[avernon]

The data is observational, which generally means you should ignore it. It is too noisy.

There were those studies that showed moderate alcohol use improved health and only heavy drinkers saw detrimental health effects. The problem was that "no drinking" group included people that weren't drinking because of poor health. Later studies compared drinking vs. a "no drinking" sample of people that drank around two glasses of wine per year. The improved health effects completely disappeared. The more you drink, the worse it is for your health.

So this study is like that in using a potentially unhealthy comparison group. They try to offset that a little by also throwing in people that quit drinking. But it is likely that some people quit drinking because of health problems. So I'd guess that this study has the same problem with an unhealthy comparison group. The study probably can't tell you what the actual relationship between alcohol use and dementia with any authority.

[KarlKemp]

Data like this is difficult, yes. Ignoring it outright may earn you accolades online where cynicism is often mistaken for intelligence. But it’s just scientific defeatism.

There is no way to study this issue, and many similar in nutrition, but with observational data.

[float4]

> The problem was that "no drinking" group included people that weren't drinking because of poor health

They've corrected for this. From the abstract:

> Adjusting for additional demographic and clinical covariates, and accounting for competing risk of death, did not substantially affect results.