Lousy epidemiologic studies with self-reported data and mouse studies are often used when ideal human clinical trial data is unavailable, and their generalizability must be carefully considered.

“So we won't have that. We'll have lousy epidemiologic studies with lousy self-reported data following people large groups for long periods of time and causal inference will be fraught. We'll have mouse studies in which we're not sure we can generalize from the mouse to the human. We'll have short-term studies of people being tubefed when we want to talk about long-term studies of people eating quote unquote ordinary foods in ordinary ways.”