Accuracy in Labeling: A Study on the Composition and Nutritional Grading of Freshly Made Beverages in China

Background: Freshly made beverages (FMBs) have become an increasingly important source of sugar and fat intake in China. However, the on-site preparation and product customization of these beverages make conventional tolerance frameworks for nutrition labeling difficult to implement. Thus, this study aimed to assess the accuracy of Nutri-Choice, a voluntary graded nutrition labeling system piloted in Shanghai, by comparing brand-registered nutrient values with laboratory-measured values under routine retail conditions. Methods: Three major chain brands involved in the initial Nutri-Choice pilot were included. The system categorizes beverages into four color-coded grades (A to D) according to predefined thresholds for non-milk extrinsic sugars (NMESs, defined as total sugars minus lactose), saturated fatty acids (SFAs), trans fatty acids (TFAs), and non-sugar sweeteners. In total, 122 beverages were selected through stratified random sampling and analyzed for NMESs, SFAs, and TFAs. Laboratory-measured values were compared with brand-registered values using relative error analysis, weighted agreement between actual grading and brand-registered grading, Cohen’s kappa statistics, and Bland–Altman analysis. Results: Among the 122 samples, pure tea accounted for 9.02%, fruit tea and fruit juice for 28.69%, and milk coffee and milk tea for 62.29%. The median contents of NMESs, SFAs, and TFAs were 3.63 [1.31, 5.72], 0.90 [0.00, 2.10], and 0.00 [0.00, 0.00] g/100 mL, respectively. Grade B beverages were the most frequent (46.72%), followed by Grade C (25.41%), Grade D (22.95%), and Grade A (4.92%). The NMES concentrations increased from Grades A to C, whereas those in Grade D were mainly determined by SFA levels. Weighted relative error analysis showed that 85.63% of the NMES measurements, 79.38% of the SFA measurements, and 98.75% of the TFA measurements satisfied the ≤120% claimed-value criterion. Bland–Altman analysis revealed a significant negative bias for NMESs, suggesting that laboratory-measured values were, on average, 0.82 g/100 mL lower than brand-registered values. No systematic bias was detected for SFAs or TFAs. The initial weighted grade agreement was 81.88% (Kappa = 0.688), which increased to 88.68% (Kappa = 0.812) after applying a 120% adjustment coefficient. Conclusions: Nutri-Choice provides a practical framework for graded nutrition labeling of FMBs, although the credibility of this framework depends on routine market sampling, laboratory verification, and evidence-based tolerance approaches. The observed variation in grading agreement among brands indicates the need for further assessment of differentiated tolerance.