This study is the first to investigate if a standardised DPM is valid and reliable for assessment of selected dietary components and the overall dietary quality of school lunch sandwiches brought from home.
The analysis of the difference between the amount of fruits and vegetables estimated from the digital images shows a difference from the weighed foods, despite almost the same medians and averages of these variables. The Bland–Altman analyses show acceptable limits of agreement for fruits (−29.4 and 20.8 g) and vegetables (−34.5 and 22.2 g), with some variability but on the same level as found by others (17
). The smaller sample of the analyses for fruits (n
=67) and vegetables (n
=130) affects the variability and the limits of agreement. The Bland–Altman plots illustrate a tendency of increasing underestimation with increasing intake when using the DPM; however, both correlation coefficients were high (r
=0.96 for both variables), and the cross-classifications illustrate that the ranking of the individual meals was good for both fruits and vegetables (100% was classified in the same or adjacent quartile).
When estimating the defined units of fat, starch, and whole grains from the digital images, no statistical difference from the weighed foods was shown. It is easier to estimate variables in household measures, because they do not require the same degree of accuracy as the variables assessed in grams. But for fish, no difference between the estimated amount from the digital images and the true weight from the food record was found, probably because it is easier to estimate the relatively small amounts of fish compared to the voluminous and especially large quantities of fruits and/or vegetables. The Bland–Altman analysis for fish shows tight limits of agreement (−14.7 to 10.0 g), but also for this food item, the Bland–Altman plots illustrate a tendency towards larger variability of the range of intake. This result must be treated with caution, since the sample for the fish analyses is relatively small (n=21). We found a difference in the saturated fat units between the methods, probably because of wrong assessment of the spread used on the bread, since it can be difficult to assess whether it is butter or, for example, margarine.
The Meal IQ consists of both the variables estimated in grams and components assessed in units. Compared to the results from the weighed food record method, the DPM was found to provide a good assessment of the overall dietary quality assessed by the Meal IQ. No difference was found between the Meal IQ score assessed using the two methods (P
=0.3394). The Bland–Altman plot shows a small bias (0.07), and the limits of agreement are sufficiently tight to suggest good agreement between the methods (−2.26 to 2.40). The Meal IQ is not influenced by the underestimation of fruits and vegetables with increasing intake. Fruits and vegetables are separate components in the Meal IQ, and each component in the Meal IQ scores from 0 to 4 points. If fruits or vegetables are not represented in the meal, the score is 0; and if the meal contains 75 g or more, 4 points are given (15
). Further analyses show that the problem of underestimating fruits and vegetables does not exist when estimating weights under 85 g of vegetables and 115 g of fruits.
The correlation coefficients between the dietary components and the Meal IQ assessed from either the DPM or the weighed food record were high (r
=0.89–0.97). Correlation analyses are often used to validate dietary assessment methods, but correlation coefficients provide only a limited measure of the level of agreement between two methods and should therefore not be used alone. Correlation coefficients depend, for example, on the range of the true quantity in the sample (20
). In this study, the correlation coefficients were supplemented with cross-classification of the individual meals. This was also good for the dietary components as well as the Meal IQ. In addition, the Bland–Altman plots used for assessment showed acceptable limits of agreement.
In this study, the interrater reliability was assessed from kappa statistics. The kappa coefficient shows a moderate strength of agreement for the assessment of starchy units by the two raters (κ
=0.59), very good agreement in estimating the amount of fruits (κ
=0.82), and good agreement for the other components (κ
=0.69–0.80) and the Meal IQ (κ
). Other studies have evaluated the interrater reliability by calculating intraclass correlation, and they also found a good interrater reliability, with intraclass correlations on the level of 0.80–0.96 for different parameters when using the DPM (12
The validity and reliability of the method are highly dependent on the skills of the image analysts. To reduce the variability caused by using many raters, intensive training of one or possibly two raters might be preferable to training many raters. Also, future training procedures of image analysts should focus on the underestimation we found, especially for the high amount for fruits and vegetables. Others have also reported underestimation when using the DPM (17
An advantage of the DPM is the opportunity to collect dietary intake data from large populations (9
) (e.g. in intervention studies where dietary data have to be collected and where data on meals should be evaluated). Another advantage is that the burden on the participants is minimal compared to that of other dietary assessment methods, and the method also overcomes the recall problems of children. The visual estimation technique is the most comparable method to the DPM. This method is also shown to be valid and reliable (12
) and would overcome some of the same challenges as the DPM. But the advantages of using the DPM instead of the visual estimation technique are rapid collection of the dietary data in the eating environment, convenience for participants and researchers, and the possibility of uninterrupted evaluations of the foods that are studied on the digital images, as opposed to evaluation in the setting for data collection (12
The most time-consuming step when using the DPM for dietary assessment is the nutritional evaluation, due to reliance on human analysts to estimate food intake and possibly subsequent calculations of the nutrient content. To make the method as cost-effective as possible, we used the Meal IQ in addition to the individual dietary components to assess the dietary quality of the lunches. The Meal IQ score is obtained easily through a simple evaluation process. There is no need to calculate the nutrient content, which would make the calculation of the total score more complex and labour intensive.
It is challenging to assess digital images of school lunch sandwiches brought from home rather than school lunches provided by the school, because recipes and product specifications are not available. But we believe that the method is appropriate for this type of meal as well, because the school lunch sandwiches brought from home normally consist of bread, spread, sliced cold meat, and a piece of fruits or some vegetables, often in relatively standardised portions. A limitation of the DPM may be the dependence of visibility of the food or nutrient of concern. The digital images do not always show details about particular foods (e.g. fat-reduced products). In this study, we used data from GfK Denmark to determine the type of product when the digital image gave too little information (18
). In addition, data from the Danish National Survey of Dietary Habits and Physical Activity were used to obtain information on the dietary composition of composite dishes or products. Composite dishes or products are not a big challenge in lunches brought from home for children aged 7–13 years, because they do not often occur. Others have also reported the challenge connected with estimating mixed dishes when assessing dietary intake (12
The DPM is very unobtrusive and would probably not influence the usual eating patterns of the children, but this is still unclear.
This study shows that the DPM in combination with the Meal IQ is valid and reliable when used to assess the quality of dietary intake from school lunch sandwiches brought from home. There is no reason to believe that the DPM in combination with the Meal IQ would be less accurate with adults. The Meal IQ has to be adjusted just a little, so the cut-off points for the different components included in the Meal IQ are adapted to the official recommendations for adults.
Compared to the more traditional dietary assessment methods, the DPM has mainly been used to collect data on individual meals. Measuring the entire diet of free-living individuals complicates the usability of the DPM. Normally, the respondents are not involved in the collection of data. If the whole diet has to be assessed, it will require that the respondents capture the digital images themselves, thereby introducing greater burden on the respondent and the possibility of increased estimation errors because of lower photo quality and a decreased standardisation of the method. The higher response burden could also affect the compliance negatively. Some studies have examined the possibility of assessing food intake among free-living people. Lassen et al. (17
) used a DPM in private homes where the participants were instructed on how to capture digital images on their evening meals to standardise the procedure. Lassen et al. concluded that the DPM for this purpose was valid and feasible. Martin et al. (24
) developed a remote food photography method that builds on the DPM. Smartphones were used to capture images of food selection and plate waste and to send the images to a server for food intake estimation. This method was developed specifically to measure energy intake in free-living adults and has proved to be valid.
When food selection and also food intake have to be measured, the standardised DPM is most appropriate when the study population eats in a cafeteria or a classroom, because this makes it possible to collect data on the leftovers. In Denmark, the oldest students often go outside the school during the lunch break, which complicates the use of a standardised DPM. Other methods that incorporate technology would be appropriate for this target group. Boushey et al. (25
) found a strong preference for technology methods among adolescents, compared to pen-and-paper records. Maybe using a smartphone as described by Martin et al. (22
) would be appropriate to take into account the eating behaviour of young people, or a personal digital assistant with a camera and mobile phone card, as described by Wang et al. (26
There is much potential in technological methods for assessment of dietary intake, and future advancements are possible (27
). Future studies have to examine the possibility of using the DPM to estimate food intake in free-living conditions among children; this aspect would be essential for the possibility to measure the entire diet. Furthermore, research on whether the dietary intake observed during one or more meals is predictive of 24-h dietary intake could also be done.
Automation of the nutrient evaluation could be developed and would improve the cost-effectiveness of the method.
In conclusion, the standardised DPM is a valid and reliable approach for assessing the dietary quality of school lunch sandwiches brought from home for children aged 7–13 years. The method does not rely on the respondents to estimate portion sizes and overcomes the recall problems that exist when collecting dietary data on children. The method is cost-effective and enables data collection for large numbers of people. The method is potentially useful for evaluating the effect of different intervention programmes on dietary behaviours from diverse population groups across different ages.