A Systematic Methodology to Estimate Added Sugar Content of Foods

Home , Added sugar, Food composition data, Sugar industry

ORIGINAL ARTICLE A systematic methodology to estimate added sugar content of foods

JCY Louie1,2,3, H Moshtaghian1, S Boylan2, VM Flood4,5, AM Rangan3, AW Barclay3,6, JC Brand-Miller2,3 and TP Gill2

BACKGROUND/OBJECTIVES: The effect of added sugar on health is a topical area of research. However, there is currently no analytical or other method to easily distinguish between added sugars and naturally occurring sugars in foods. This study aimed to develop a systematic methodology to estimate added sugar values on the basis of analytical data and ingredients of foods. SUBJECTS/METHODS: A 10-step, stepwise protocol was developed, starting with objective measures (six steps) and followed by more subjective estimation (four steps) if insufﬁcient objective data are available. The method developed was applied to an Australian food composition database (AUSNUT2007) as an example. RESULTS: Out of the 3874 foods available in AUSNUT2007, 2977 foods (77%) were assigned an estimated value on the basis of objective measures (steps 1–6), and 897 (23%) were assigned a subjectively estimated value (steps 7–10). Repeatability analysis showed good repeatability for estimated values in this method. CONCLUSIONS: We propose that this method can be considered as a standardised approach for the estimation of added sugar content of foods to improve cross-study comparison. European Journal of Clinical Nutrition (2015) 69, 154–161; doi:10.1038/ejcn.2014.256; published online 17 December 2014

INTRODUCTION remain inconsistent and most methods require a high level of The term ‘added sugar’ is usually understood to mean sugar understanding of food composition to make subjective decisions, added to foods during processing. In many countries, the majority or they require additional data from the food industry. For of sugar added to food is in the form of refined sucrose, but it may example, the 59-step method proposed by Roodenburg et al.19 include other monosaccharides and disaccharides containing used average values (as proportion of total sugars) for many of the ingredients such as glucose and fructose (and their syrups), as packaged foods, for example, canned vegetables in syrup and well as corn syrup or high-fructose corn syrup (used more cornflakes, as well as data from the food industry. Many steps in regularly in the United States of America).1 Added sugar is a prime that method are also specific to a single food group. On the other target for nutrition intervention, as it provides ‘empty calories’,or hand, the method used by the US Department of Agriculture calories with little or no associated nutrients. Studies have shown (USDA) was not outlined in detail,11 precluding adoption by other that a high intake of added sugar (e.g., 420% of energy) can – researchers. Differences in product formulations between dilute the nutrient content of the diet2 5 and increase the total 22 6 7 countries also mean that methods based on food composition daily energy intake, potentially resulting in weight gain. Limiting are unlikely to be reliable in other countries without further the intake of added sugar has been advocated by many key fi 8–10 modi cation. government and public health agencies. Newer analytical methods that allow the amount of individual fi Dietary intake of added sugar is dif cult to assess accurately, as sugars to be quantified using high-performance liquid there are no analytical methods that distinguish between added chromatography23 have been incorporated in some sugar and naturally occurring sugars such as those in fruits, methods,11,19 improving the reliability of estimating added sugar vegetables and milk. Largely for this reason, the provision of content of foods. It should be noted that although high- added sugar content on food labels is not mandatory. Several performance liquid chromatography is able to identify individual countries have attempted to provide food composition databases with added sugar values estimated from the combined composi- types of sugars, it is still unable to distinguish between naturally tion and ingredient lists provided by food manufacturers.11–13 This occurring sugars and added sugars. In some cases, individual has enabled the exploration of the association between added sugar types could reasonably be assumed to be naturally sugar and health outcomes.14–18 Unfortunately, in many other occurring, for example, lactose in dairy foods. The high cost countries, including Australia, such information is not readily associated with high-performance liquid chromatography makes available. testing every single food item in the food composition database Various methods to estimate added sugar contents of foods prohibitively expensive. As a result, database providers such as have been described previously.11,19–21 Although the principle of Food Standards Australia and New Zealand have prioritised their added sugar estimation of these methods is the same, that is, laboratory testing to examine foods that are likely to contain added sugars = total sugars − naturally occurring sugars, they different types of sugars.24

1School of Medicine, Faculty of Science, Medicine and Health, The University of Wollongong, Wollongong, NSW, Australia; 2Boden Institute of Obesity, Nutrition, Exercise and Eating Disorders, The University of Sydney, Sydney, NSW, Australia; 3School of Molecular Bioscience, The University of Sydney, Sydney, NSW, Australia; 4Faculty of Health Sciences, The University of Sydney, Sydney, NSW, Australia; 5St Vincent Hospital, Sydney, NSW, Australia and 6Australian Diabetes Council, Glebe, NSW, Australia. Correspondence: Dr JCY Louie, School of Molecular Bioscience, The University of Sydney, Level 4 East, The Hub, D17 Charles Perkins Centre, Sydney, NSW 2006, Australia. E-mail: [email protected] Received 29 May 2014; revised 9 October 2014; accepted 15 October 2014; published online 17 December 2014 Method for estimating added sugar content JCY Louie et al 155 To provide reliable between-country comparisons of added excluded, as some of them may also be yeast-free, meaning that all sugars sugar intake, a standardised objective methodology that allows for added to the recipe serve only as a sweetener and not as a processing aid. the differences in product formulation and cultural preparation of Pastries with dried fruits and/or nuts were excluded from this step, as dried foods is required. The aim of this study was to develop a fruits containing added sugar may be used, and some pastries with fillings systematic methodology to estimate added sugar values on the tend to be sweetened with sugars, and hence the group is not homogeneous; another subsequent step that can correctly take these basis of analytical data and ingredients in food products. This into consideration (e.g., step 4) should be used to estimate their added methodology was then applied to an Australian food composition 25 sugar content. table to estimate added sugar values for all foods. Step 3: Assign 100% of total sugars as added sugar for foods in the following food groups: MATERIALS AND METHODS (a) All confectionery except those containing dairy products such as fudge Definition of added sugar and chocolate. In this work, the term ‘added sugar’ was defined similarly to that used by (b) Breakfast cereals and cereal bars without fruits, chocolate, dairy or milk the USDA—that is, refined sugars added during cooking or solids. manufacturing.11 By using this definition, the following sweeteners are (c) Coffee and beverage base with no milk solids, dry or made up considered added sugars: sugar (granulated (sucrose), brown, powdered with water. and maple); monosaccharides and disaccharides (e.g., fructose, lactose, (d) Crumbed/battered meat and seafood. maltose, glucose (dextrose)); single-ingredient syrups (light corn, dark corn, (e) Processed meats. fl high-fructose corn, maple, malt, sorghum); honey and molasses; and (f) Regular soft drinks, sport drinks, avoured water and non-fruit-based maltodextrin. Despite being used as sweetening agents in some foods, energy drink. sugar alcohols were not included as added sugars in this definition, (g) Savoury biscuits, sweet biscuits, cakes and buns, donuts and batter- because they are not monosaccharides or disaccharides, and thus they are based products that do not contain fruit, chocolate or dairy products. not normally considered as ‘sugars’, such as in the Australian and New (h) Soy beverages and soy yoghurt without added fruits. Zealand Food Standards Code.26 In line with the approach of the USDA11 (i) Stock powder. and Somerset,27 undiluted fruit juice concentrate was considered as added (j) Sugar and syrups. sugar in this definition, whereas diluted fruit juice concentrates were considered to have no added sugar. This is because diluted fruit juice These food groups were selected as they contain minimal amounts of concentrates have similar composition to normal fruit juices, where the naturally occurring sugars—for example, the sugar content of plain sugar content by weight is low, making them ineffective as sweeteners. wheat flour (used in biscuits and so on) or soya beans is negligibly low 24,28 Products sweetened only with low-energy sugar substitutes (intense (o0.5 g/100 g); therefore, most, if not all, of the sugars present are sweeteners) were considered to have no added sugar. likely to be added. Step 4: Calculation based on standard recipe used in the food composition database—proportioning method where added sugar con- Proposed methodology for estimating added sugar content of tents of ALL ingredients were available from steps 1 to 3 foods Added sugar per 100 g (AS100 g) is given by the following formula: The following process outlines the methodology that we propose for – Pj estimating the added sugar content of foods, in which steps 1 6 were ´ – Wi ASi considered to be objective and steps 7 10 were considered subjective. ¼ AS ¼ i 1 Derivation of formulas used in steps 4, 5 and 6 and worked example of 100g Pj steps 4–9 are provided in Online Supplementary File 1. Wi ´ ðÞ100% þ %W Step 1: Assign 0 g added sugar to foods with 0 g total sugars. i¼1 Step 2: Assign 0 g added sugar to foods in the following food groups: where Wi is the weight of the ith ingredient in recipe, ASi is the added sugar content per 100 g of the ith ingredient and %WΔ is the percentage (a) 100% Fruit/vegetable juice and juice/cordial base sweetened with change in weight on cooking. fi arti cial sweeteners only. Step 5: Calculation based on comparison with values from the (b) All spices and herbs. unsweetened variety. (c) All fats and oils. Added sugar per 100 g (AS100g) is given by the following formula: (d) All plain cereal grains, pastas, rice and flours. (e) Eggs and egg products (except egg-based desserts). ´ ðÞ- ¼ 100 Sus Stotal (f) Fresh fruit, fresh vegetables (including salads with no dressing), fresh AS100g Sus - 100 meat, fresh seafood and tofu. fi (g) Fruits canned in 100% fruit juice or liquid sweetened with arti cial where Sus is the total sugar content per 100 g of the unsweetened variety sweeteners only. of the food and Stotal is the final listed sugar content. (h) Intensely sweetened jam and beverage base (without added sugar). Step 6: Decision based on analytical data. (i) Legumes (fresh, dried and/or processed, except sweetened varieties). If analytical data for lactose are available, and the ingredients do not (j) Mixed meat dishes with no added sugar (decided on the basis of include dried fruits or malted cereals, added sugar content is calculated as ingredient information; e.g., recipe). total sugars − lactose. If the food contains malted cereals and lactose and (k) Non-sweetened alcoholic beverages. maltose data are available, added sugar content is calculated as total (l) Non-sweetened coffees and tea. sugars − lactose − maltose. (m) Non-sugar-sweetened milk and buttermilk; breast milk. Step 7: Use borrowed values from similar products from steps 1 to 6 or (n) Non-sugar-sweetened dairy products (including yoghurts sweetened from overseas databases. with artificial sweeteners only). Values from similar product(s) within local food composition databases (o) Nuts (except sweetened varieties and nut bars), coconut (and products (e.g., AUSNUT2007) should preferably be chosen in this step. If no similar except sweetened varieties) and seeds. product is available in the local database, values from an alternative (p) Oats (and porridge) with no added sugar (decided based on ingredient database are borrowed, and the proportion of total sugars as added sugar information; e.g., ingredient list). is calculated for the borrowed food. The added sugar content of the target (q) Plain pastries without filling (such as chocolate, dried fruit and/or nuts). food is then estimated as total sugars × proportion of sugars as added (r) Plain breads (except gluten-free), English muffin, bagels, pizza bases (calculated from the borrowed food). The choice of the foreign database to and naan. borrow data from is dependent on the similarity of the food supply (s) Unsweetened dried fruits. between the countries (e.g., type of foods available), and in the current example, data from the last updated version USDA added sugar database11 These food groups were selected because they are either unprocessed were used given the similarity in the types of food available in Australia or minimally processed with no added sugar. Gluten-free breads were and the United States.

© 2015 Macmillan Publishers Limited European Journal of Clinical Nutrition (2015) 154 – 161 Method for estimating added sugar content JCY Louie et al 156 Yes Step 1. The food contain 0 g total sugars. Added sugar = 0

No Yes Step 2. Does the food belong to one of the no added sugar food groups1? No Yes Step 3. Does the food belong to one of the 100% added sugar food groups2? Added sugar = 100% total sugars Highest confidence No OBJECTIVE STEPS Step 4. Are added sugar content of all ingredients in the standard Yes Calculate added sugar based on recipe known? proportioning method3

No Yes Calculate added sugar based on Step 5. Is there a comparable unsweetened variety of the food? unsweetened variety method4 No Yes Step 6. Are analytical data of individual sugar types available? Estimate based on analytical data5

No Yes Step 7. Are there any similar products with known added sugar Borrow value from similar foods6 content (from overseas database or from steps 1 – 6)?

No Step 8. Could the added sugar content of the food be subjectively Yes Subjectively estimate the added estimated based on available information?7 sugar content No

Step 9. Are added sugar content of all ingredients in the standard Yes Calculate added sugar based on recipe known after steps 5 - 8? proportioning method3 No SUBJECTIVE STEPS Lowest confidence Step 10. 50% of the total sugars as added sugars

Figure 1. Decision algorithm. Black box indicates decision end points. 1Include 100% fruit/vegetable juice and intensely sweetened juice/ cordial base, non-sugar-sweetened milk, buttermilk, breast milk, non-sugar-sweetened dairy products (including intensely sweetened yoghurts), oats and porridge with no added sugars, fresh fruit, vegetables (including salads with no dressing), meat, seafood and tofu, fruits canned in juice or intensely sweetened liquid, dried fruits, eggs and egg products (except egg-based desserts), all spices and herbs, all oil and fats, all plain cereal grains, pastas, rice and flours, nuts (except sweetened varieties and nut bars), coconut (and products) and seeds, non-sweetened alcoholic beverages, legumes, non-sweetened coffees, mixed meat dishes with no sugary ingredients, plain bread (except gluten-free), English muffin, bagels, pizza bases and naan, plain pastry, intensely sweetened jam and beverage base. 2Include sugar and syrups, regular soft drinks, sport drinks, flavoured water and non-fruit-based energy drink, coffee and beverage base with no milk solids, dry or made up with water, breakfast cereals and cereal bars without fruits, chocolate or milk solids, processed meats, stock powder, savoury biscuits and sweet biscuits, cakes and buns, donut and batter-based products that do not contain fruits, chocolate or dairy products, all 3 confectionery except fudge, crumbed/batteredP meat and seafood, soy beverages and yoghurt. Added sugar per 100 g (AS100 g)isgivenby j W ´ AS ¼ ÀÁP i¼1 i i the following formula: AS100g j ,whereW is the weight of the ith ingredient in recipe, AS is the added sugar content per ´ ðÞ%þ% i i ¼ Wi 100 W i 1 4 100 g of the ith ingredient and %WΔ is the percentage change in weight on cooking. Added sugar per 100 g (AS100 g) is given by the following 100 ´ ðÞSusÀStotal formula: AS100g ¼ ,whereS is the amount of sugar in the unsweetened variety of the food, and S is the final listed sugar SusÀ100 us total content. 5If analytical data for lactose are available, and the ingredients do not include dried fruits or malted cereals, added sugar content was calculated as total sugars − lactose. If the food contains malted cereals and lactose data are available, added sugar content was calculated as total sugars − lactose − maltose. 6Values from foods with similar nutritional compositions and, where possible, within the same food group were borrowed. The proportion of total sugars as added sugar was calculated for the borrowed food. The added sugar value of the target food will then be estimated as total sugars × proportion of sugars as added (calculated from the borrowed food). 7Informationonthe ingredients list was used to guide the decision. Foods were deemed to have no added sugar if the ingredients listed did not contain added sugar. If the ingredients contained added sugar, the proportion of sugary ingredient, for example, the percentage of sweetened raspberry in a raspberry-flavoured muesli bar, was used to inform the estimation. If information on proportion was not available, the order of appearance of sugary ingredients and common recipes were used to inform decisions. For non-packaged foods, estimation was based on common recipes.

Step 8: Subjective estimation on the basis of ingredients and/or common Step 10: Assign 50% of total sugars as added sugar. recipes (e.g., obtained from popular recipe books). If estimation of added sugar content is impossible from steps 1 to 9, Information on the ingredients list is used to guide the decision. Foods then added sugar is assumed to be 50% of total sugars. This is because are deemed to have no added sugar if the ingredients listed do not foods with very high or very low amounts of added sugar would likely have contain added sugars. If the ingredients contain added sugars (or had their added sugar content estimated at an earlier step. ingredients with added sugars), the proportion of sugary ingredient, for A flow diagram illustrating the decision algorithm is available as example, the percentage of sweetened raspberry in a raspberry-flavoured Figure 1. muesli bar, is used to inform the estimation. If information on proportion is not available, the order of appearance of the sugary ingredients and common recipes were used to inform decisions. For non-packaged foods, Applying the method to an Australian food composition database estimation is based on common recipes. The proposed method was initially applied to the AUSNUT2007 food Step 9: Calculation based on the standard recipe that includes composition database by a dietitian (JCYL). AUSNUT2007 is a food ingredients with values assigned at steps 5–8, using the proportioning composition database compiled by Food Standards Australia and New method. Zealand for the analysis of the 2007 Australian National Children’s Nutrition Step 4 is repeated here where more foods have their added sugar and Physical Activity Survey.29 It contains complete data for 37 common contents estimated after steps 5–8. nutrients for 3874 foods, as well as predefined linkages of branded items

European Journal of Clinical Nutrition (2015) 154 – 161 © 2015 Macmillan Publishers Limited Method for estimating added sugar content JCY Louie et al 157 with generic items. AUSNUT2007 was the preferred choice of food RESULTS composition database for many Australian nutrition professionals, as it Estimation of added sugar content of foods in AUSNUT2007 contains a large range of commercially available foods and is the most database current complete Australian food composition database. Foods in Overall, 2977 foods (77% of all foods in AUSNUT2007) were AUSNUT2007 were classiﬁed into 23 broad food categories.30 Percentage change in weight owing to cooking is also provided. Nutrient data of 1233 assigned an estimated value based on objective criteria (steps – (32%) foods in AUSNUT2007 were based on data from NUTTAB2006, a food 1 6), and 897 (23%) were assigned a subjectively estimated composition database with mainly chemically analysed data. Data for the value (steps 7–10).Thecompletelistofestimatedaddedsugar remaining 2641 foods were derived using a ‘recipe approach’ based on content of all foods in AUSNUT2007, together with the steps standard recipes (n = 2153); from food labels (n = 236); calculated or that the values were based on, are available as Online imputed (n = 105); borrowed from the 1995 National Nutrition Survey Supplementary File 2. The added sugar content of a sample of database (n = 81) or similar items in overseas food composition databases commonly consumed foods estimated using our method is (n = 61); or obtained from food manufacturers (n = 5). showninTable1.

Inter-researcher agreement of choice of steps Inter-rater repeatability of the proposed methodology Results of agreement analysis between two researchers are To examine the inter-rater repeatability of the method, a second showninTable2.κ-Coefficient for consistency between 25 researcher (HM) applied the same method to the same database. The researchers for steps chosen to estimate added sugar was 0.71 added sugar content of individual foods estimated by the two researchers (Po0.001). The biggest discrepancy in the choice of steps was compared using paired sample t-tests and Pearson’s correlations. between the researchers occurred with steps 6, 8 and 10. For A Bland–Altman plot was also created to assess the level of agreement between the two sets of values. The results were further stratified, where 76% of food items (n = 2946), both researchers used the same applicable, by steps chosen by researchers to estimate added sugar steps to estimate the added sugar values. Even though the two content. As added sugar values were constant for food items in steps 1, 2, researchers used different steps to estimate added sugar 3 and 10, stratified analyses were not used. κ-Statistics were used to assess content in the remaining 24% of food items, there was no the consistency between two researchers in steps chosen, and the number significant difference between estimated values (mean differ- of foods in which the two researchers used different steps to estimate the ence − 0.15 ± 6.40 g; P = 0.46), and the Pearson’s correlation added sugar value was counted. coefficient was 0.92 (Po0.001).

Table 1. Added sugar content of selected processed foods in Australia estimated using the 10-step method

Food name Added sugar Total sugar content (g/100 g) content (g/100 g)

Bar, cherry and coconut centre, dark chocolate-coated 47.2 48.6 Bar, fudge centre, milk chocolate-coated 73.6 78.1 Bar, nougat, caramel and peanut centre, milk chocolate-coated 42.6 47.8 Biscuit, savoury cracker, puffed, flavoured 1.9 1.9 Biscuit, sweet, Anzac or butternut style 26.1 26.1 Biscuit, sweet, chocolate-coated 27.6 27.6 Bread, from white flour 0.0 3.3 Breakfast cereal, flakes of corn, unfortified 9.3 9.3 Breakfast cereal, puffed or popped rice, cocoa coating, added vitamins B1, B2, B3 and folate and Fe 44.0 44.0 Breakfast cereal, whole wheat, biscuit, added vitamins B1, B2 and B3 2.8 2.8 Cake, banana, iced, homemade 31.5 34.8 Cake, fruit, rich style, iced 42.0 46.3 Cake, lamington, with dairy cream filling 26.3 26.8 Chocolate, milk, with nuts 37.6 44.8 Cordial base, lime fruit juice 40.2 40.9 Crisp or chip, potato, flavoured 0.8 1.7 Doner kebab, lamb in flat white bread with lettuce, tomato, onion & sauce 0.7 2.4 Doughnut, dusted with cinnamon and sugar 13.7 13.7 Extruded snack, cheese-flavoured 0.4 2.5 Hamburger, beef pattie with cheese, lettuce, onion and sauce, takeaway style 1.6 3.2 Ice cream, regular fat, chocolate flavour 12.0 18.0 Mayonnaise, full fat, commercial 14.4 14.4 Milk, cow, fluid, flavoured, chocolate, regular fat 3.9 8.2 Mousse, chocolate mud, commercial 16.9 18.8 Muffin, English-style, dried fruit, toasted 8.8 10.5 Pie, meat 0.0 0.9 Pizza, ham and pineapple topping, tomato sauce, takeaway style 1.0 2.7 Sauce, barbecue, commercial 39.2 40.2 Sauce, tomato, commercial 20.7 23.3 Scone, white flour, plain 6.5 6.5 Soft drink, cola flavour 8.8 8.8 Soup, chicken noodle, prepared with water 0.0 0.6 Sugar confectionery, jelly varieties 50.6 50.6 Tart, jam 18.9 34.6 Yoghurt, extra creamy (~4.5% fat), vanilla flavoured 8.4 12.0

© 2015 Macmillan Publishers Limited European Journal of Clinical Nutrition (2015) 154 – 161 Method for estimating added sugar content JCY Louie et al 158 Inter-researcher repeatability necessary to progress research around nutrition and public health 32–38 There was no significant difference for added sugar values issues, which has so far been limited, owing to the lack of estimated between two researchers for all 3874 food items reliable data. This is especially important, as the USDA has recently (P = 0.89). Findings also showed a significant correlation for these decided to withdraw their added sugar database, citing difficulties values between two researchers (r = 0.97, Po0.001). Results of in updating the database with product formulation constantly Bland and Altman assessment31 for all items are shown in Figure 2, changing and food manufacturers being reluctant to share and Bland and Altman plots stratified by steps are available as ‘proprietary’ information with the public.11 Online Supplementary File 3. Only 73 (1.9%) of the 3874 foods had With this standardised method, there is scope to compile a a difference between the values estimated by the two researchers regularly (e.g., annually or biannually) updated added sugar outside of the limits of agreement (i.e., mean difference ± 1.96 × s.d.). 2 database for both branded and generic food items that would The R linear for the plot is 0.004, indicating that there is no allow monitoring of the levels of added sugar in different food significant systematic bias. The difference between estimated o products in the food supply across the years, as well as values between the two researchers was ± 0.5 g per 100g in longitudinal examination of the association between added sugar 83.2% of the 3874 food items in the data set. The discrepancies intake and health outcomes. Regular collection of food label data observed were resolved via discussions between the two 39 researchers, with the values presented in the Online that include ingredient data, as is the case in Australia, would Supplementary File 2 considered as final. facilitate the regular updating of the database using our proposed Findings of repeatability analyses stratified by steps in the 2946 method. Changing the food labelling law to require added sugar food items where both researchers used the same steps, are content to be disclosed on food labels, as recommended by the 40 shown in Table 3. For all steps analysed, the correlation recent review of the American food labelling regulations, will coefficients are above 0.97, indicating excellent correlation further assist in the compilation and maintenance of an added between the two sets of values. In addition, although for some sugar database. steps the paired t-test showed a statistically significant difference, the magnitude of the difference is negligible. Out of these 2946 foods, 2702 (91.7%) had a less than ± 0.5 g difference between the R² = 0.004 values estimated by the two researchers. 100 Mean difference = -0.007 80 Upper Limit of Agreement = 6.31 Lower Limit of Agreement = -6.32 DISCUSSION 60 Using this method, we were able to estimate objectively the 40 added sugar content of more than three-quarters of the 3874 20 foods in AUSNUT2007, which significantly minimises subjective – bias. Unlike previous methods,11,19 21 most of the steps in this 0 method do not require sophisticated understanding of composi- -20 tion of individual foods, with the exception of food grouping, which is usually readily available in national food composition -40

25 two researchers (g/100g) databases. This method therefore presents a simple way with -60 reasonable accuracy and good repeatability to estimate the added sugar content of foods. In addition, step 5 of this method is a new -80

approach to objectively estimate added sugar content when the Difference of added sugar values estimated by -100 total sugars content of the unsweetened variety is available, which 0 20 40 60 80 100 could signiﬁcantly improve the accuracy of the estimation. Mean of added sugar values estimated by two researchers There is currently no analytical method to distinguish between (g/100g) naturally occurring and added sugars (and it is unlikely that any Figure 2. Bland and Altman plot for difference in added sugar values feasible method will be available in the future). Our approach will between two researchers for 3874 food items. Solid line: mean allow accurate and reliable estimation of added sugar contents of difference; short strip lines: 95% limits of agreement; long strip lines: foods, especially processed and packaged foods, that are ﬁt line.

Table 2. Agreement between two researchers for steps chosen to estimate added sugar values

Steps researcher 1

Steps researcher 2 1 2 345678910

1 100.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 2 0.0% 90.0% 1.2% 9.8% 0.0% 1.0% 0.0% 8.9% 3.9% 5.3% 3 0.0% 2.8% 88.1% 2.9% 10.4% 4.8% 0.0% 19.2% 4.3% 5.3% 4 0.0% 3.7% 1.9% 51.5% 0.0% 0.0% 0.0% 0.0% 12.1% 0.0% 5 0.0% 0.4% 0.0% 0.4% 58.3% 1.0% 0.0% 3.2% 0.8% 0.0% 6 0.0% 0.1% 1.9% 0.0% 6.3% 45.2% 0.0% 5.7% 0.0% 0.0% 7 0.0% 0.8% 2.6% 0.2% 20.8% 36.5% 95.2% 36.5% 0.8% 57.9% 8 0.0% 0.7% 1.9% 0.0% 2.1% 2.9% 4.8% 16.5% 0.0% 0.0% 9 0.0% 1.5% 2.3% 35.2% 0.0% 1.9% 0.0% 0.8% 78.0% 5.3% 10 0.0% 0.1% 0.2% 0.0% 2.1% 6.7% 0.0% 9.2% 0.0% 26.3% Total n 841 1006 427 551 48 104 21 370 487 19 κ-Value = 0.71.

Table 3. Comparison and relationship between added sugar values for foods with same step between two researchers

Step n Mean ± s.d. P-valuea rP-valueb

Researcher 1 Researcher 2

1 841 0 0 N/Ac N/Ac N/Ac 2 905 0 0 N/Ac N/Ac N/Ac 3 376 11.88 ± 23.37 11.88 ± 23.37 N/Ac N/Ac N/Ac 4 284 6.31 ± 12.50 6.22 ± 12.50 0.06 0.99 o0.001 5 28 10.40 ± 13.66 11.04 ± 13.53 0.02 0.99 o0.001 6 47 18.61 ± 18.49 18.79 ± 18.20 0.31 0.99 o0.001 7 20 11.23 ± 15.13 11.70 ± 15.09 0.01 0.99 o0.001 8 61 10.57 ± 15.33 11.15 ± 15.03 0.18 0.97 o0.001 9 379 14.30 ± 13.94 14.49 ± 13.67 0.05 0.99 o0.001 10 5 8.23 ± 6.49 8.23 ± 6.49 N/Ac N/Ac N/Ac

Overall 2946 4.67 ± 12.61 4.71 ± 12.59 0.01 0.99 o0.001 aPaired t-test. bPearson's correlation, correlation is signiﬁcant at 0.01. cNot applicable as the values are constant.

It should be noted that the AUSNUT2007 database provided Limitations of the proposed method with the current paper is intended for use to analyse its As with other methods used to derive food composition data, corresponding survey data set,41 and as such it does not require except perhaps laboratory testing, the current proposed method constant updating, although practitioners may choose to use it to has a number of limitations. First, data for individual types of analyse dietary intake of individuals in their practice. They are sugars are sometimes unavailable. However, advances in analytical alerted to the limitations of the data set (as outlined in the online techniques in recent years mean that these data are now more Supplementary Table). easily obtainable. If there is no source of naturally occurring sugars (e.g., fruit/dairy) listed as an ingredient, it can be reasonably safe to assume that any sugar detected will be added. Strengths of the proposed method Second, assumptions were made around the consistency of On comparing the current proposed method with those devel- added sugar content of food groups in steps 2 and 3 (0 or 100% oped previously,11,19–21 our method appeared to be simpler and sugars as added sugar). For example, all plain breads were more accurate. In particular, our method has used only two food considered to contain no added sugar, because in most cases all group-specific steps (steps 2 and 3) that are used to assign 0 or sucrose added to bread in the baking process will be used up in 100% total sugars as added sugar, an approach that is also used in the fermentation of the dough. However, this may not be true in some cases owing to differences between manufacturers; for other methods.19,20 In addition, we have also minimised the use of example, in the United States some ‘plain’ breads have extra blanket estimation—that is, the proportion of total sugars as 22 19 sugars added. Users (especially international users) may need to added sugar. In the method developed by Roodenburg et al., choose another suitable step to estimate the added sugar content, the blanket approach was applied to 19 out of 59 steps. For or attempt to obtain the required information to allow better example, canned fruits in syrup were assigned a blanket value of estimation. In addition, in some countries the addition of refined 48% total sugars as added, compared with an average of 45% sugars to 100% fruit juices without labelling is permitted by food (range: 0–95%) in the current method. Other notable differences law; for example, up to 4 g/100 g refined sugars could be added to between the current method and the method proposed by 100% fruit juices in Australia,42 to allow standardisation of sugar Roodenburg et al.19 include the following: fruit drinks (current: content in fruit juices across seasons/years. Although technically 77%, range: 46–100% vs Roodenburg et al.:19 70%), baked beans in these sugars are ‘added sugars’, it is in our opinion that they tomato sauce (current: 80.4% vs Roodenburg et al.:19 39%) and should be excluded from the definition of ‘added sugar’ as they do fruit pie filling (current: 15.9% vs Roodenburg et al.:19 53%). not increase the overall sugars or energy content of the fruit juices Although these differences may be true owing to differences in above their natural level. On the other hand, food groups assumed product formulation, it highlights the limitation of using a blanket to have 100% sugars as added sugar (step 3) were selected on the approach that it is likely to result in overestimation and under- basis of the assumption that the only source of sugars in these estimation of added sugar content given the high variability in the foods is added. Most of the food groups listed are either proportion of total sugars as added sugar in food products. beverages with no fruit or milk content or cereal-based products with no fruit or milk content. For the latter, as the sugar content of The other strength of this method is its good repeatability. o Agreement in choosing steps and inter-rater agreement for the cereals is negligibly low, for example, 0.5 g/100 g for wheat flour, it is not unreasonable to assume that all sugars contained in values indicate high repeatability of the method and demonstrate these foods are added. the facilitation in decision-making process. Using this systematic Third, to produce reliable estimations using this method, the method, two researchers chose the same steps for the majority of food composition database should at least provide data for food items, and the difference between estimated values was lactose. Availability of data for other types of sugars may further o 4 0.5 g in 80% of cases. Other methods for estimating added improve the decision-making process. Differences in legislation 11,19 sugar did not report the repeatability of their methods or and unavailability of analytical data for individual types of sugars reported low repeatability owing to the lack of clarity in the may also render some steps (e.g., step 6) in the current method estimation process.20,21 Therefore, we propose that this systematic unusable. Furthermore, in this method, all lactose in foods were method with good repeatability can be used as a standard assumed to be naturally occurring (as a component of milk or method to estimate added sugar content of food items. other dairy products), and it was assumed there was neither loss

© 2015 Macmillan Publishers Limited European Journal of Clinical Nutrition (2015) 154 – 161 Method for estimating added sugar content JCY Louie et al 160 nor hydrolysis of sugars on cooking. Step 6 of this method endorsement programme in Australia. The remaining authors declare no conflict of therefore tends to overestimate or underestimate the added sugar interest. content, because it is based on analysed lactose content, which may be hydrolysed during cooking, especially in acidic ACKNOWLEDGEMENTS environment,43 and/or used as a sweetening agent or an excipient to a sweetening agent in some countries despite its relatively low No external funding was applied to this project. Permission has been granted from Food Standards Australia and New Zealand to reproduce the food ID, food name and sweetness compared with other sweeteners.44 total sugar content of the foods in AUSNUT2007. The authors are responsible for the Fourth, in order for step 8 to work, ideally the ingredients list accuracy of the estimated added sugar content presented in the added sugar should provide the actual proportion of characteristic ingredients, database (Online Supplementary File 2). The added sugar database is not produced 45 as is the case in Australia. Alternately, the ingredient list should by, or is in anyway associated with or endorsed by, Food Standards Australia and at least list the ingredients in descending order by weight, New Zealand. allowing some assumptions to be made (e.g., if salt and food additives were listed higher than sugar, then it could be AUTHOR CONTRIBUTIONS reasonably assumed that the proportion of added sugar was small). The estimated added sugar content of foods derived from a JCYL, SB, VMF, AMR, AWB, TPG and JCBM were involved in the conception of recipe may also be significantly different from the true analytical the study. JCYL was responsible for the early development of the methodology. values owing to variations in recipes and formulations, sampling All authors provided substantial intellectual input into the further development time frame, seasonality of fresh produce, fermentation processes and refinement of the methodology. JCYL estimated the added sugar content and other factors, but this is true for all estimated nutrients. of the food items in AUSNUT2007 using the method so developed and drafted Finally, although this method tried to minimise the subjectivity the manuscript. HM was responsible for the repeatability analysis. All authors of decision-making, some subjective decisions in choosing steps reviewed the final values in the database, were involved in the subsequent are inevitable, as shown in the repeatability analysis. Further edits of the manuscript and have read and approved the final manuscript details in the description of subjective steps may help alleviate this limitation because the most disagreement between research- REFERENCES ers applying this method was found in the choice of subjective steps (8 and 10), as identified by the extreme values in the Bland 1 Ng SW, Slining MM, Popkin BM. Use of caloric and noncaloric sweeteners – 112 and Altman assessment (Figure 2). Despite disagreement in in US consumer packaged foods, 2005 2009. J Acad Nutr Diet 2012; : 1828–1834. e1–6. choosing steps for some food items, the differences between 2 Joyce T, Gibney MJ. The impact of added sugar consumption on overall the estimated added sugar values for most of these items were dietary quality in Irish children and teenagers. J Hum Nutr Diet 2008; 21: small, although some extreme differences were identified in the 438–450. current study. In the discussion to resolve the discrepancies, it was 3 Rennie KL, Livingstone MBE. Associations between dietary added sugar intake revealed that these extreme differences arose from the misunder- and micronutrient intake: a systematic review. Br J Nutr 2007; 97:832–841. standing of the properties of a small number of foods by one of 4 Forshee RA, Storey ML. The role of added sugars in the diet quality of children and 20 – the researchers—for example, the use of dextrose and/or adolescents. J Am Coll Nutr 2001; :32 43. 5 Cobiac L, Record S, Leppard P, Syrette J, Flight I. Sugars in the Australian maltodextrin, both considered added sugars in our definition, as diet: results from the 1995 National Nutrition Survey. Nutr Diet 2003; 60: bulking agents in an intense sweetener formulation, which 152–173. resulted in a 97 g/100 g discrepancy. This was resolved easily 6 Drewnowski A. The real contribution of added sugars and fats to obesity. after clarification. We therefore propose that as a best practice two Epidemiol Rev 2007; 29:160–171. persons should conduct the estimations independently using the 7 Morenga LT, Mallard S, Mann J. Dietary sugars and body weight: systematic proposed method, at least for the subjective steps, and any review and meta-analyses of randomised controlled trials and cohort studies. BMJ anomalies between the two sets of data should be resolved via 2013; 346: e7492. discussion, as we have done here. A third researcher can act as the 8 Johnson RK, Appel LJ, Brands M, Howard BV, Lefevre M, Lustig RH et al. Dietary sugars intake and cardiovascular health: A Scientific Statement From the adjudicator should the discrepancy be unresolvable via discussion, American Heart Association. Circulation 2009; 120:1011–1020. similar to the standard approach in systematic reviews to reduce 9 Fitch C, Keim KS. Position of the Academy of Nutrition and Dietetics: 46 the risk of bias. Nevertheless, the number of food items that had use of nutritive and nonnutritive sweeteners. J Acad Nutr Diet 2012; 112: been considered for subjective steps were relatively small, as it 739–758. was possible to estimate values for the majority of foods based on 10 Faculty of Public Health of the Royal Colleges of Physicians of the United King- objective steps. dom. Sugar—a position statement, 2007. Available at: http://www.fph.org.uk/ uploads/ps_sugar.pdf (last accessed 23 January 2013). 11 United States Department of Agriculture. USDA Database for the Added Sugars CONCLUSIONS Content of Selected Foods, Release 1, 2009. Available at: http://www.ars.usda.gov/ services/docs.htm?docid = 12107 (last accessed 18 February 2013). The method presented in this paper provides an objective 12 The Norwegian Food Safety Authority, The Norwegian Directorate of Health, and estimation of added sugar content in over three-quarters of the The University of Oslo. The Norwegian Food Composition Table, 2006. Available foods contained in an Australian food composition database, and at: http://www.matportalen.no/matvaretabellen (last accessed 18 October 2011). thus subjective approaches were applied to only 23% of foods. 13 National Food Institute, Technical University of Denmark. Danish Food We propose that this method be considered as a standardised Composition Databank, 2009. Available at: http://www.foodcomp.dk/v7/fcdb_ approach in epidemiological studies for the estimation of added default.asp (last accessed 2 May 2012). 14 Nicklas TA, O'Neil CE, Liu Y. Intake of added sugars is not associated with weight sugar content of food to improve cross-study comparison. measures in children 6 to 18 years: National Health and Nutrition Examination Surveys 2003-2006. Nutr Res 2011; 31: 338–346. CONFLICT OF INTEREST 15 Sæland M, Haugen M, Eriksen FL, Wandel M, Smehaugen A, Böhmer T et al. High sugar consumption and poor nutrient intake among drug addicts in JCBM is a coauthor of The New Glucose Revolution book series (Hodder and Oslo, Norway. Br J Nutr 2011; 105:618–624. Stoughton, London; Marlowe and Co., New York; Hodder Headline, Sydney; and 16 Welsh JA, Sharma A, Cunningham SA, Vos MB. Consumption of added sugars and elsewhere), is the director of a not-for-profit GI-based food endorsement programme indicators of cardiovascular disease risk among US adolescents. Circulation 2011; in Australia and manages the University of Sydney GI testing service. AWB is a 123:249–257. coauthor of the Diabetes & Pre-diabetes Handbook, which is part of The New Glucose 17 Welsh JA, Sharma AJ, Grellinger L, Vos MB. Consumption of added sugars is Revolution book series, and is a consultant to a not-for-profit GI-based food decreasing in the United States. Am J Clin Nutr 2011; 94:726–734.

European Journal of Clinical Nutrition (2015) 154 – 161 © 2015 Macmillan Publishers Limited Method for estimating added sugar content JCY Louie et al 161 18 Welsh JA, Sharma A, Abramson JL, Vaccarino V, Gillespie C, Vos MB. Caloric 31 Bland JM, Altman DG. Statistical methods for assessing agreement between two sweetener consumption and dyslipidemia among US adults. JAMA 2010; 303: methods of clinical measurement. Lancet 1986; 1: 307–310. 1490–1497. 32 Borgen I, Aamodt G, Harsem N, Haugen M, Meltzer HM, Brantsæter AL. Maternal 19 Roodenburg AJC, Popkin BM, Seidell JC. Development of international criteria for sugar consumption and risk of preeclampsia in nulliparous Norwegian women. a front of package food labelling system: the International Choices Programme. Eur J Clin Nutr 2012; 66: 920–925. Eur J Clin Nutr 2011; 65: 1190–1200. 33 Bao Y, Stolzenberg-Solomon R, Jiao L, Silverman DT, Subar AF, Park Y et al. Added 20 Kelly SAM, Summerbell C, Rugg-Gunn AJ, Adamson A, Fletcher E, Moynihan PJ. sugar and sugar-sweetened foods and beverages and the risk of pancreatic Comparison of methods to estimate non-milk extrinsic sugars and their cancer in the National Institutes of Health-AARP Diet and Health Study. Am J Clin application to sugars in the diet of young adolescents. Br J Nutr 2005; 94: Nutr 2008; 88:431–440. 114–124. 34 Larsson SC, Bergkvist L, Wolk A. Consumption of sugar and sugar-sweetened 21 Kelly SA, Moynihan PJ, Rugg-Gunn AJ, Summerbell CD. Review of methods foods and the risk of pancreatic cancer in a prospective study. Am J Clin Nutr used to estimate non-milk extrinsic sugars. J Hum Nutr Diet 2003; 16: 2006; 84:1171–1176. 27–38; quiz 39–41. 35 Galeone C, Pelucchi C, La Vecchia C. Added sugar, glycemic index and load in 22 Dunford E, Webster J, Metzler AB, Czernichow S, Ni Mhurchu C, Wolmarans P et al. colon cancer risk. Curr Opin Clin Nutr Metab Care 2012; 15:368–373. International collaborative project to compare and monitor the nutritional 36 Wang H, Steffen LM, Zhou X, Harnack L, Luepker RV. Consistency between increasing composition of processed foods. Eur J Cardiovasc Prev Rehabil 2011; 19:1326–1332. trends in added-sugar intake and body mass index among adults: the Minnesota 23 Food Standards Australia New Zealand. Explanatory Notes for Use with NUTTAB Heart Survey, 1980–1982 to 2007–2009. Am J Public Health 2013; 103:501–507. 2010. FSANZ: Canberra, ACT, Australia, 2011. 37 Song WO, Wang Y, Chung CE, Song B, Lee W, Chun OK. Is obesity development 24 Food Standards Australia New Zealand. NUTTAB2010 Food Composition associated with dietary sugar intake in the U.S.? Nutrition 2012; 28: 1137–1141. Database, 2011. Available at: http://www.foodstandards.gov.au/science/ 38 Welsh JA, Cunningham SA. The role of added sugars in pediatric obesity. Pediatr monitoringnutrients/nutrientables/nuttab/Pages/default.aspx (last accessed 8 Clin N Am 2011; 58: 1455–1466. xi. November 2013). 39 Bupa Australia. FoodSwitch, 2013. Available at: http://www.bupa.com.au/health- 25 Food Standards Australia New Zealand. AUSNUT2007 Food Composition Data- and-wellness/tools-and-apps/mobile-app 2014s/foodswitch-app (last accessed 8 base, 2008. Available at: http://www.foodstandards.gov.au/science/monitor November 2013). ingnutrients/ausnut/pages/default.aspx (last accessed 8 November 2013). 40 US Food and Drugs Administration. Proposed Changes to the Nutrition Facts Label, 26 Food Standards Australia New Zealand. Australia New Zealand Food Standards 2014. Available at: http://www.fda.gov/Food/GuidanceRegulation/GuidanceDocu Code—Standard 1.2.8—Nutrition Information Requirements. FSANZ: Canberra, ACT, mentsRegulatoryInformation/LabelingNutrition/ucm385663.htm (last accessed 23 Australia, 2013. April 2014). 27 Somerset SM. Refined sugar intake in Australian children. Public Health Nutr 2003; 41 Sobolewski R, Cunningham J, Mackerras D. Which Australian food composition 6: 809–813. database should I use? Nutr Diet 2010; 67:37–40. 28 U.S. Department of Agriculture. USDA National Nutrient Database for Standard 42 Food Standards Australia New Zealand. Standard 2.6.1—Fruit juice and vegetable Reference version 25, 2013. Available at: http://ndb.nal.usda.gov/ (8 November juice. FSANZ: Canberra, ACT, Australia, 2008. 2013). 43 Panpae K, Jaturonrusmee W, Mingvanish W, Nuntiwattanawong C, Chunwiset S, 29 Commonwealth Scientific Industrial Research Organisation (Australia), Santudrob K et al. Minimization of sucrose losses in sugar industry by PH and Preventative Health National Research Flagship, and The University of South temperature optimization. Malay J Anal Sci 2008; 12:513–519. Australia. 2007 Australian National Children's Nutrition and Physical Activity Surveys 44 Aires CP, Tabchoury CP, Del Bel Cury AA, Cury JA. Effect of a lactose-containing —Main Findings. Department of Health and Ageing (Australia): Canberra, sweetener on root dentine demineralization in situ. Caries Res 2002; 36:167–169. Australia, 2008. 45 Food Standards Australia New Zealand. Australia New Zealand Food Standards 30 University of South Australia, Australian Commonwealth Scientific and Research Code—Standard 1.2.10—Characterising Ingredients and Components of Food. Organization, and i-View Pty Ltd. User Guide—2007 Australian National Children’s FSANZ: Canberra, ACT, Australia, 2011. Nutrition and Physical Activity Survey, 2009. Available at: http://www.health.gov. 46 Moher D, Liberati A, Tetzlaff J, Altman DGThe PG. Preferred Reporting Items for au/internet/main/publishing.nsf/Content/AC3F256C715674D5 Systematic Reviews and Meta-Analyses: The PRISMA Statement. PLoS Med 2009; 6: CA2574D60000237D/$File/user-guide-v2.pdf (last accessed1 July 2009). e1000097.

Supplementary Information accompanies this paper on European Journal of Clinical Nutrition website (http://www.nature.com/ejcn)