In recent research on ultra-processed foods (UPFs), the NOVA classification system has been widely applied; however, classification uncertainty frequently arises during its application due to ambiguous criteria and inherent limitations of dietary inta...
In recent research on ultra-processed foods (UPFs), the NOVA classification system has been widely applied; however, classification uncertainty frequently arises during its application due to ambiguous criteria and inherent limitations of dietary intake survey data. This study aimed to identify and categorize the major sources of uncertainty in the application of the NOVA system and to examine how estimates of UPF intake and associated dietary characteristics among Korean adults vary when such uncertainty is accounted for through scenario-based analyses. Data from the 9th Korea National Health and Nutrition Examination Survey (KNHANES, 2022–2023) were used. A total of 2,665 food items reported in the dietary intake survey were independently classified into NOVA groups by two researchers, who also assessed their confidence in each classification. Food items for which discrepancies remained in either NOVA classification or confidence ratings after additional review and discussion, as well as items initially judged as difficult to classify with certainty by both researchers, were defined as “classification-uncertain foods.” As a result, 360 food items (13.5%) were identified as uncertain. The main sources of uncertainty were insufficient information on additive use (44.4%), preparation or manufacturing context (42.8%), and processing methods (12.8%). In the case of preparation/manufacturing context and processing methods, uncertainty arose not only from information gaps but also from the ambiguity of the NOVA framework itself in distinguishing levels of processing based on these criteria. Approximately 43.6% of uncertain foods were traditional Korean foods, including rice cakes, kimchi, fermented soybean products, and pickled or braised dishes. Because uncertain foods could plausibly fall within a range of NOVA groups, their final classification may vary depending on researcher judgment. To account for this variability, three classification scenarios were established for uncertain foods: a lower-bound scenario, in which foods were assigned to the least processed plausible NOVA group; an upper-bound scenario, in which foods were assigned to the most processed plausible group; and a consensus scenario, in which foods were classified based on researcher agreement after comprehensive consideration of processing methods, ingredient information, and the domestic food production and distribution context. To estimate UPF consumption under each scenario, the NOVA classifications were applied to individual dietary intake data, and the percentage of total energy intake derived from UPFs was calculated for 10,598 adults aged 19 years and older. Participants were divided into quartiles based on UPF energy contribution and categorized into low (Q1), medium (Q2–Q3), and high (Q4) UPF consumption groups. Differences in general characteristics, nutrient intake, food group intake, and dietary behaviors across groups were analyzed. Categorical variables were compared using Rao–Scott chi-square tests, and continuous variables were analyzed using analysis of covariance (ANCOVA) adjusted for sex, age, or total energy intake as appropriate. Logistic regression analyses were conducted to examine associations between UPF consumption and adequacy of nutrient and food group intake. Estimated UPF energy contribution among Korean adults was 21.5% in the lower-bound scenario, 33.1% in the consensus scenario, and 38.1% in the upper-bound scenario, showing a maximum difference of 16.6 percentage points across scenarios. Across all scenarios, higher UPF energy contribution was consistently observed among men, adults aged 19–49 years, urban residents, individuals with higher educational attainment, and economically active populations. Compared with the low-consumption group, the high UPF consumption group had higher intakes of total energy, fat, calcium, riboflavin, and milk and dairy products, but lower intakes of vitamin A, thiamine, potassium, phosphorus, fruits, and vegetables. Higher UPF consumption was also consistently associated with dietary behaviors such as skipping breakfast and more frequent eating out. Logistic regression analyses showed that adequate nutrient intake (Mean Adequacy Ratio, MAR ≥ 0.75) was negatively associated with increasing UPF energy contribution in the lower-bound scenario (OR = 0.96), while no significant associations were observed in the consensus or upper-bound scenarios. Adequate food group intake (Dietary Diversity Score, DDS = 5) was negatively associated with UPF energy contribution in the lower-bound and consensus scenarios (OR = 0.93–0.95), but a positive association was observed in the upper-bound scenario (OR = 1.04). In conclusion, uncertainty in applying the NOVA classification system primarily arose from insufficient information on additive content, cooking or manufacturing environment, and processing methods. Scenario analyses accounting for this uncertainty demonstrated that estimates of UPF consumption varied substantially depending on classification assumptions. Nevertheless, across all scenarios, individuals with higher UPF consumption consistently exhibited lower fruit and vegetable intake and poorer intake of key micronutrients such as vitamin A, thiamine, and potassium. These findings suggest that reducing UPF consumption remains an important strategy for improving diet quality. However, higher adequacy ratios for riboflavin and calcium observed among high UPF consumers, as well as positive associations between UPF consumption and dietary diversity in certain scenarios, indicate that some UPFs—such as milk and dairy products—serve as important nutrient sources. Therefore, policy approaches aimed at reducing UPF consumption should consider the nutritional characteristics and functional roles of individual foods rather than adopting a uniform exclusion strategy.