Observational Studies vs Trials in Nutrition: Why Food Evidence Is Hard

The question

Why do food studies so often say "linked to" instead of "causes"? Because most of the research is observational, and diet is one of the hardest things to study well. This page explains the two main designs and what each can and cannot tell you. It is information, not advice about what to eat.

What the two designs are

• Observational study. PubMed's definition is a study in which "the investigator does not assign participants to specific interventions". People eat as they choose and researchers follow them, often for years (a cohort study). [1]
• Randomized controlled trial. Participants are assigned to a test treatment or a control "by a random process", then followed. [1]
• Systematic review and meta-analysis. These gather and combine many studies of one question. Their strength depends on the studies they gather. See Human trials and meta-analyses of foods.

Why food research leans on observational studies

A primer on nutritional epidemiology says randomized trials provide the strongest evidence of causality, but the expense and difficulty of sustaining adherence to dietary interventions are substantial barriers. Observational designs, particularly prospective cohort studies, are therefore commonly used to study long-term links between diet and disease. [2]

A group of nutrition researchers adds a further point: they argue that criticism of the field partly comes from using the "drug trial paradigm" where it does not fit, because diet is a complex system of interacting components rather than a single treatment. [3]

What can go wrong in observational food studies

"Confounding" means that something else about the people who eat a food (for example other habits) may explain the result. The sources we read do not say that observational findings are always wrong; they say the design has known weak points. [3]

What can go wrong in food trials

• Adherence. It is hard to keep people on an assigned diet for long. [2]
• Blinding. In the CORDIOPREV diet trial the clinicians judging outcomes were masked, but participants were not. [5]
• Randomization problems. The PREDIMED trial was withdrawn and re-analyzed after the authors found protocol deviations, including people enrolled or assigned without randomization. [6]
• Surrogate measures. Many trials measure blood markers, not illness. A Cochrane review of Mediterranean-style diets found that the trials in one primary-prevention comparison did not report clinical endpoints such as heart attacks or strokes. [7]

Do the two designs agree?

A meta-epidemiological study matched 97 diet-disease pairs where both trials and cohort studies existed. On average the difference between the pooled results was small, but the prediction interval was wide, so large differences in individual comparisons could not be excluded. [8] When the type of intake or exposure was identical in both designs, the estimates were similar. [8] Most of the difference came from comparing unlike things, such as dietary supplements in trials with nutrient status in cohort studies. [8]

This is one study, not a rule. It suggests that the question asked matters as much as the design.

An example: from cohort hints to trials

The Cochrane review of Mediterranean-style diets notes that later observational studies confirmed benefits of a Mediterranean pattern for cardiovascular risk factors, while clinical trial evidence is more limited. [7] The trials that followed have mixed and qualified results; see Dietary patterns studied. The MIND diet trial, which followed observational hints, found no significant difference in thinking scores over 3 years. [9]

How news can change the message

In a study of 462 press releases from 20 UK universities, 33% contained exaggerated causal claims drawn from correlational research. When the press release exaggerated, 81% of news stories did too. [10] A headline that says a food "prevents" something may therefore come from an observational link. See How to read nutrition research headlines.