Applying the Nova food classification to food product databases using discriminative ingredients: a methodological proposal

Abstract

Background: Growing interest in the Nova food classification system surged among various stakeholders, driven primarily by compelling evidence linking the consumption of ultra-processed foods (UPF) to negative health outcomes. However, the conventional classification process often leaves room for uncertainty and operational challenges. Objective: This study aimed to develop and test a replicable method to identify UPF based on discriminative food composition ingredients using the 2017 Brazilian Food Labels Database and evaluate the sensitivity of this method in comparison with the identification of UPF by food name and food category. Methods: We created six scenarios to identify UPF using food additives and food substances used in the definition of UPF and compare them with the classic methodology of Nova classification based on product name and food categories. We estimated the proportion of foods and beverages identified as UPF according to the different scenarios based on the presence of these discriminative ingredients, total and per food category. Using a diagnostic test and a receiver operating characteristic (ROC) curve, we compared the UPF identified through each of the six scenarios with the ones identified through the classic method. Results: We found variations in UPF prevalence from 55% to 72% across scenarios, compared to 70% using the classic method in Brazilian packaged foods. Despite its cautious approach, the scenario using food additives with exclusively cosmetic functions and food substances effectively identified a significant portion of UPF, while maintaining satisfactory sensitivity and specificity, and a better performance on the ROC curve. Conclusion: This methodological study emphasizes the importance of detailed criteria to identify UPF, offering researchers alternative and standardized methods for safe decision-making.