Nutrient criteria need to be based on a unit of measure. Most systems are calculated based on per 100 kcal, per 100 g, or per serving. U.S. food labeling regulations specify reference amounts customarily consumed (RACC), providing standard serving sizes that can be used for FOP systems. In the European Union, labeling information is provided per 100 g, and many of the FOP systems in Europe use this as their basis. Each method has advantages and disadvantages as described below.
The basis of per 100 kcal allows for comparisons with nutrient recommendations and guidelines, typically expressed per 2,000 kcal. However, foods with low energy content and high nutrient content will be scored disproportionately high (Drewnowski, Maillot, & Darmon, 2009). Energy density of foods is defined as energy or calories per unit of weight. Foods with high water content are often low in energy density because water contributes weight but no calories. Green leafy vegetables have low energy density because of their high water content and are typically consumed in much smaller portions than 100 kcal, but their high nutrient content can cause extremely high scores on a per 100 kcal basis. One cup of raw spinach provides about 7 calories, and a 100 kcal portion of spinach would be about 14 cups, much more than anyone would consume; therefore, a high food score on a per 100 kcal basis would be unrealistic. For example, the NRFI score for spinach is 694.8 per 100 kcal and 135.8 per RACC (see online supporting material in Fulgoni et al., 2009). Some low-fat spreads such as light mayonnaise may exceed sodium criteria on a per 100 kcal basis but would not on a typical portion size basis (personal communication, Lisa Sutherland).
Scoring foods on a per 100 g basis, which is consistent with nutrient labeling in European countries, allows for comparisons of foods in the same category that would be served in similar amounts. However, it is not very useful for across-the-board approaches because different foods are consumed in very different amounts than 100 g (Drewnowski et al., 2009). For example, typical servings of soup or beverages may be approximately 250 g, while a typical serving of oil or butter may be less than 10 g. The water content of foods greatly influences the nutrient content per unit by weight. Sweetened beverages would benefit from a system based on 100 g that includes a limit on sugar because the sugar content of 100 g may meet the criteria. A version of the Food Standard Agency's Ofcom model makes an allowance for this by scoring beverages on a 200 g basis (Scarborough, Boxer, Rayner, & Stockley, 2007).
Both per 100 kcal and per 100 g may be difficult for consumers to understand. Europeans are used to seeing label information on a per 100 g basis. Some people understand calories, but many do not know how 100 kcal relates to a typical portion of food. Calculation on a per-serving basis makes sense because it generally represents how people eat. However, this assumes consumption patterns are similar to standardized or reference portions.
To examine the performance of four versions of a nutrient-rich scoring system calculated on a per 100 g, per 100 kcal, and per-serving basis, Drewnowski et al. (2009) calculated scores using a food list of 378 foods from a food frequency questionnaire. A version of the algorithm that used only negative nutrients expressed on a per 100 g basis penalized energy-dense foods such as butter, oils, and cheeses because of the high amount of negative nutrients in these foods and no balance of positive nutrients. The algorithm expressed on per 100 kcal produced better scores than the per 100 g algorithm for some energy-dense foods that have an energy content greater than 100 kcal per 100 g (e.g., chips and chocolate bars). Algorithms that used positive nutrients only performed similarly when calculated on 100 kcal and on RACC. A version of the NRFI that contained a combination of negative and positive nutrients performed similarly when expressed per RACC and per 100 kcal in models where food scores from the index were regressed on an overall index of dietary quality, the Healthy Eating Index (HEI) (Fulgoni et al., 2009). However, models per 100 kcal explained slightly more of the variation in HEI scores.