[drgo]

We have been here before many times. Nutritional epidemiology studies have a terrible track record of establishing causal relationships (e.g., Beta-carotene and lung cancer, selenium and prostate cancer, etc all were not replicated when the definitive clinical trials were done). The problem is that statistical models with questionable and often untestable assumptions are used, but the results are reported as if these models were fault-less. The result is overly optimistic estimates of statistical significance and inflated confidence in study findings.

[KempyKolibri]

I would disagree with this. While we can always point to examples where epi did not align with RCTs, this doesn’t capture how discordant (or not) this relationship is in the aggregate.

Thing is, we actually have empirics on this, and in reality observational studies comparing intake to intake are concordant in over 90% of cases, so I think we actually have a very strong case for making causal inferences based on replicated epi findings:

https://www.bmj.com/content/374/bmj.n1864