Observational epidemiology [...] should only be used to generate hypothesis and design interventional studies.
I feel like I've seen this same thing said at least three times in the past few days. It used to be that we all just went "correlation does not imply causation". What changed?