Should Atypical and Non-Representative Studies Such as NutriNet Santé Be Used to Drive Public Health Policy?

Background: Findings from the NutriNet-Santé studies have been used to drive public health policy in France, the European Union, and globally. The fact that NutriNet-Santé studies are not generalizable is a matter of concern. Objectives: We aimed to compare the characteristics and diet quality of consumers and non-consumers of low-calorie sweeteners (LCS) within the National Health and Nutrition Examination Survey (NHANES) sample to published within-cohort findings from NutriNet Santé. Methods: Dietary intake data for the US from two 24 h dietary recalls in 4 cycles of the NHANES 2011-18 study (n = 17,252) were used to identify LCS consumers and non-consumers. Results: LCS consumers in NHANES were more likely to be overweight, were of higher education and incomes, and had lower intakes of added sugar and higher HEI 2020 diet quality scores compared to LCS non-consumers. Based on published reports, higher LCS consumers in NutriNet Santé were normal weight, did not differ in education, did not consume less added sugar, and had lower quality diets overall. Whereas LCS consumers in NHANES were less likely to be current smokers, higher LCS consumers in NutriNet Santé were more likely to be current smokers. Based on published estimates, mean aspartame intake in NutriNet Santé was only 3 mg/day (0.045 mg/kg/day) for lower consumers and 47 mg/day (0.71 mg/kg/day) for higher consumers. Conclusions: Minimal LCS exposure and likely floor effects can be sources of statistical biases in studies of diet and health. NutriNet Santé is a large volunteer cohort of thin, educated, weight conscious French women who diet and smoke. Extreme caution is warranted when findings from atypical and non-representative samples are used to support policies in global public health.