Nutrition/Obesity
Meta-analyses of nutritional exposures must identify and distinguish between study estimands: a pilot study of an illustrative review Natalia Ortega* Natalia Ortega Peter Tennant Octavio Pano Georgia Tomova
Background: Answering causal questions is a common aim of nutrition research. Meta-analyses (MAs) aim to produce definitive estimates of causal effects by synthesising findings from multiple studies. Unfortunately, few nutrition studies are explicit about their target causal estimands, making it challenging to appropriately pool their estimates and interpret them. In dietary data, it is particularly important to consider the energy adjustment strategy; because adjusting for, or dividing by, total energy will change the effect from an additive to a substitutive effect. It is unclear, however, whether such issues are widely considered in primary or secondary research.
Methods: To explore the reporting of target estimands in nutrition research, and the extent that MAs pool different estimands, we systematically searched for all MAs on the illustrative topic of saturated fats and cardiovascular disease incidence. For this pilot study, we identified the most cited MA and doubly-extracted data from all contributing primary studies.
Results: Among the 21 primary studies included in the illustrative MA, only one defined a target estimand within their aim. Thirty-eight unique models were examined, of which 32 targeted substitutive effects and 6 additive effects. Among the studies with substitutive models, 70% did not mention the energy adjustment strategy of choice, nor mentioned a comparator. Substitution models typically adjusted for alcohol, fiber, vegetables, fruits and intake of different types of fat. Only 5 out of 32 models were correctly interpreted as substitutive by the authors, and 2 out of 6 as additive in the results. Twenty out of 38 models did not provide an interpretation for the estimate.
Conclusion: MAs should differentiate between study estimands and aim to pool estimates for the same estimand where possible, or as a minimum – differentiate between substitutive and additive effects.