Empirical Comparison Of Approaches To Mediation Analyses In Nutrition Research With Applications To The Quality Cohort

Background: Adiposity may mediate the effect of dietary glycemic load (GL) on lipid profile in children as studies have shown an association between dietary GL and adiposity and between adiposity and lipid profiles. While several approaches for conducting mediation analysis have been proposed, given the strict assumptions and restrictions for each method, it is not clear which method is ideal in nutrition research. Our objective was to compare the conventional approach and the causal approach using marginal structural models (MSM) with inverse probability weights (IPW) to examine adiposity as a mediator in the association between baseline dietary GL and lipid profile after 2 years. Methods: The QUALITY cohort included 630 children, 8-10 years at recruitment with at least one obese parent followed-up 2 years later. Three separate 24-hour dietary recalls were administered by a dietitian at baseline. Child and parent characteristics were obtained through direct measurement (blood lipids, anthropometrics) or questionnaires (socio-economic characteristics). Indicators of adiposity, including BMI z-score and percent fat mass, were the mediators of interest. A conventional approach of statistical adjustment for the mediator was used. As well, we used MSM to estimate the controlled direct effect (CDE) between GL and blood lipids 2 years later not mediated by adiposity. The MSM models were constructed using IPW truncated at 5-95 percentile. Results: Mean age of children at baseline was 9.6 years and 33% were overweight or obese. Both mediation analysis methods revealed that most of the effect of GL on blood lipids was mediated by adiposity. Results obtained from both methods differed; while the MSM with IPW did not show evidence of a direct effect of GL on TG not via percent fat mass (TG: β=0.06, 95%CI=-0.01, 0.12; HDL: β=-0.01, 95%CI=-0.04, 0.03), the conventional method did (TG: β=0.03, 95%CI=0.003, 0.06; HDL: β=-0.02, 95%CI=-0.03, -0.001). Conclusion: In conclusion, our results suggest that adiposity contributes substantially to the association between GL and blood lipids after 2 years. MSM with IPW may not be appropriate in nutritional studies with small sample sizes.