Nutritional epidemiology has long relied on standard nutritional models to examine associations between dietary exposures and health outcomes, often interpreting model coefficients with causal intent. In this paper, we use the target trial framework to clarify common causal questions in nutritional research and connect them to established tools from causal inference. Using chicken and fish as motivating examples, we define two key causal estimands relevant to dietary strategies: the total effect of chicken consumption and the comparative effect of fish versus chicken. We then use the g-formula to estimate average causal effects of the target trials and examine how common nutritional models relate to the conditional outcome model used in the g-formula estimation. We show that several standard models are re-parameterizations of the same conditional outcome model used in g-formula estimation. We also caution that standard multivariate and residual models should not be used directly within g-formula implementations without updating total energy intake under each intervention, as doing so leads to inaccurate conditional mean counterfactual outcome predictions. By providing a unified framework that links traditional models to a causal framework, our findings offer guidance on models that yield valid estimates and align with actionable dietary interventions.
Chiu et al. (Tue,) studied this question.