The school has been recognized as an important setting for health promotion, especially eating habits among children(1). In Denmark, it has been common practice for most children to bring their lunch to school from home, but more recently several initiatives with school food programmes have been introduced, one of the main objectives being to improve the dietary habits of school-aged children.
To investigate if school food programmes actually improve children's dietary intake at school, it is important to have appropriate tools for evaluating such health promotion initiatives. However, one of the challenges with regard to diet is the lack of simple and valid dietary assessment tools to monitor possible differences in the dietary quality between lunches provided as part of a school food programme and lunches brought from home.
Dietary quality indices have received increased attention and may be used as a simple and quick assessment of overall diet quality in order to evaluate adherence to dietary guidelines or guidelines for the prevention of a specific disease, as well as to monitor dietary changes(Reference Fransen and Ocke2).
A variety of dietary indices have been developed to assess overall dietary quality based on different assessment methods and data for a varying number of days. The dietary indices have mainly been proposed for adults(Reference Haines, Siega-Riz and Popkin3–Reference Biltoft-Jensen, Fagt and Groth10), but indices specifically for children have also been developed. The Preschoolers Diet–Lifestyle Index (PDL-index)(Reference Manios, Kourlaba and Grammatikaki11) and the Revised Children's Diet Quality Index (RC-DQI)(Reference Kranz, Hartman and Siega-Riz12) focus on pre-school children. The Youth Healthy Eating Index (YHEI) has been developed and used for children and adolescents(Reference Feskanich, Rockett and Colditz13).
Most indices assess the dietary quality of the total diet, whereas indices reflecting the nutrient quality of single meals, including school lunches, remain limited. A Simple Healthy Meal Index (SHMI) was developed to reflect the nutrient profile of single meals provided by canteens for adults(Reference Lassen, Biltoft-Jensen and Hansen14). Kremer et al. developed a school food checklist, with food and beverage categories, which was designed to estimate children's average energy intake from foods and beverages available in a school setting(Reference Kremer, Bell and Swinburn15). There, the focus was on the quantity, measured in the energy content of the meal, and not on the quality. However, there is a need for a tool for scientific purposes where the focus is on the dietary quality of a single meal for school-aged children. The requirements for such a tool are that it has to be simple; it must be flexible with regard to the different types of meals; and it must also be sensitive enough to measure relevant differences when children eat lunches provided by the school instead of lunches brought from home.
The aim of the present study was to develop and validate an index for assessment of dietary quality of school lunches, either brought from home or provided by the school.
Experimental methods
Study sample
Data for the validity study came from a school food programme intervention study in which eight schools from all over Denmark participated. A standardized digital photographic method(Reference Sabinsky, Toft and Andersen16) was used to collect data on the lunches brought from home by students in the second and third grades (7–10 years) and fifth and sixth grades (10–13 years). The digital images were used to assess typical lunches among schoolchildren in Denmark in the present study. A sample of 191 lunches brought from home was selected randomly out of a total of 6061, taking into consideration the school and age of the children. To validate the developed Meal IQ, it was necessary to have weighed food records. Based on digital images of the lunches brought from home, an identical double portion of the meal was produced and the weight of the lunches’ various food items was recorded. In the intervention schools, thirty-one different lunches provided by the schools were served. Recipes and product specifications for these lunches were collected. Two of each of the school meals were bought and the food items were weighed and registered in order to obtain the weighed food records. The data were collected during August–December 2008 and February–April 2009.
To increase the number of lunches provided by schools and thereby ensure greater representativeness, another thirty-two provided school meals were included in the study sample. These meals were collected in another Danish study in eighteen public schools, representative for Danish schools in terms of degree of urbanization and size (numbers of pupils)(Reference Christensen, Hansen and Sabinsky17). Weights of the food items in the lunches provided at the schools were recorded and recipes and product specifications were collected. These data were collected during November 2007–April 2008.
In total, the study sample consisted of weighed food records for 254 school lunches: 191 lunches brought from home and sixty-three lunches provided by schools.
Development of the Meal Index of dietary Quality
Overall model selection
The Meal Index of dietary Quality (Meal IQ) scoring system was developed to provide a simple measure of dietary quality of school lunches for children aged 7–13 years. The steps in the development of the Meal IQ were inspired by a nutrient profiling approach(Reference Verhagen and van den Berg18) and included: selection of variables; selection of measures for assessing the variables; definition of scoring systems and thresholds; and validation of the Meal IQ (Fig. 1).
Fig. 1 Steps in the development and validation of the Meal Index of dietary Quality (Meal IQ). Modified from Verhagen and van den Berg(Reference Verhagen and van den Berg18)
Selection of variables
The selection of variables was based on the dietary issues that are particularly relevant to the lunches and general food intake of children aged 7–13 and also on knowledge of the association between nutrients/food groups and chronic diseases.
Data from the Danish National Survey of Dietary Habits and Physical Activity revealed that to meet the official nutrition recommendations(19) and the dietary guidelines(Reference Astrup, Andersen and Stender20), Danish children should eat less fat, especially saturated fat, and less added sugar. Furthermore, children should increase their intakes of fruits, vegetables, whole grain and fish(Reference Pedersen, Groth and Christensen21–Reference Mejborn, Biltoft-Jensen and Trolle23). These considerations led to a Meal IQ consisting of seven components, which reflect the following nutrients and food groups: fat, saturated fat, sweet snacks as a proxy for added sugar, whole grain, fish, fruits and vegetables.
Measurement of variables
The Meal IQ components, i.e. fruit, vegetables and fish, were estimated in grams. To estimate total fat, saturated fat, whole grain/potatoes and snack products in the lunch meals, unit sizes were defined in terms of household measures such as slices, cups and pieces(Reference Lassen, Biltoft-Jensen and Hansen14). For validation purposes, weights in grams were assigned to each of the units; and from the weighed food records, the different number of units could be estimated.
In the development of the Meal IQ, nutrition criteria such as balance, moderation and adequacy were used to ensure the recommended macronutrient distribution within the meal. As a measure of the relative content of fat in the meal, fat units were combined with the number of starchy food units to ensure the right balance. The number of fat units was subtracted from the number of starchy food units. A fat unit was defined as 5 g of fat. This corresponds to approximately 50 g of a medium-fat product with about 10 % fat (e.g. chicken with skin, meat used for skewers); 20 g of a high-fat product (e.g. liver pâté, sausage, feta cheese: approximately 25 % fat); 10 g of a very-high-fat product (e.g. bacon, pepperoni, regular vinegar/oil salad dressing: approximately 50 % fat); or 5 g of solid fats and oils (e.g. butter, oil, mayonnaise: 80 % or more fat). Low-fat products (e.g. lean ham, cottage cheese: 5 % fat or less) do not contribute to the fat unit accounts. Furthermore, fish and plain nuts were not counted as fat units, regardless of fat content, as these foods are considered part of a healthy diet. A starchy unit corresponded to 50 g of bread, 75 g of pasta or rice, 150 g of potatoes, 300 g of vegetables, 200 g of fruits and 35 g of dried fruits, which corresponded to an energy content of about 400–500 kJ (about 25 g of starch per unit).
If the fat units were animal-based they were counted and used as an approximation of the content of saturated fat in the meal. Whole grain and potatoes were combined in the same score, since potatoes (cooked, baked or mashed) are a common accompaniment to hot meals as an alternative to rice or pasta, and it is recommended to eat potatoes several times weekly(Reference Astrup, Andersen and Stender20). The number of starchy food units, which consist of a wholegrain product or potatoes, was counted to reflect the content of healthy starch units (whole grain) in the meal. A wholegrain product was defined as containing ≥51 % DM(Reference Mejborn, Biltoft-Jensen and Trolle23) (e.g. rye bread, wholemeal pasta and brown rice). Snack products were used as a proxy for the content of added sugar. A snack product was defined as having a nutrient content beyond the following limits: fat >10 g/100 g and/or saturated fat >4 g/100 g and/or added sugar >10 g/100 g(Reference Mejborn, Biltoft-Jensen and Ygil24). The starchy units in snack products often consist of more added sugar. The starchy units in snack products were counted separately, and this assessment was relevant for the differentiation of the score for this component of the Meal IQ. The contribution of fat units and saturated fat units from the snack products was also counted, and was added to the total fat units.
To make the assessment of fat units, saturated fat units and starchy units from snack products as simple as possible, lists were made to support the process. The lists for assessment of fat units contained the most common fat-containing products, with information on the fat content per 100 g of the product and the quality of the fat; and the list of the most used snack products contained information about the contents of starch and fat and fat quality of one snack product or 100 g of the product.
Defining scoring systems and thresholds
Each of the seven components of the Meal IQ was scored from 0 (lack of compliance) to 4 (full compliance) with intermediate scores reflecting level of attainment towards dietary recommendations(19, Reference Astrup, Andersen and Stender20), but intake level in the population was also taken into consideration, especially when cut-offs for the components, which build on units, were defined. However, for snack products, the score was assessed somewhat differently. If no snack product was present in the meal, 4 points were given. If the meal contained a snack product, then the score 0 or 2 could be given, depending on the contribution of starchy units it contained. If the contribution of starchy units was ≥0·5 units, the score would be 0, but if the content of starchy units was <0·5 units then the score of 2 was given. The value 0·5 units was used to define the cut-off because it represents a relatively high contribution of starch from the snack product, about 10 % of the energy content of the meal. If most of the starch is added sugar, the content of added sugar meets the maximum level, according to the official nutrition recommendation(19). The total score for the Meal IQ ranged from 0 to 28. The construction and criteria for scoring each component are listed in Table 1.
Table 1 The Meal Index of dietary Quality (Meal IQ) scoring criteria for each of the seven components
*Wholegrain products have a content of whole grain ≥51 % of DM(Reference Fagt, Christensen and Groth22).
†If no snack product is present in the meal, 4 points is achieved. If the meal contains a snack product, then the score 0 or 2 could be given; if the contribution of starchy units is ≥0·5 units the score would be 0, but if the amount of the snack product is so limited that the contribution of starchy units is <0·5 units then the score of 2 is given. A snack product is defined as having a nutrient content beyond the following limits: fat >10 g/100 g and/or saturated fat >4 g/100 g and/or added sugar >10 g/100 g(Reference Mejborn, Biltoft-Jensen and Trolle23)
‡Limits for children up to 11 years of age.
§Limits for children from 11 years.
Validation of the Meal IQ
The Meal IQ was tested on 254 calculated meals (191 lunches brought from home and sixty-three lunches provided by schools) for its ability to assess dietary quality. The nutrient content of the meals was calculated from the weighed food records of the meals and the recipes and product specifications. The nutrient calculations were conducted using the computer program GIES (General Intake Estimation System; National Food Institute, Søborg, Denmark)(Reference Andersen, Biltoft-Jensen and Christensen25). First, the single components in the Meal IQ were tested to examine if the components correlated with the nutrient concerned. Then, the Meal IQ score was estimated from the weighed food record of the lunches and validated against the calculated nutrient content of these meals.
Statistical analysis
To investigate if the selected components in the Meal IQ reflected the nutrients of concern, correlation coefficients between the estimated components and the objective measures were assessed. Because the data on dietary intake were not normally distributed, Spearman's correlation coefficient was used(Reference Brown and Benedetti26). The estimated components were classified into quartiles. Gross misclassification was defined as classification in the opposite quartile observed in the highest and lowest quartile. Correlations between the Meal IQ score and the calculations of the nutrient content were assessed. The sample was divided into four categories according to the total Meal IQ score; and mean values of energy and nutrient content of the meals were compared by ANOVA, after testing for equality of variances, or using the Kruskal–Wallis test(Reference Altman27, Reference Kruskal and Wallis28). Bonferroni correction was used to account for increase in type I error due to multiple comparisons(Reference Abdi29). Linear trends across the categories were tested by modelling the score as a continuous variable in the model and testing for model reduction(Reference Altman27).
All reported P values were based on two-sided hypotheses and compared with a significance level of 5 %. Statistical analyses were carried out using the SAS statistical software package version 9·2 (SAS Institute Inc., Cary, NC, USA).
Results
Correlation between the Meal IQ components and calculated nutrient content in the school lunches
Each of the components not measured in grams (total fat, saturated fat, whole grain/potatoes and snack products) was estimated from the weighed food records and validated against the relevant calculated nutrient content (Table 2). The component estimating fat content was highly correlated to the percentage of energy from fat (r = −0·77). Numbers of saturated fat units were correlated to the percentage energy from saturated fat (r = 0·76). The number of whole grain units was correlated to dietary fibre (r = 0·56); and the snack component was correlated to added sugar in the meals (r = 0·57).
Table 2 Correlations between the calculated nutrient content of the meal components and the nutrient components of the Meal Index of dietary Quality (Meal IQ; n 254)
%E, percentage of food energy.
The proportion of meals which were classified into the same or adjacent quartiles from the measured component and the nutrient of concern ranged from 91 % (whole grain units v. fibre) to 98 % (starch units in snack product v. added sugar). Gross misclassification was found in one of the 254 cases for the components measuring added sugar, saturated fat and total fat; and for the whole grain score correlated with fibre, misclassification was found in 2 % of the meals.
Correlation between the Meal IQ score and nutrient content
The results presented in Table 3 are based on comparisons of the assessed Meal IQ score from the weighed food records of the lunches brought from home or provided by the school. A higher Meal IQ score was significantly associated with lower intakes of total and saturated fat and sugar and with higher intakes of fibre, fish, fruits, vegetables and various vitamins and minerals. The Spearman's correlation coefficients between the score for Meal IQ and the nutrient content of the nutrient-calculated meals varied between dietary components and were highest for energy density (r = −0·61) and lowest for energy (r = 0·04) where no association was measured. The Spearman's correlation coefficients were calculated separately for the lunches brought from home and the lunches provided by the schools in order to examine if the results were similar. The separate analyses showed the same tendencies as the combined analysis of the lunches except for the association between the Meal IQ score and the energy density in the lunches provided by the school (results not shown). The correlation was not as strong (r = −0·13) as seen in the lunches brought from home (r = −0·73).
Table 3 Correlations between energy and nutrient content of the meals and the total score of the Meal Index of dietary Quality (Meal IQ; n 254)
%E, percentage of food energy; RE, retinol equivalents.
*Spearman's correlation coefficient analyses are made using the 28-classed score.
Table 4 illustrates the P values for trend across the Meal IQ categories and the P values for differences between the categories. The linear trend was highly significant for the percentage of energy from fat and saturated fat, fibre, vitamin E, vitamin K, vitamin B6, folic acid, vitamin C, fruits and vegetables. There was also a trend for vitamin A (P = 0·0274). The result of the analysis for trend showed no significance across the Meal IQ score for energy, energy density, percentage of energy from carbohydrate and added sugar, Ca, Fe and fish. The P values for the ANOVA showed a significant difference between the score categories for all the nutrient and food groups except energy and vitamin A, and after Bonferroni correction, there was no difference between the categories for Ca either.
Table 4 Energy and nutrient contents of the meals by categories according to the total score of the Meal Index of dietary Quality (Meal IQ; n 254)
%E, percentage of food energy; RE, retinol equivalents.
a,b,c,dMean values within a row with unlike superscript letters were significantly different after Bonferroni correction for multiple comparisons (P < 0·05).
*Excluding potatoes.
Discussion
The results obtained in the present study indicate that the Meal IQ is a valid and useful tool for providing information on the dietary quality of school lunches brought from home or provided by the school, and thus a useful evaluation tool for school food programmes. The Meal IQ score is a good measure of dietary quality, as higher values of the score are strongly associated not only with the nutrient and food groups included in the index, but also with selected nutrients.
We found a linear trend across the Meal IQ categories for percentage of energy from fat and saturated fat, fibre, vitamin A, vitamin E, vitamin K, vitamin B6, folic acid, vitamin C, fruits and vegetables. But we did not find a trend for all nutrient and food groups. The missing trend for sugar and fish might be explained by the fact that a high content of added sugar was found only in the meals with low scores, and fish was mainly present in the lunches with high Meal IQ scores. This was also the result of the ANOVA after Bonferroni correction for multiple comparisons, where a significant difference was found between the content of added sugar in the category for meals with the lowest Meal IQ score and the other three categories. The same picture was seen when analysing the content of fish in the four categories, where the meals with the highest Meal IQ score had a significantly higher amount of fish compared with the other three categories. This could also explain the absent trend for vitamin D, as this micronutrient is highly present in fish. The content of Ca in the meals did not show either a trend or any differences between the categories. The reason for this might be that all types of meals contain cheese, for example, and the one scoring lowest also contains spread and cheese snacks which all contribute to the Ca content; and the meals with high scores also contain Ca from vegetables. The lack of trend for Fe is due to the content of Fe in meat, which is present in most of the lunches, but the amount of fat in the meat may vary. Fortification could have had an influence on the results of trends across the categories for micronutrients, but in Denmark fortification is not common. The non-significant trend across categories for energy density was unexpected, but it may be due to a wide variation in energy content. The ANOVA showed a highly significant P value (P < 0·0001) for the general difference among the categories for energy density. Pair-wise differences between the categories after Bonferroni correction showed a significant difference between all the groups (P < 0·05) except for the last two. Lassen et al.(Reference Lassen, Biltoft-Jensen and Hansen14) did not do a trend analyses when analysing the energy density across categories for the score of the SHMI, but they found a significant difference between categories in the total score of the SHMI for energy density using ANOVA; after Bonferroni correction, however, there was no difference between the two middle categories.
Because nutrient intake is positively correlated with energy intake, a diet quality index could overrate high-energy diets, especially if nutrient adequacy is weighted more heavily than moderation(Reference Guenther, Reedy and Krebs-Smith30). Several indices show an association between the score and total energy intake(Reference Kennedy, Ohls and Carlson4, Reference Toft, Kristoffersen and Lau8, Reference Biltoft-Jensen, Fagt and Groth10). In the Meal IQ, both nutrient adequacy and moderation are represented among the chosen components. We did not find any correlation between the Meal IQ score and total energy intake. This should be noted as an advantage, because the Meal IQ can assess diet quality independently of diet quantity. Another methodological issue concerning dietary indices is how to combine the different components into one measure(Reference Fransen and Ocke2). Often the components incorporated in the indices are considered equally important. In the Meal IQ, we attempted to focus more on fruits and vegetables and fat by dividing fruits and vegetables into two components, and also having separate components for total fat and saturated fat. Another issue is the scoring of each component (binary, proportional or other)(Reference Fransen and Ocke2, Reference Waijers, Feskens and Ocke31). In the Meal IQ, we have mainly used a proportional approach on the assumption that the difference from 0 to 1 is the same as from 1 to 2. This may not be completely true. If total fat goes from 25 to 20 % of energy, the effect is not the same as from 40 to 35 % of energy. Future study could do further work on differentiations in the single component scores in the Meal IQ, and also on different strategies for scoring of the variables. For instance, the correlation with regard to total fat may be better described by a U-shaped relationship than by the proportional approach.
Not all aspects of nutritional recommendations are implemented in the Meal IQ. Beverages were not included as a component in the Meal IQ, because the relationship between energy density and macronutrient content in beverages is more complex than in individual foods or meals(Reference Crowe, LaFontaine and Gibbons32). Besides the focus was on developing a tool for assessing differences between lunches brought from home and lunches provided by the schools, and since beverages are not part of the school food programme in Denmark, this component was not included. Beverages may contribute significantly to total energy intake, but data from the Danish National Survey of Dietary Habits and Physical Activity show that more than half of school-aged children (7–13 years) drink water or low-fat milk for lunch at school(Reference Fagt, Christensen and Groth22). Incorporating beverages as a component in the Meal IQ could be relevant in future studies. In addition, the index does not deal with salt content, which should be developed and tested for future extension of the Meal IQ. Further research is needed to determine the dietary elements that are most related to health among children/youth.
The Meal IQ focuses on the overall dietary quality of a meal. The official recommendations are valid for the average intake for a longer period, at least a week, since the dietary composition may vary from meal to meal and from day to day(19). It is therefore a challenge to establish dietary guidelines for a single meal, but this was done by defining cut-off points for the Meal IQ components based on the official recommendations and dietary guidelines. When defining the official recommendations, the current food consumption patterns were taken into account. This was also done when defining the Meal IQ scoring system for whole grain: if the starting point was that children should receive 25 % of their daily energy needs from the lunch they eat at school, then the upper cut-off for whole grain should be about 25 % of the recommended intake. But for whole grain, the limit was set above 25 % of the recommended daily intake because Danish children eat rye bread at lunch(Reference Mejborn, Biltoft-Jensen and Trolle23), which is an important source of whole grain; therefore, it is likely that the lunch would provide more whole grain. From this point of view, it would be appropriate to develop specific indices for the different types of meals and for specific target groups.
A limitation of using the Meal IQ is the need for recipes and product specifications, especially when assessing the Meal IQ components for hot meals, for which it can be difficult to assess the content of e.g. fat and the amounts of vegetables. It is relatively easy to get information on the lunches provided by the schools, because recipes are available and the meals are often standardized.
One of the advantages of the Meal IQ is that the score is easily obtained through a simple evaluation process. The seven components incorporated in the Meal IQ can be determined from a weighed food record. There is no need for calculations of the nutrient content, which would make the calculation of the total score more complex and labour-intensive as in the Healthy Eating Index (HEI)(Reference Kennedy, Ohls and Carlson4) and the RC-DQI(Reference Kranz, Hartman and Siega-Riz12), among others.
In conclusion, the new Meal IQ is an easily applied evaluation tool for assessing the dietary quality of lunches provided by schools or brought from home. The method is valid, simple, flexible and sensitive. The Meal IQ is a tool that can be used by health professionals at various levels to evaluate health promotion interventions in schools.
Acknowledgements
The project is part of the EVIUS study funded by Danish Food Industry Agency, DTU National Food Institute and Department of Nutrition, National Food Institute. The authors declare no conflict of interest. All the authors contributed to the study concept and design, interpretation of data and preparation of the manuscript. The authors thank all participants of this study. They also thank Bent Egberg Mikkelsen, Food, People & Design, Department of Development and Planning, Aalborg University and Anne Dahl Lassen, DTU National Food Institute for advice and manuscript comments.
