Healthier carbohydrate (carb)-rich foods are essential for health, but practical, validated indices for their identification are not established. We compared four pragmatic metrics, based on, per 10g of carb:(a) 1g fiber (10:1 carb:fiber), (b) 1g fiber and <1g free sugars (10:1:1 carb:fiber:free sugars), (c) 1g fiber and <2g free sugars (10:1:2 carb:fiber:free sugars); and (d) 1g fiber and, per each 1 g of fiber, <2g free sugars (10:1 carb:fiber, 1:2 fiber:free sugars; or 10:1|1:2). Using 2013–2016 National Health and Nutrition Examination Survey /Food and Nutrient Database for Dietary Studies, we assessed, overall and for 12 food categories, whether each metric discriminated carb-rich products higher or lower (per 100g) in calories, total fat, saturated fat, protein, sugar, fiber, sodium, potassium, magnesium, folate,and 8 vitamins/minerals. Among 2,208 carb-rich products, more met 10:1 (23.2%) and 10:1|1:2 (21.3%), followed by 10:1:2 (19.2%) and 10:1:1 (16.4%) ratios, with variation by product sub-categories. The 10:1 and 10:1|1:2 ratios similarly identified products with lower calories,fat, free sugars, and sodium; and higher protein, fiber, potassium, magnesium, iron, vitaminB6, vitamin E, zinc and iron. The 10:1:2 and 10:1:1 ratios identified products with even larger differences in calories and free sugars, but smaller differences in other nutrients above and lower folate, thiamine, riboflavin, and niacin; the latter findings were attenuated after excluding breakfast cereals (~9% of products). These novel findings inform dietary guidance for consumers, policy, and industry to identify and promote the development of the healthier carb-rich foods.
Clearly, practical, validated metrics are needed to help consumers identify and guide the food industry to develop and promote healthier carbohydrate products. A recent convening by the U.S. Inter agency Committee on Human Nutrition Research concluded that definitions of healthier grain foods are not established and are urgently needed . As one potential metric,the American Heart Association (AHA) has recommended choosing carb-rich foods with at least 1 g of fiber for every 10 g of total carbohydrate (carb:fiber ratio <10:1), based on the ratio of carbohydrate-to-fiber in whole wheat [9, 10]. This ratio aims to incorporate the overall relative amounts of starch and sugars vs. whole grains, bran, and other sources of fiber (e.g. added seeds). Researchers have found that the 10:1 ratio is more effective at identifying cereals-based products with better nutritional quality than other approaches that utilize the ingredients [6,11].
While the 10:1 ratio appears to be a valid, practical metric for identifying healthier carb-
rich foods, several questions remain. First, prior studies have evaluated available products in major supermarkets, rather than actual national intakes of products. Second, the 10:1 ratio explicitly does not differentiate contents of starch vs. sugar. With the growing emphasis on reducing added or free sugars [1, 12, 13], it is unclear whether modified ratios with additional restrictions on free sugar content may improve performance and/or limit available product choices for consumers. Given the public health impact of carbohydrate quality, it is crucial to test practical metrics for identifying carb-rich products with higher nutritional quality.
To address these gaps in the knowledge, we investigated how four different metrics, including the 10:1 ratio and three additional metrics further limiting free sugars, related to the nutritional quality of carb-rich products in the US.
Materials and methods
We evaluated four metrics of carbohydrate quality based on the ratios of total carbohydrate, fiber, and/or free sugars. These metrics were selected as pragmatic in that they are widely measured, based on nutrients commonly listed on product labels, and more straightforward than complex nutrition profiling scoring schemes. The first was the original validated 10:1 carb: fiber ratio (10:1),[14–17] defined as �1 g of fiber per 10 g of carbohydrate. We evaluated three additional metrics incorporating additional restrictions on free sugars contents, per 10g of carb, including:
• �1 g of fiber and < 1 g of free sugars (10:1:1 ratio of carb:fiber:free sugars, or 10:1:1);
• � 1 g of fiber and < 2 g of free sugars (10:1:2 ratio of carb:fiber:free sugars, or 10:1:2); and• �1 g of fiber and, per each 1 g of fiber, <2g free sugars (10:1 ratio of carb:fiber + 1:2 ratio of fiber:free sugars; or 10:1|1:2).
The 10:1:1 was selected based on the recommendation of consumption of 50% of energy from carbohydrates, 5% of energy from free sugars and 30g of fiber [18,19]. The 10:1:2 was similarly selected, but with the more permissive assumption that the 10% energy limit on added sugars would largely be derived from these carb-rich foods . The 10:1|1:2 was selected to limit the amount of free sugars depending on the fiber (rather than carbohydrate)content of the product, based on the WHO recommendations to consume 30 g/d of fiber and no more than 10% energy (60 g on a 2400 kcal/d diet) from free sugars .
Identification of carb-rich products consumed in the US
We combined data from the two most recent versions of the Food and Nutrient Database for Dietary Studies (FNDDS, 2013–2014 and 2015–2016) , corresponding to the National Health and Nutrition Examination Survey (NHANES) for this period. These data were supplemented with information from the Food Patterns Equivalents Database (FPED) , which includes the amounts of grains and other food components in these products. Information on free sugars content was adapted from the UK definition  for application in FNDDS using added sugars (e.g., honey, white sugars, syrups), sugars from fruit juice, other sugars present in beverages (excluding natural dairy sugars), and sugars from extruded fruit/vegetable products.We selected categories of carb-rich products for which metrics of nutritional quality would be most relevant to consumers and industry, based on the What We Eat in America (WWEIA)
food classification developed by the United States Department of Agriculture (USDA) .
We selected 12 total categories, including breads, rolls, and tortillas; quick breads; ready-to-eat(cold) cereals; cooked cereals; cooked grains; sweet bakery products; savory snacks; crackers; snack/meal bars; smoothies and grain drinks; baby food: cereals, snacks, sweets; and relevant mixed dishes (e.g., rice/pasta mixed dishes). We focused on these variably processed foods, and did not include unprocessed foods (e.g., fruits, vegetables, legumes). The flowchart of the numbers of food items included is shown in S1 Fig.
We evaluated the identified products in these categories both overall and weighted by the frequency of consumption among US adults and children from the two most recent corresponding cycles of NHANES (2013–2014, 2015–2016). Nationally weighted analyses were considered overall and separately for adults age 20+ years and children age 2–19 years.
To assess the nutritional profiles of each product, we utilized FNDDS data to quantify the contents of calories, total fat, saturated fat, protein, total sugar, free sugars, added sugar, fiber, sodium, potassium, magnesium, folate, thiamine, riboflavin, niacin, vitamin B6, vitamin B12, vitamin E, zinc and iron. All nutrients were selected a priori based on those commonly contained in whole grains or fortified in grain-rich foods. To account for varying serving sizes of different products, all analyses were standardized to 100g servings.
In addition to the individual nutrients above, we also evaluated the extent to which each
product met selected nutrient profiling systems (NPS) currently used in practice, including from the UK Ofcom (the Office of Communications) , FSANZ (Food Standards Australia New Zealand)  and the Health Star Rating (Australia New Zealand) . We recognized that these nutrient profiling systems do not capture all aspects of nutritional quality, in particular higher contents of many healthful compounds. Yet, these systems are currently being used by governments, for example Ofcom and FSANZ Nutrient Profiling Scoring Criterion (NPSC)to determine if a product can carry a particular ‘health claim’ [29,30]; and the Health Star Rat-ing (derived from NPSC with a slightly differing algorithm) for front-of-pack labeling for packaged foods . In brief, foods were first classified into specific categories with two for the Ofcom model, three for the FSANZ NPSC and six for the HSR. Then, a summary score was calculated for each food product based on points for both nutrients to limit (energy, saturated fat, total sugar, and sodium) and nutrients or food components to encourage (protein, fiber,
and percent composition of fruits, vegetables, nuts and legumes (FVNLs) in a product). Points are allocated based on the nutritional content in 100g of a food or drink. The details of components and scoring of these three nutrient profiling systems are presented in the Supplemental Material.
For all eligible carb-rich products, we assessed the numbers and proportions of products meeting each metric, both overall (all available products) and weighted by their consumption levels in the U.S. Kappa statistics were calculated to assess chance-corrected agreement between the metrics.
Our primary outcomes were the nutritional differences according to each metric for all
carb-rich foods combined, which can be considered the mean differences in nutrient contents if consumers were to make selections based on these metrics among different carb-rich products. The dependent variables were each of the proposed nutrients in nutritional profile. The independent variable was a dichotomous indicator of products meeting or not meeting each ratio. We used robust standard error of variance in the regressions to account for non-normally distributed nutrients . In secondary analyses, we repeated the analyses among each of the 12 food categories, which can be considered the mean differences in nutrient contents if consumers were to make selections based on these metrics within specific food categories. To assess and compare the proportion meeting each of the three NPS thresholds, we used similar approaches with logistic regression.
A 2-tailed P value < .05 was taken to indicate statistical significance. All statistical analyses were conducted in Stata software, version 14.2 (Stat Corp LLC, College Station, Texas), and R, version 3.4.2 (R Foundation for Statistical Computing).
A total of 2,208 available carb-rich products were identified in FNDDS, including 386 (17.5%) sweet bakery products; 206 (9.3%) breads, rolls, and tortillas; 198 (9.0%) cold cereals; 197 (8.9%) cooked cereals; 177 (8.0%) savory snacks; 143 (6.5%) quick breads and bread products;82 (3.7%) cooked grains; 80 (3.6%) crackers; 46 (2.1%) snack/meal bars; 19 (0.86%) smoothies and grain drinks; 55 (2.5%) baby foods; and 619 (28.0%) mixed dishes (Table 1).
Proportions of food products meeting each proposed metric
Among these food products, more than 3 in 4 (>76.8%) did not meet any of the metrics. The highest number met the 10:1 (23.2%), followed by the 10:1|1:2 (21.3%), 10:1:2 (19.2%), and
10:1:1 (16.4%) (overall agreement across metrics [kappa] = 0.88) (Table 1). Considerable variation was identified across the different food categories. For example, 43.9% of cold cereals, 38.6% of cooked cereals, and 31.1% of breads, but only 2.6% of sweet bakery products, met the 10:1. Agreement across the metrics was also highly variable by food category. For example, agreement was very high (kappa> = 0.98) for cooked grains, savory snacks, and mixed dishes.Agreement was lowest (kappa< = 0.50) for cold cereals, snack/meal bars, and smoothies/grain drinks. Generally, the 10:1 and 10:1|1:2 identified a more similar proportion of products meeting their criteria, except for cold cereals (43.9 vs. 34.3%, respectively), snack/meal bars (23.9 vs. 13.0%), and smoothies/grain drinks (42.1 vs. 0%). The 10:1:1 was most restrictive, with largest differences compared to other metrics in products identified for cold cereals (e.g., only 9.1% met 10:1:1, vs. 20.7% for 10:1:2), crackers (21.3 vs. 31.3%), and breads (24.8 vs. 31.1%). Across all products, inter correlations between the four metrics ranged from 0.79 for 10:1 and 10:1:1 to 0.95 for 10:1 and 10:1|1:2 (S1 Table in S1 File).
Weighted by the frequency consumption, the general patterns across metrics and food categories were similar (Table 2). However, the overall percentages of products meeting each metric were lower when weighted by consumption as compared to available choices, indicating that products not meeting the metrics were more frequently consumed. For example, while 16.4% of available products met the 10:1:1 ratio (Table 1), only 9.7% of products met this metric when weighted by actual consumption levels. The percentage of food products meeting each metric was also generally lower among foods consumed by children as compared with those consumed by adults. For example, overall, 15.5% of products consumed by children vs.19.7% of products consumed by adults met the 10:1 ratio.
Nutritional quality of food products meeting or not meeting each proposed metric
When we assessed the nutrient contents of products meeting or not meeting each proposed metric, all metrics each identified products with significantly lower calories, fat, and sugars, and higher protein, fiber, potassium, and magnesium. The 10:1 and 10:1|1:2 metrics each could further identify products with higher vitamin B6, vitamin E, zinc, and iron. The 10:1, 10:1:1 and 10:1|1:2 metrics each could identify products with lower sodium. The 10:1 metric is the only one, which can identify higher values of vitamin B12 (Table 3). Compared to the 10:1 and 10:1|1:2 ratios, the 10:1:2 and 10:1:1 metrics identified larger differences in contents of calories and sugars, but smaller differences in several of the other nutrients above including protein, fiber, potassium, and magnesium. In addition, the 10:1:2 and 10:1:1 ratios also identified products with significantly lower folic acid, thiamine, riboflavin, and niacin than products not meeting these ratios.
In sensitivity analyses, different food categories were separately evaluated. Excluding cold cereals, the different levels of folic acid, thiamine, riboflavin, and niacin among products meeting vs. not meeting the 10:1:2 and 10:1:1 ratios were smaller and were also more similar to findings for the 10:1 and 10:1|1:2 ratios (S3 Table in S1 File). Excluding mixed dishes, 10:1 and 10:1|1:2 ratios performed similarly for identifying products with higher nutritional quality except not as well for folic acid, thiamine, riboflavin, and niacin (S4 Table in S1 File). Findings for each of the individual food categories were shown in (S5–S14 Tables in S1 File); because some of these categories had relatively small numbers of products reported consumed, these subgroup findings should be interpreted in that light.
Relationships of each metric with selected nutrient profiling systems
When we evaluated how the metrics identified carb-rich products meeting NPS-defined criteria for healthfulness, each of the four metrics generally performed similarly (Fig 1, S2 Table in S1 File). Across the NPSs, about 70–85% of products that met each metric also met desirable thresholds for the NPSs (highest for Health Star Rating, lowest for Ofcom). In contrast, among products that did not meet each metric, only about 38–45% met desirable thresholds for the NPSs. Generally, products consumed by children were less likely to meet the NPSs compared to products consumed by adults. The relationship between each metric and meeting each of the three NPSs across the 12 food categories was shown in S15 Table in S1 File and comparing
available carb-rich products meeting each of the three NPSs, each of the four metrics was shown in S16 Table in S1 File.
DiscussionIn this study based on a nationally representative U.S. sample, we investigated how four pragmatic metrics assessed the nutritional quality of 2,208 carb-rich foods and beverages, including the previously validated 10:1 carb:fiber ratio and three additional metrics that further restricted free sugars. Across the four metrics, between 1 in 4 (23.2%; 10:1) and 1 in 6 (16.4%; 10:1:1) products met the criteria. All proposed metrics identified products with generally higher nutritional quality; overall, the 10:1 and 10:1|1:2 ratios appeared to perform best in terms of nutrient differences as well as greater consumer product choices.
Meaningful differences were also identified by food categories. Generally, cold cereals,
cooked cereals, cooked grains, breads, and crackers had the largest proportions of products meeting these metrics, while sweet bakery products, quick breads/bread products, and snack/meal bars had the lowest. These findings identify food categories where carbohydrate quality is poor and thus indicate where industry could improve carbohydrate quality of products by increasing the fiber and reducing sugar contents. Although baby food was included in our categorization, our expectation was that few baby foods would meet any of the metrics because formulations may be limited by regulatory guidance . Differences between metrics were also most pronounced in certain food categories. For example, compared to the 10:1 ratio, restricting sugars as a percentage of total carbs (10:1:1, 10:1:2) greatly diminished the number of cold cereals meeting the criteria, while restricting sugars as a ratio to fiber (10:1|1:2) had
smaller effects. Restricting the content of free sugars (as opposed to the sum of starch + sugar)also eliminated all smoothies and (except for 10:1|1:2) all snack/meal bars.
The interpretation of the value of these different identifications partly depends on the scientific assessment of relative harms of refined starch vs. sugar. If refined starches vs. sugars in
foods are considered metabolically similar for health (e.g., based on glycemic responses and associations with long-term weight gain), then the additional product eliminations may not be useful [34, 35]. If free sugars are considered worse for health than refined starch (which appears true for sugar-sweetened beverages, but more uncertain and controversial for sugars vs. starch in foods), then the 10:1:1, 10:1:2, and 10:1|1:2 ratios may be preferable . Among these latter three, our findings highlight a key question for future research, that is whether bio-logic effects of sugars should be considered relative to overall carbohydrate (10:1:1, 10:1:2) or to dietary fiber (10:1|1:2).
The 10:1:2 and 10:1:1 ratios identified products with smaller differences in several nutrients(protein, fiber, potassium, magnesium), and lower levels of other nutrients (folic acid, thiamine, riboflavin, niacin), largely due to restriction of cold breakfast cereals. Among such cereals, many that are rich in whole grains also contain some added sugars to increase palatability,while those with no added sugars are often essentially 100% refined starch (e.g. corn starch).
These findings raise the important question of focusing on added sugars alone, which may drive consumers and industry formulation toward cereals high in refined starches and low in whole grains. Our results also suggest a need to understand whether consumers will use such labels or any other alternatives to select carb-rich foods overall or within specific subcategories.
To our knowledge, this is the first investigation to evaluate the healthfulness of various
carb-rich products as consumed by American adults and children. By all of the metrics, children consumed a smaller proportion of healthier carb-rich products than did adults.
These findings provide necessary quantitative data on the conventionally recognized gaps in the nutritional quality of carb-rich products promoted to and consumed by children in the US, including which food categories are in particular need of improvement. Each of the four metrics performed reasonably well, and choices of which to use may partly depend on which nutrients are prioritized for any particular population or health priority. In addition, most consumers may be unlikely to calculate these metrics for themselves other