Causal criteria in nutritional epidemiology.

Making nutrition recommendations involves complex judgments about the balance between benefits and risks associated with a nutrient or food. Causal criteria are central features of such judgments but are not sufficient. Other scientific considerations include study designs, statistical tests, bias, confounding, and measurement issues. At a minimum, the set of criteria includes consistency, strength of association, dose response, plausibility, and temporality. The current practice, methods, and theory of causal inference permit flexibility in the choice of criteria, their relative priority, and the rules of inference assigned to them. Our approach is as follows. Consistency across study designs is compelling when the studies are of high quality and are not subject to biases. A statistically significant risk estimate with a > 20% increase or decrease in risk is considered a positive finding. A statistically significant linear or otherwise regularly increasing trend reinforces the judgment in favor of a recommendation. A plausible hypothesis likewise reinforces a recommendation, although the rules of inference for biological evidence are highly variable and depend on the situation. Temporality is, for nutrition recommendations, more a consideration of the extent to which a dietary factor affects disease onset or progression. Evidence supporting these criteria provides a strong basis for making a nutrition recommendation, given due consideration of the balance between presumed benefits and presumed harms. Recommendations should make clear their breadth of application; a narrow recommendation involves a single disease or condition whereas a broad recommendation involves all relevant diseases or conditions.