[neuroma]

They did adjust for confounders available in the Framington heart study.

Observational studies are less good at showing causation, sure.

Whilst I can't see the original article, the news summary insinuated a dose related effect. A sample size of >2000 is also decent.

Naturally in an ideal world you'd want a 3 arm, 20 year long, randomised placebo controlled double blinded study. But who would ever 1. Pay for it, 2. Wait that long.

[stewbrew]

The problem with large sample sizes is that random deviations show up as significant difference unless you have a responsible statistician in the team.