Why "Linked To" Is Not "Causes" in Nutrition Coverage

Why "Linked To" Is Not "Causes" in Nutrition Coverage

Nutrition stories arrive with a familiar verb. A food is "linked to" a lower risk of something, or "tied to" a higher risk of something else. The phrase is doing careful work: it signals an association observed in a population, not a demonstrated cause. Almost everything else in the article, and certainly the headline, tends to forget that distinction within a paragraph or two.

Where the numbers come from

Most diet-and-disease findings come from large cohorts in which participants filled out a food frequency questionnaire — how often, over the past year, they ate each item on a long list. Those answers are converted into estimated nutrient intakes and compared against health outcomes recorded over the following years. The method has produced genuinely useful knowledge. It also carries measurement error that nobody has solved: people do not remember a year of eating accurately, and the direction of the error correlates with weight, education and how virtuous a food is considered.

The company a food keeps

People who eat more of one thing eat less of another, and they differ in many other ways at the same time: smoking, physical activity, income, and how often they see a doctor at all. Statistical adjustment can reduce the influence of confounders that were measured. It cannot touch the ones that were not. This is the main reason observational nutrition results so often fail to reproduce when the same question is finally put to a randomised trial.

Reading the size of the effect

A relative risk of 1.2 reads as a substantial hazard and may correspond, in absolute terms, to a change of a fraction of a percentage point over a decade. Both descriptions are accurate. Only the second tells a reader what the finding would mean in a life. If the absolute rates appear nowhere in the coverage, that absence is worth noting before the finding is repeated.

A short checklist

Ask how the diet was measured and over what period. Ask how many food-outcome pairs the investigators examined, because a large table of comparisons will produce striking results by chance alone. Ask whether any randomised evidence exists and whether it agrees. When it does not, the observational result is best held as a hypothesis rather than a conclusion.

Our working notes on individual nutrition studies, each traced back to the paper it came from, are published in the MedProofs nutrition section. This article is about how evidence is read; it is not dietary advice and recommends no change to anyone's diet.