Modified HSR
We applied the original HSR algorithm to items in the AUSNUT 2011-13 database using Microsoft Excel (Microsoft 365 MSO, Version 16.0.13426.20306) informed by the HSR guide [
20]. In brief, to calculate a star rating, foods/beverages were categorised into one of six categories: (1) Non-dairy beverages, 1D. Milk and Dairy beverages (and alternatives), (2) Foods, 2D. Dairy foods (and alternatives), (3) Oils and Spreads, 3D. Cheese. Category 2 foods are the only foods which contain whole grain and therefore only these foods were modified in this modelling study. HSR ‘baseline points’ were calculated using components linked with increased risk of non-communicable diseases. This includes energy (kJ) and total sugars (g) content (for Category 1), as well as saturated fat (g) and sodium (mg) content (for the remaining categories), per 100 g/ml of the food item.
‘Modifying points’ were then allocated in relation to the percentage of fruits, vegetables, nuts and legumes (FVNL) of a product, referred to as ‘HSR V points’. These points are calculated based on the concentrated and/or non-concentrated FVNL content of a food. There are no clear criteria of what ingredients can contribute to the concentrated or non-concentrated FVNL of a food. To ensure consistency, we used the HSR guide [
20] and the “Example of potential additional guidance on eligibility for FVNL and concentrated FVNL” from the HSR Technical Advisory Group [
21], when calculating the “V points” of food items. The Australian Healthy Survey – Australian Dietary Guidelines (AHS-ADG) database [
22] and the AUSNUT 2011-13 food recipe file [
23] were used to determine the concentrated and non-concentrated FVNL content of each food and the equation provided in the HSR guide [
20] was used to calculate the FVNL percentage foods containing both concentrated and non-concentrated FVNL. The AHS-ADG is based on estimations, therefore, some items received greater than 100% FVNL content. These products were assigned the maximum V points.
Categories 1D-3D may also receive additional modifying points for protein, ‘HSR P points’ or dietary fibre, ‘HSR F points’ content, which were calculated. Point allocation for each food is dependent on its categorisation. Additional considerations outlined in the HSR guide were also applied, for example, not assigning ‘P points’ to items with greater than 12 baseline points, unless they obtained five of more ‘V points’. These modifying points were then subtracted from the baseline points to produce a HSR score, which is coherent to a HSR rating [
20].
Cut-offs are then applied to the total score to determine the number of stars a product is awarded. Here, the original HSR algorithm was modified by including and assigning modifying points for the whole-grain content of food items, ‘WG points’. According to the WGI, foods with ≥ 25% whole-grain ingredients are eligible for stating the presence of whole grains on FOPL, and a food can only be labelled as a ‘whole-grain food’ if it is composed of a ≥ 50% whole-grain ingredients, based on dry weight [
24]. These percentage values were used to determine the WG points an item could score based on their whole-grain content. The method of allocating WG points was modelled on previously trialled methods [
17]. Foods containing 25–100% whole grain scored up to 10 WG points, with higher percentages equating to more points (Table
1). The 25% whole-grain cut-off approximately correlates to 8 g whole grain per serve, which was the minimum amount used in early studies to show the relationship between whole-grain intake and positive health outcomes [
25]. No points were applied for foods below 25%, as this is widely considered the minimal acceptable amount to make an impact on whole-grain intake and the higher cut-off may encourage manufacturers to produce items with a greater whole-grain content [
24].
Table 1
Whole-grain cut-off points to create modified HSR algorithm
Whole-grain percentage (dry weight)a | < 25 | | ≥ 25 | | | | ≥ 50 | | | | 100 |
The whole-grain content of items was determined, using the whole-grain values from the Expanded Australian Whole-grain Database. WG points were applied to the HSR score equation in the same way as the other modifying points. Category 2 foods in the HSR system includes all foods other than beverages, dairy and oils, therefore the whole-grain points were applied only to grain foods in Category 2, which in effect, was only grain foods which contained ≥ 25% whole-grain ingredients. The following formula was used to determine HSR score.
Equation 1
HSR score = HSR baseline points – (V pointsa+ P pointsb+ F pointsc+ WG pointsd).
Equation 1: Modified HSR equation to determine the HSR score of an item.
aPoints for fruits, vegetables, nuts and legumes content.
bPoints for protein content.
cPoints for fibre content.
dPoints for whole-grain content.
Initial models indicated that modifying only the points assigned to foods, but maintaining the same cut-offs for scores to award the stars tended to poorly differentiate between items with differing amounts of whole-grain content and nutrition composition. Therefore, to account for the addition of ‘WG points’, different ranges to assign HSR scores to a rating were trialled (Table
2). We compared shifting cut off ranges by 3 points or 10 points (maximum as we had added up to 10 whole grain points) to consider methods to maximise the differentiation between refined and whole-grain foods. Similar to the application of modified points, in this comparison food items were considered ‘whole-grain’ items if they contained ≥ 25% whole-grain ingredients and the rest as ‘refined grain’ items.
Table 2
Health Star Rating (HSR) score cut-offs and corresponding final HSR for Category 2 gra in foods
5.0 | ≤-11 | ≤-14 | ≤-21 |
4.5 | -10–7 | -13–10 | -20–17 |
4.0 | -6–2 | -9–5 | -16–12 |
3.5 | -1–2 | -4–1 | -11–8 |
3.0 | 3–6 | 0–3 | -7–4 |
2.5 | 7–11 | 4–8 | -3–1 |
2.0 | 12–15 | 9–12 | 2–5 |
1.5 | 16–20 | 13–17 | 6–10 |
1.0 | 21–24 | 18–21 | 11–14 |
0.5 | ≥ 25 | ≥ 22 | ≥ 15 |
To ensure non-grain containing items were not unnecessarily shifted down in star ratings the algorithm was only applied to major grain containing groups, including “cereals and cereal products” (Group 12 in the AUSNUT database), “cereal based products and dishes” (Group 13) and “confectionary and cereal/nut/fruit/seed bars” (Group 28). Group 28 consists of items with and without grains (whole and refined). Non-grain containing items were identified and excluded using the AUSNUT 2011-13 food recipe file [
23] and professional judgement. For example, if flour was in an item’s ingredient list on the recipe file, it was considered a grain containing item and was included. Each cut-off test was applied to each grain containing food and analysed within major food groups (those in the database represented on a 2-digit level – e.g. cereals and cereals products is Group 12), as specified in the AUSNUT database. Frequency tables were created on Microsoft Excel (Microsoft 365 MSO, Version 16) to compare the shift in star rating within each major group when applying each modification.
To further examine the shift in HSR, between whole-grain and refined grain foods, we also considered foods at a sub-major group (3 -digit) level. “Regular breads, and bread rolls” (Group 122), “English-style muffins, flat breads, and savoury and sweet breads” (Group 123), “Breakfast cereals, ready to eat” (Group 125) and “Breakfast cereals, hot porridge style” (Group 126), contain items that contribute the most to whole-grain intake in Australia [
9], therefore, these groups were combined based on similarities (i.e. 122 and 123, 125 and 126) and analysed in the same way as the 2-digit groups. Hereafter the grouping “Regular breads, and bread rolls” (Group 122) and “English-style muffins, flat breads, and savoury and sweet breads” (Group 123) is referred to as “Bread items” and the grouping of “Breakfast cereals, ready to eat” (Group 125) and “Breakfast cereals, hot porridge style” (Group 126) as “Breakfast cereals”.
Statistical analysis
Statistical analysis was conducted on IBM SPSS Statistics Version 28. Data was checked for normality using Kolmogorov-Smirnov and descriptive statistics were used to determine the median, interquartile range (IQR) (for non-normally distributed data) and range of whole-grain content (per 100 g) and HSR for whole-grain and refined grain items for each major and sub-major group specified above. Independent-samples median tests were conducted to compare differences in HSR between refined and whole-grain items in each group. Two commonly consumed items (one refined and one whole-grain) from each major group were selected as examples to compare and convey the changes in HSR when applying the different algorithms. This included a refined and whole-grain version of a bread roll (Group 12), a savoury biscuit (Group 13) and a muesli bar (Group 28).
Spearman’s correlation was then used to measure the strength of the association between components of an item’s nutrient composition (whole grain, fibre, protein, fruit/vegetable/nuts/legumes, energy, saturated fat, total sugar and sodium) and the item’s HSR using each algorithm. A p-value of < 0.05 was used to determine statistical significance of the component contributing to overall HSR.