I just read the Methods section of that article. It does not use a RCT design; it's an observational study based on questionnaire data. All standard confound caveats apply (with the exception of the controls accounted for in the paper).

Thanks for explaining. After I posted I read the methods and that became clear, but the explanation of confounding is definitely helpful.

Willett says in his book, "When the first reports appeared linking moderate alcohol consumption with lower rates of heart disease, many doctors and scientists thought that some other habit shared by drinkers, not the drinking accounted for the benefit. Today the evidence strongly points to alcohol itself."

Unfortunately I don't have the copy of the book with me to check, and that study was the only one I could find by searching the book through Amazon. I would be interested to see what other evidence he points to.