A new paper argues that foods alone do not confer certain health benefits. What matters is what you replace on your plate, and that change can change how you interpret nutritional evidence.
Opinion: Is this food healthy? Reconstructing nutritional evidence through counterfactual comparisons. Image credit: Anna Puzatykh / Shutterstock
Recent opinion papers published in magazines clinical nutrition We advocate reframing nutrition research through counterfactual comparisons.
Nutritional science, despite decades of research, continues to produce conclusions that are often perceived as context-dependent or inconsistent. Nutritional science often relies on surrogate results, such as intermediate physiological measurements and biomarkers, whose interpretation depends on underlying causal contrasts. Dietary interventions are structural in nature, meaning that increasing intake of one food requires decreasing intake of another.
Therefore, studies evaluating the same food may reflect heterogeneous causal contrasts defined by different contexts and comparison targets. Pooling these heterogeneous contrasts in a meta-analysis can produce estimates that can obscure the relationship between health and diet. In this paper, the authors argue that improvements in the synthesis of nutritional evidence necessitate meta-analytic reframing through the lens of causal inference that integrates the context of comparison.
Counterfactual framework in nutritional science
To move beyond associative or descriptive interpretation to causal inference, we need to account for a counterfactual framework. Modern causal reasoning emphasizes that causal relationships are defined in relation to specific interventions or alternatives. Therefore, meaningful causal interpretation depends on the identification of exposures and counterfactual comparisons.
The assumption of consistency is an important requirement. This indicates that the results observed for a particular exposure reflect the potential consequences associated with that intervention. It requires that exposures represent well-defined interventions so that different treatment versions should not yield systematically different outcomes. Otherwise, the estimated effects will be ambiguous.
For example, red meat intake may refer to unprocessed red meat, processed meat, or meat consumed with refined carbohydrate-rich foods and vegetables. Although these scenarios have the same exposure label, they are different interventions with different health effects and biological mechanisms. Therefore, treating such different versions as interchangeable risks undermines causal inferences.
Meal substitution and relationship health effects
Many studies treat foods as having intrinsic benefits independent of dietary context. Nevertheless, the compositional nature of the diet questions this assumption. Changing a single element of your diet is not done in isolation. Instead, it addresses specific scenarios based on how other dietary component changes are allowed.
This characteristic affects the definition of the estimate and the interpretation of the results. This paper distinguishes between effects that allow for broader dietary changes and substitution effects that reflect the substitution of one food for another under constant intake. The health effects of a particular food correspond to specific substitutes rather than to the inherent properties of the food itself. For example, a randomized controlled trial (RCT) compared consumption of raw ham to consumption of cooked ham (control).
Although this intervention appeared to produce favorable changes in metabolic markers compared to controls, the interpretation critically depends on the nature of the substitutions tested. If the comparable effect is less favorable, the observed benefit would represent a relative improvement over the alternative food rather than the intrinsic cardioprotective properties of the intervention.
Network meta-analysis and causal inference
Meta-analyses of RCTs often provide the best evidence when the underlying studies investigate the same causal question. However, many nutrition meta-analyses estimate effects from contrasting different diets without making like-for-like counterfactual comparisons. Therefore, the pooled estimates lack explicit causal interpretation.
These challenges do not mean that evidence synthesis is inherently flawed. Rather, they suggest that traditional meta-analytic methods may be inadequate for compositional exposures such as diet. In contrast, network meta-analysis (NMA) provides a methodological framework that addresses some of these limitations by incorporating multiple comparators. NMA can keep dietary interventions relevant by modeling competing alternatives.
In the examples discussed by the authors, NMA reveals comparator-specific differences, whereas traditional meta-analysis may report minimal or neutral effects. Of note, NMA does not eliminate all challenges, as valid causal interpretation requires three important assumptions to be met: consistency (indirect and direct evidence are consistent), transitivity (studies are comparable across treatment contrasts), and clinical comparability (interventions are sufficiently homogeneous).
The impact of counterfactual nutrition research
Taken together, the limitations described here do not indicate a failure of meta-analysis per se. Rather, they reflect a mismatch between the causal structure of dietary exposure and the conventional evidence synthesis. Approaches like NMA preserve the comparator structure and thus work better with causal inference.
Nevertheless, methodological tools alone are not sufficient. Therefore, to improve nutritional science, research questions must be reframed to reflect clearly defined counterfactual contrasts. The authors also call for clearer exposure definitions and more transparent reporting of alternative situations and energy balances. From “Is that food healthy?” to “Is this food healthy compared to what?” Nutrition translations can improve relevance, consistency, and interpretability.
