We have read with interest the letter by Gomes et al. (1) suggesting that methodological artifacts biased some of our analyses on the relationship between ultra-processed food (UPF) subgroups and risk of type 2 diabetes (T2D) (2). We would like to take the opportunity to demonstrate how that is unlikely to be the case.
Potential multicollinearity between total UPF, UPF groups, and UPF subgroups was the first concern. In the main analyses, total UPF intake was quantified using servings/day (Table 2 in our article). In sensitivity analyses, total UPF intake was modeled using four alternative metrics: calories (kcal) from UPF/day, percentage of kcal from UPF/day, percentage of grams from UPF/day, and energy-adjusted servings of UPF/day (Supplementary Table 5 in our article). None of these analyses were affected by multicollinearity, as only total UPF intake was included in the models. We additionally investigated the relationships between nine different UPF groups and T2D risk (Fig. 1 in our article). The groups were simultaneously included in the models, without total UPF intake. Correlation coefficients between intakes of these nine subgroups were considerably low, ranging from −0.02 to 0.35. Finally, we repeated the analyses by expanding three groups (ultra-processed breads and cereals, packaged sweet snacks and desserts, and artificially and sugar-sweetened beverages) into subgroups, including a total of 14 groups and subgroups in the models. Again, the correlations between groups and subgroups were minor, with coefficients ranging from −0.23 to 0.35, suggesting that multicollinearity would not have undermined the reliability of our results.
Concerns with risk of false-positive findings due to multiple testing in group/subgroup analyses were also raised. UPF groups/subgroups with a significant relationship with T2D risk had P values <0.0001. The only exceptions were yogurt and dairy-based desserts and ultra-processed dark breads and whole-grain breads, for which the P value was 0.005. Setting statistical significance to 0.006 (i.e., 0.05/9 comparisons) to account for multiple testing would still allow for these results to be considered significant or marginally significant. As mentioned in the article, the inverse associations between T2D risk and packaged sweet snacks and desserts as well as packaged savory snacks remain unclear, and residual confounding cannot be ruled out.
Third, it was highlighted that non-UPF consumption was similar across quintiles of UPF intake, and, as a result, the use of servings/day instead of proportional contribution of non-UPF to total energy introduced confounding. In fact, we repeated all our analyses using four alternative metrics for UPF, including with proportional contribution to total energy as well as by controlling for non-UPF intake rather than total energy (Supplementary Tables 5 and 6). The consistency in our findings across multiple metrics and approaches underscores the confidence in our original results.
Fourth, the letter offers suggestions for isocaloric replacement analyses. Our article did not include such analyses, although we could certainly consider substitution analyses in future research.
Overall, we are confident in our conclusions that are supported by the data. Our study supports the recommendations of limiting total UPF consumption, especially those associated with a higher risk of T2D.
Article Information
Funding. The original analysis discussed in this letter was conducted in the Nurses’ Health Study (NHS), NHSII, and Health Professionals’ Follow-up Study cohorts. The NHS and NHSII studies and the Health Professionals’ Follow-up Study are supported by National Institutes of Health grants UM1 CA186107, P01 CA87969, R01 CA49449, R01 HL034594, R01 HL088521, U01 CA176726, R01 CA67262, U01 CA167552, R01 HL035464, R01 HL060712, R01 DK120870, and U01 HL145386. J.-P.D.-C. is a research scholar of the Fonds de Recherche du Québec–Santé (Quebec Health Research Funds).
Duality of Interest. J.-P.D.-C. received speaker and consulting honoraria as well as investigator-initiated research funding from the Dairy Farmers of Canada in 2016, 2018, and 2021, outside the submitted work. No other potential conflicts of interest relevant to this article were reported.
