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 (UPFs) to negative health outcomes. This growing interest underscores the potential value of identifying clear markers to classify UPFs, particularly to support research and regulatory efforts. OBJECTIVE: To propose replicable methods to identify UPFs, by testing the sensitivity and specificity of these methods using a large sample of packaged foods from the 2017 Brazilian Food Labels Database. METHODS: We created five scenarios to identify UPFs using substances of rare culinary use and food additives typically found in UPFs and compared them with the Nova food classification process based on the product name and food categories, considered the classic method to identify UPFs. We estimated the proportion of foods and beverages identified as UPFs using the different scenarios based on the presence of discriminative ingredients. We used a diagnostic test and a receiver operating characteristic (ROC) to understand which of the five scenarios performed better compared to the classic method to identify UPFs. Finally, we conducted a sensitivity analysis to test the role of vitamins and minerals in identifying UPFs. RESULTS: We found variations in UPFs prevalence from 47 to 72% across the five scenarios, compared to 70% using the classic method to identify UPFs in Brazilian packaged foods. The scenario using food additives of a sole cosmetic function and substances of rare culinary use (scenario 3) identified a 65% UPF, while maintaining reasonable sensitivity and specificity, and the best-performing ROC curve. There was no significant difference in identifying UPFs when comparing the addition of vitamins and minerals to the food additives with sole cosmetic function. CONCLUSION: This study shows that using ingredient-based criteria, specifically cosmetic additives and substances of rare culinary use, can reliably identify UPFs, offering reproducibility, and supporting its use in research and policy applications.