Mendelian randomization part, there has been a lot of debate around the sick quitter effect that all of the potential protective effect is due to people being actually sick, so reducing or stopping their consumption, which renders low to moderate drinkers seeming to be at lower risk than those who completely abstain from alcohol.

And so Mendelian randomization studies have been changing the field quite a lot and the argumentation quite a lot, as Jürgen has mentioned, and basically moved a lot of opinion and kind of science basis towards the reasoning that there's no protection of alcohol for ischemic heart diseases and related outcomes.

The problem with Mendelian randomization studies, however, there are a lot of very strong assumptions that go into this analysis, and with alcohol, which is a time-varying risk factor, so you certainly change your consumption over time, you initiate not at birth, but you initiate consumption at, say, 18, 21, and these create a lot of Subtle problems when trying to estimate the effect of alcohol consumption on outcomes like ischemic heart disease, and we concluded that there's either no or somewhat harmful effect given that, and also the underlying problems of Mendelian randomization when studying risk factors like alcohol consumption or other time-varying risk factors, that the evidence from cohort studies that we have so far, even if certainly imperfect, we propose to proceed with those for comparative risk assessments to assess how much, uh, of the burden is due to consumption.

Uh, is there anything else that you would like to add, Sinclair?

I think good next steps is to refine methods that we have, especially in the Mendelian randomization realm, to re-tackle the questions that have been answered before, but now in a even more rigorous way, and as Jurgen has mentioned, to triangulate findings between different sources of evidence, and try to harmonize the causal questions we can answer between them, and make them as explicit as possible.

Reuse observational data, I think there's, again, a lot of potential to, to employ the target trial emulation framework here, and it certainly doesn't, um, bypass issues like unmeasured confounding or measurement error, but there are certain- certainly many biases that it can avoid, and where it can help to generate evidence that is also more actionable.

For, for drinking, for example, we want to know what happens if we change our drinking in a certain way, if we keep kind of doing that change, and not what the status at a, at a given point in time is.