Causality or confounding? Applying E values to examine associations between ultra-processed food consumption and risk of weight gain

Abstract

Abstract Background Ultra-processed food (UPF) consumption is associated prospectively with weight gain and obesity in observational studies of adults. Unaccounted for confounding is a risk when attempting to make causal inference from observational studies. Limited research has examined how feasible it is that unmeasured confounding may explain associations between UPF consumption and weight gain in observational research. Methods We introduce the E value to obesity researchers. The E value is defined as the minimum hypothetical strength of association that one or more unaccounted for confounding variables would need to have with an exposure (UPF consumption) and outcome (weight gain) to explain the association between the exposure and outcome of interest. We meta-analysed prospective studies on the association between UPF consumption and weight gain in adults to provide an effect estimation. Next, we applied the E value approach to this effect estimate and illustrated the potential role that unmeasured or hypothetical residual confounding variables could theoretically have in explaining associations. Results Higher consumption of UPFs was associated with increased weight gain in meta-analysis (RR = 1.14). The corresponding E value = 1.55, indicating that unaccounted for confounding variables with small-to-moderate sized associations with UPF consumption and weight gain (e.g., depressive symptoms, trait overeating tendencies, access to healthy and nutritious food) could individually or collectively hypothetically account for observed associations between UPF consumption and weight gain. Conclusions Unaccounted for confounding could plausibly explain the prospective association between UPF consumption and weight gain in adults. High quality observational research controlling for potential confounders and evidence from study types devoid of confounding are now needed.