Almost everything that circulates about coffee and health comes from observational studies: one records what tens or hundreds of thousands of people say they drink, follows what happens to them, and compares. The result is an association, never a proof, and three mechanisms can produce it without any coffee having protected anyone.The confounding factorheavy drinkers also differ by age, income, tobacco, alcohol, sleep, occupation, and statisticians only correct for what they have measured.Reverse causationsomeone who starts to feel unwell cuts back on coffee, so that light drinkers fall ill more often, and the conclusion is reversed.The questionnaireone reports a number of cups from memory, sometimes over a year. An association repeated on several continents, with a dose relationship, that withstands corrections, becomes a serious clue; it remains a clue. The 2020 collection itself, source of the most-cited cardiovascular and metabolic figures, contains no original study on these subjects: fourteen of its twenty articles deal with athletic performance, and the general-public data come from a literature review inserted in a chapter on consumer perceptions.A collection that cites a review that cites meta-analyses that compile studies: four layers, and at each layer caveats get dropped.

MDPI 2020, pp. 207-210; LC ch. 19