Testing nutrient profile models using data from a survey of nutrition professionals.
Scarborough P., Boxer A., Rayner M., Stockley L.
OBJECTIVE: To compare nutrient profile models with a standard ranking of 120 foods. DESIGN: Over 700 nutrition professionals were asked to categorise 120 foods into one of six positions on the basis of their healthiness. These categorisations were used to produce a standard ranking of the 120 foods. The standard ranking was compared with the results of applying eight different nutrient profile models to the 120 foods: Models SSCg3d and WXYfm developed for the UK Food Standards Agency, the Nutritious Food Index, the Ratio of Recommended to Restricted nutrients, the Naturally Nutrient Rich score, the Australian Heart Foundation's Tick scheme, the American Heart Association's heart-check mark and the Netherlands tripartite classification model for foods. Rank correlation was assessed for continuous models, and dependence was assessed for categorical models. RESULTS: The continuous models each showed good correlation with the standard ranking (Spearman's rho = 0.6-0.8). The categorical models achieved high chi(2) results, indicating a high level of dependence between the nutrition professionals' and the models' categorisations (P < 0.001). Models SSCg3d and WXYfm achieved higher scores than the other models, implying a greater agreement with the standard ranking of foods. CONCLUSIONS: The results suggest that Models SSCg3d and WXYfm rank and categorise foods in accordance with the views of nutrition professionals.