Validity of the Australian Recommended Food Score as a diet quality index for Pre-schoolers

Dietary intake was assessed using the Australian Eating Survey Pre-schooler Version (AES-P). Given the age of the pre-schoolers, a caregiver (i.e. parent or guardian) recorded the child’s frequency of consumption of a comprehensive 120-item semi-quantitative FFQ. Further details of the development of the AES FFQ have been published elsewhere [4, 14], demonstrating acceptable accuracy for ranking nutrient intakes in children and adolescents 9–16 years of age [14]. The AES-P was previously shown to be a valid estimate of total energy expenditure in children 3 years of age [4], and children 8–11 years of age using doubly labeled water [17].

Caregivers recorded their child’s frequency of consumption of a comprehensive, defined list of foods over the previous 6 months. The frequency options ranged from ‘never’ to ‘4 or more times per day’, to ‘7 or more glasses per day’ for beverages. Questions from the FFQ were grouped categorically into food groups and subgroups including those energy dense nutrient poor foods now referred to as ‘ discretionary choices’ [18]. Due to the seasonal availability of some fruits, a separate section was included in the FFQ for seasonal fruit. The frequency categories were listed as for other food items, with the question, “when the following fruit is in season, how often do you usually eat it?” to capture the usual consumption of the fruit when it is in season. Seasonal availability was determined by contacting the food markets, Sydney, NSW and obtaining information about the wider availability in other markets and supermarkets during the year, in addition to referring to supermarket literature that indicated the months of the year different seasonal fruit was available.

Pre-school age portion sizes in grams for individual foods items, was derived from the 2007 National Children’s Nutrition and Physical Activity Survey, purchased from the Australian Social Science Data Archive at the Australian National University [19]. Nutrient intakes from the AES-P FFQ were computed from the most current food composition database of Australian foods available, the Australian AusNut 2007 database (All Foods) Revision 17 and AusFoods (Brands) Revision 5 [20] to generate individual mean daily macro-and micronutrient intakes. The estimated daily intakes for 20 macro- and micronutrients were calculated using FoodWorks [21] and compared to age-specific nutrient targets. Adequacy of nutrient intake was assessed using Recommended Daily Intake (RDI) and Adequate Intake (AI) targets [22], while intakes exceeding recommendations were defined by Upper Limits (UL) [22] and the 2013 Australian Dietary Guidelines [18].

The ARFS-P was designed as a brief, culture specific, food based diet quality tool for young children, focusing on dietary variety within recommended food groups. It was modeled on the Recommended Food Score [23] and the Australian Recommended Food Score (ARFS) [24], and utilizes a subsample of questions from the AES-P FFQ consistent with the 2013 Australian Dietary Guidelines [18]. The ARFS-P includes seventy questions relating to eight food group components; vegetables (n = 20), fruit (n = 12), meat (n = 7), meat alternatives (i.e. non-meat protein) (n = 6), breads/cereals (n = 12), dairy (n = 10), water (n = 1) and condiments (n = 2). The procedure used to calculate ARFS has been published elsewhere [13]. Briefly, points were awarded for foods consumed according to frequency of consumption, with healthy foods receiving more points (i.e. vegetables, low-fat dairy, whole grains), and point caps on foods that should not be consumed overly often (e.g. meat, whole milk). An ARFS component was only calculated for those responses with not more than one question missing; one child was excluded for this reason. The ARFS total possible score ranges from zero to 73, and the component scores from zero to the number of questions in that component, plus one more possible for vegetables, low-fat dairy, whole grains.

Scores were described by medians and interquartile range (IQR) and Fisher’s exact test and Wilcoxon rank-sum test were used to compare two age groups in univariate analyses. Groups were younger children (<4 years of age) and older children (≥4 years of age), so that the same RDI, AI and UL values would apply to all participants in a group. Linear regressions were used to estimate the beta-coefficients (which can be thought of as correlations) of ARFS-P scores with FFQ components, whilst controlling for demographic factors. More precisely, agreement was assessed using linear regression models with AES-P FFQ food group, macro and micronutrient intakes as response variables and ARFS-P components as explanatory variables. Demographic variables, age, gender and parent education, were controlled for by being kept in the model if significant (too few participants were Aboriginal or Torres Strait Islander to be able to estimate the effect of race). The total energy intake was included as an explanatory variable if significant. By this we mean that backwards stepwise with p = 0.05 for removal was employed to remove variables that added no predictive value to the model. This was followed by adding any variable back in that was significant at the 0.05 level, so that that smallest model with the best predictive ability was identified. Both ARFS-P and FFQ were standardized (subtract mean and divide by standard deviation) so the model coefficient was the beta-coefficient. A square root transform was applied to some variables to improve the normality of the residuals. Kolmogorov-Smirnov tests and normal probability plots were used to assess normality of residuals. Square root transform of response variables was used if necessary to improve residual normality. Statistical significance was at the 5% level. All data manipulation and statistical analysis was undertaken using Stata MP v12 [25].

Table 4 demonstrates that total ARFS-P and the ARFS-P components were significantly related to AES-P FFQ intake of the majority of macro and micronutrients. Total ARFS-P was positively related to protein (ρ = 22% CI_95% 14-30), cholesterol (ρ = 27% CI_95% 15-40), fibre (ρ = 12% CI_95% 3-22), vitamin A (ρ = 40% CI_95% 26-54), β-carotine (ρ = 58% CI_95% 44-71), niacine equivalents (ρ = 20% CI_95% 10-30), folate (ρ = 14% CI_95% 2-26), vitamin C (ρ = 44% CI_95% 31-58), calcium (ρ = 16% CI_95% 4-29), magnesium (ρ = 16% CI_95% 9-24), phosphorus (ρ = 13% CI_95% 6-21), potassium (ρ = 27% CI_95% 19-35) and zinc intake (ρ = 17% CI_95% 10-25), while negatively related to carbohydrate (ρ = -10% CI_95% -16- -5).

Table 4 also demonstrates that total ARFS-P was significantly related to eating patterns. It was positively related to total consumption of nutrient-dense foods, as well as food sub-groups including vegetables, fruit, meat and meat alternatives. In contrast, total ARFS-P was negatively related to total consumption of discretionary choices, and subgroups such as confectionary, sugar sweetened drinks, packaged snacks and take-away foods. Figure 1 displays these beta-coefficients between total ARFS-P and AES-P FFQ reported intake, together with their 95% confidence intervals.

Anthropometric data, such as weight, were not obtained from this population, since the primary outcome measure of the FHFK randomized controlled trial was changes to dietary intake. As a consequence, the relationship between weight status and ARFS-P could not be explored. However, unpublished data collected in 2008 as part of the Before School Screening program for 4–5 year olds (n = 571) in the same study locations suggested that rates of overweight and obesity were higher than the NSW average [29], with 26.8% of children overweight or obese (27.3% male, 26.3% female). Despite a previous study suggesting that the FFQ was relatively accurate when assessing total energy intake at the group level [4], over –and under-reporting of dietary intake and physical activity level were not assessed in the current study. The relatively small sample from rural locations reduces the generalizability and hence results should be interpreted with caution.

The effect of parental bias must also be considered when interpreting results as a caregiver completed the AES-P on their child’s behalf. This may exacerbate the over reporting bias already associated with the use of an FFQ [30] and may increase the chance that the data is incomplete or that intake has been underestimated [31]. Future studies should consider obtaining dietary intake reports from both parents, or where appropriate, with the child present to try to reduce reporting biases [32, 33].

Some limitations of calculating and using ARFS to determine usual diet quality have been previously published [13], including that the national nutrient databases used in this study does not accurately reflect population consumption of folate as fortification of breads, cereals and cereal products is not accounted for within the database, hence results obtained for FFQ folate intake are likely to be lower than true intake.

The ARFS-P allows for the rapid measurement of diet quality of children aged 2–5 years in a single continuous variable. It was derived from a validated FFQ for young children, and provides researchers the opportunity to use an independent measure of intake, or calculated secondarily, from the AES-P FFQ. Its calculation is less onerous than methods that require derivation of nutrient based sub-scales. This means that the provision of feedback based on overall score can potentially be in real-time.

Based on the finding in the current study, the ARFS-P method for scoring diet quality could be applied to other FFQs, in other populations, age groups, and in other settings. This could include testing its use as a self-monitoring tool or within clinical practice. If validated in these settings it would be a convenient tool that could be used by a variety of health professionals to assess and monitor dietary patterns.