I'm not buying the Mendelian randomization concept. It seems like you are setting up a bias in how people process alcohol which seems more like something to control for rather than something to use to make cohorts.

I don't know why we can't get a randomized trial if its something people volunteer to do. The connection to an alcohol lobby in the first case is something that can be accounted for and weighted after the fact. A potentially flawed study is better than no study if it's all we've got of that form and we are aware of the flaws. Information can always be disregarded later so the introduction of new information never hurts.

The other interesting question is as they mention light drinking is associated with higher income and they worry that becomes an uncontrolled factor. But have they considered that drinking habits can impact income? Drinking impacts lifestyle as much as lifestyle impacts drinking.

Not only that but these are things that can be controlled for. Pretty easily. Similarly they mentioned that you could have former heavy drinkers among the abstainers. That is also something that you can control for. Instead of gathering a minimum of information from your participants and then investing a lot into tracking them and analysis and trying to band-aid data you don't have with other studies which will result in more analysis debating if they way you did that is correct. You could just do the cheap thing of asking more questions.

This is why I'm excited for the use of AI modeling. We can treat everyone as being in a cohort of one, or a member of many overlapping cohorts, and then find the partial derivative of any one factor on the set or the partial derivative of a elastic factor given other set factors.

Then you can capture things like their ethnicity, specific genes, what time of day they drink, do they drink a small amount every day or much more but on the weekends, are they an always abstainer or current abstainer, did they ever have a habit of heavy drinking, other co-morbidities.

For ethnicity this is particularly interesting to this subject because they have found that the j-curve only can be observed with Europeans, possibly because we can process alcohol better. Or possibly because we have different socio-economic tendenies associated with alcohol. For non-europeans it may be possible that the negative effects hit immediately, whereas for Europeans we may have a threshold amount where the negative effects are mitigated and the positive effects out-weigh.

This is why I didn't like when they showed old studies that showed a j-curve, then they contrasted that with other methodologies that also used global data and said see, the prior methods were wrong because we got different results. In one study Europeans could have been 80-90% of the population, and in the other Europeans would be as low as 8% of the population. And you are saying that the difference is because of better methods? Might be. But there is kind of another big factor in there.

But either way we will never know until we get the two things we need. A complete observation that includes everything. And two an actual randomized trial. So much science endlessly debates over junk data when we could stop pussy-footing around and just get the data we need.

Best part of a real observation is the same data is useful for other mortality associations. We can do it all at once for everything. That's the thing I wish I could get the ear of RFKJ on. He wants health to be more lifestyle focused. He's going to need the data to support it and help people decide what matters and how much. Let's build this model. Because it would be a long term longitudinal model taking up to 30 years, we can't delay. It's a negligent failure of science if someone isn't already doing that having started some 30 years ago.