The research was a part of the OPUS School Meal pilot study, which was conducted to test multiple measurement procedures, logistics, cooking, and serving of NND meals for schoolchildren. The data collection was performed in January (baseline) and in February/March (intervention) 2011. The full pilot study design is illustrated in Fig. 1. The schoolchildren had their usual packed lunches during the baseline period and were served the NND, including a mid-morning snack, an ad-libitum hot lunch meal, and an afternoon snack, during the intervention period. The menu plan for the NND served can be seen in the Appendix. The children, assisted by parents, recorded their diet in the evening using the WebDASC during each of the time periods of seven consecutive days: baseline and intervention. In the same periods, the children wore accelerometers. Detailed oral and written instructions on how to record food intake and how to use the accelerometer were given individually to all the children and their parents. In addition, trained personnel interviewed parents on their social background, health issues, and attitudes and knowledge about food and health. After the baseline assessment, participants completed a user acceptability questionnaire about the WebDASC, and trained personnel in the OPUS mobile laboratory took anthropometric measurements of body weight and height.

This study was conducted in accordance with the guidelines laid down in the Declaration of Helsinki and the Biomedical Research Ethics in the Capital Region of Denmark approved all procedures involving human subjects.

Children in 3rd and 4th grade at a school in north-eastern Denmark, in total 105 pupils aged 8–11, and their families were invited to take part, and 81 gave their written informed consent.

WebDASC was developed as an interactive food record–recall method. Participants recorded their diet in WebDASC in the evening after the final eating occasion on each day for seven consecutive days.

WebDASC guides respondents through six daily eating occasions (breakfast, morning snack, lunch, afternoon snack, dinner, and evening snack). For the diet recording, a database of 1,300 food items was available, either through category browsing or free text search, aided by a spell-check application. It was possible to type in foods not otherwise found through category browsing or text search. The amount consumed was estimated by selecting the portion size from four different digital images among 320 photo series. Furthermore, participants recorded any intake of supplements, whether a recording day represented usual or unusual intake, and reasons for unusual intakes, such as illness.

WebDASC includes internal checks for frequently forgotten foods (spreads, sugar, sauces, dressings, snacks, candy, and beverages).

To make the interface appealing for children, WebDASC uses an animated armadillo as a guide and the following features to create motivation: a food-meter displaying the total amount of food recorded so far, a most-popular-food ranking, and a computer game with a high score list. The rank list and game is accessible after completing one recording day. The opening screen and food search and selection screen are illustrated in Figs. 2 and 3. If a participant failed to report food intake one day, parents were reminded the next day by e-mail. If a participant failed to record food intake for a day within 48 h, the recording day was automatically closed for further registration, and the next recording day was opened.

For participants to be included in the analyses, the WebDASC had to be completed for at least three weekdays and one weekend day. The EI was calculated for each individual using the software system GIES (Version 1.000 d – 2010-02-26) developed at the National Food Institute, Technical University of Denmark, and the Danish Food Composition Databank (version 7; Søborg; Denmark; 02-03-2009).

The children were instructed to wear the accelerometer (ActiGraph^TM GT3X, Tri-Axis Accelerometer Monitor, Pensacola, FL) 24 h a day for seven consecutive days. The accelerometer was worn in an elastic belt on the right hip, also when asleep, and the participants were instructed to remove it only during activities involving water, as when showering or swimming. The freely available software Propero Actigraph analysis software (version 1.0.18; http://sourceforge.net/projects/propero/files/) was used to analyse the accelerometer data. As recommended by Treuth and colleagues (26), time periods of at least 20 consecutive minutes of zero counts were considered as representing periods when the monitor was not worn or when the child was sleeping and were thus disregarded before analysis. Derived variables were mean counts per min (cpm) based on the vertical axis. Criteria for a successful recording were a minimum of four days of 13 h wear-time per day including at least three weekdays and one weekend day as for the diet assessment.

Activity energy expenditure (AEE) was derived from the mean cpm using Ekelund et al.'s modified prediction equation (27): AEE (kcal/day) = 66.847+(cpm×0.953)–(176.91×gender). TEE was then calculated as AEE plus BMR plus DIT, which was assumed to account for 10% of TEE (28) (TEE = AEE + BMR + DIT). BMR was calculated using the Schofield equation as improved by Henry (29) based on age, sex, height, and weight.

After the baseline diet and activity reporting, and after fasting overnight, participants were weighed once, without shoes and in light indoor clothing, to the nearest 0.1 kg on an electronic digital scale (Tanita BWB-800S, Tokyo, Japan). Their height was measured without shoes to the nearest 0.1 cm with a stadiometer (CMS Weighing Equipment LTD, London, UK).

At the personal interview, each child was given a questionnaire to be completed with the assistance from one parent after the baseline recordings and returned to give us qualitative feedback on the user acceptability of the WebDASC. The questionnaire contained 18 questions, with 12 of the questions using rating scales, and the rest using either multiple choice or open responses. The questions included the amount of help provided by parents, the time spent recording information on the first day and the following days, how acceptable this time was, the preferred search functionality for the child and for the parents, how helpful the images of portion sizes were when estimating portion sizes, the usefulness of the guidance given, self-assessed reactivity, and questions about the interface design, the game, and suggestions for improvements.

Our definition of under-, acceptable, and over-reporters of recorded EI was assessed using the confidence limits of agreement between the EI and TEE recorded at the individual level as suggested by Black (30). This method has been found suitable for adults. As far as the authors are aware, no cut-off methods specifically designed for children have been published. Including data from both measurement periods, we defined misreporters of EI using the ratio EI:TEE, and acceptable reporters were defined as having a ratio of EI:TEE in the range 78–122%. We defined under-reporters as having EI:TEE < 78% and over-reporters as having EI:TEE > 122%. Differences in the number of under-, acceptable, and over-reporters between baseline and intervention were evaluated using χ² test.

The agreement between EI and TEE at group level was evaluated by comparing means, using the paired-sample t-test. Limits of agreement and distribution of bias over the range of energy values were assessed using a modified Bland–Altman plot for repeated measurements, which takes into account both a random error within and between individuals (31), which makes it possible to include each subject twice. The limits of agreement were defined as twice the corrected standard deviations of the differences above and below the mean.

The agreement between EI and TEE at the individual level was evaluated using the cross-classification of EI and TEE divided into quartiles and applying kappa statistics. Pearson's correlation coefficients were also calculated.

The repeatability of EI between baseline and intervention was assessed using the Intra Class Correlation Coefficient (ICC).

Linear-mixed models were used to assess a potential intervention effect (the period factor: baseline; intervention), the effects of gender, parental education, BMI, age, illness reported as affecting eating, and the mutual interactions of all these on EI:TEE. The fixed-factor effects in the model were gender, parental education, BMI, age, illness that affected eating, and measurement period, and their two-way interactions. To adjust for dependency in repeated measures within subjects, random effects were added for subject. The homogeneity of variance and normality of the residuals were examined using graphical methods.

In all the statistical analyses, a significance level of 5% was applied. Data were analysed using SPSS for Windows version 19.

At group level, we found no differences between EI and TEE at baseline (−0.02 MJ/d; p=0.92) or at intervention (−0.06 MJ/d; p=0.790) (Table 2). EI, EI:TEE, EI:BMR did not differ significantly at baseline and intervention (7.1 vs. 7.3; 1.0 vs. 1.0; 1.42 vs. 1.47; p>0.2). The Bland–Altman plot of repeated measures demonstrated positive differences between EI and TEE at higher values of EI and TEE and negative differences at lower values (Fig. 4). The 95% limits of agreement were 3.48 and 3.44 MJ/d for both periods.

Including both recording periods, the Pearson's correlation coefficient between EI and TEE was 0.31 (p<0.001). The proportion of participants appearing in the same or adjacent quartile for both EI and TEE was 73%; 20% was misclassified and 7% was grossly misclassified. The value for the Kappa coefficient was 0.128 (p=0.047) indicating slight agreement (33).

Separated by periods, the Pearson's correlation coefficient was 0.32 at baseline (p=0.004) and 0.30 at intervention (p=0.011). For the baseline period, 77% were classified into correct or adjacent quartiles and, for the intervention period, 69% were classified into correct or adjacent quartiles.

Approximately 20% of the study population were defined as under-reporters and 20% as over-reporters of EI at baseline and intervention, reflecting the trend seen in the Bland–Altman plot. There was no significant (p=0.80) difference in the proportion of under-, acceptable, and over-reporters between baseline and intervention (data not shown). A significantly higher proportion of under-reporters were affected by illness than for acceptable and over-reporters (Table 3). More girls than boys were misreporters; under-reporters had significantly higher BMI compared to acceptable and over-reporters, while over-reporters had lower BMI compared to acceptable and under-reporters.

The results from the linear-mixed models show that main effects of age, parental educational level, and measurement period (baseline vs. intervention) were not significantly associated with EI:TEE. Respondents who were not affected by illness had an 11% higher EI:TEE than those who were affected, mainly because of a higher EI (1,233 kJ/d higher). Boys had a 10% lower EI:TEE than girls, because they had a 1,682 kJ/d higher TEE. EI:TEE decreased 4% for every unit increase in BMI (Table 4).

The ICC between baseline and intervention for EI was 0.45 (95% CI: 0.25–0.61) indicating moderate agreement.

Seventy-four out of 81 who completed the dietary assessment at baseline returned the usability questionnaire. Ninety percent of the children received some help from parents to complete the WebDASC. The average time spent completing the WebDASC the first day was 35 min and 15 min on the following days. A total of 80% found the recording duration acceptable, and 85–90% found the task of the diet recording easy. Children preferred the browse search by category, whereas parents preferred the free text search. Both liked the user interface design.

Approximately 20% were classified as over-reporters and 20% as under-reporters. This is different from other validation studies using motion sensors to validate EI in children, which argue that under-reporting is a large problem (23, 24) (35). Data from studies documenting over-reporting of EI in children are sparse. In the review of Forrestal (36), only two studies described substantial over-reporting in children aged 9 and 4–11, respectively. Using portion size images for portion size estimation of all foods and beverages recorded in the WebDASC may have made it easier to be consistent in either under- or over-reporting compared to other similar dietary assessment methods using a mixture of standard portions and portion size images (10, 24) (25).

In the present study, the children characterized as over-reporters had lower BMI compared to acceptable reporters and under-reporters. This was also observed in a study among children aged 4–11, in which over-reporters weighed less than under- and accurate reporters (37). These findings support the idea that children with a low BMI (or their parents) may be more likely to report larger portion-sizes as they wish (for their child) to increase growth and eat more. It has been shown that maternal pressure to eat is inversely correlated to children's BMI (38).

Under-reporters were more likely than both acceptable and over-reporters to report that illness affected eating during the recording periods. This was confirmed by the results from the linear-mixed models that showed that absence of illness influenced EI:TEE positively. This has also been reported in a study with adults, in which illness during the recording period had a significant impact on under-reporting (39). A German validation study with children and adolescents (1–18 years) found, in agreement with these findings, that acceptable reporters had a higher percentage of normal recording days (36).

Under-reporters had higher BMI and recorded less EI compared to acceptable and over-reporters. Other studies have also shown that under-reporters aged 7–11 have higher BMI and are more worried about weight than acceptable reporters and over-reporters (40–42). Recording accuracy may also be compromised by overweight parents underestimating their children's intake (42). Parents in the present study were more likely to be overweight than a representative sample of parents for the same age group in the Danish National Survey of Diet and Physical Activity (DANSDA) 2003–2008 (54% vs. 42%).

The repeatability of EI between baseline and intervention was moderate. There were some limitations with the repeatability assessment, because the conditions between the baseline and intervention period differed with the NND served for school lunch and snacks in the intervention period. It seems reasonable to expect that reported EI should be approximately at the same level during the two periods since the intervention only covered school meals where food were offered ad-libitum. However, it may have been more difficult for the children and their parents to report the NND in the WebDASC than their usual packed lunches. This may be reflected in the moderate repeatability.

Due to limited resources and the design of the pilot study, the children were only weighed once (after the baseline reporting period). Energy balance could therefore not be confirmed in the present study. However, the dietary assessment period was too short for energy imbalance to present as notable weight change.

During growth and development children are normally in a positive energy balance, but energy accretion is about 1–2% of EI (34, 43), which would not influence the overall results.

One major strength in the present study lies in the use of a reference method to derive TEE, which do not have any errors correlated with the dietary assessment method, as would be a risk if another dietary assessment method was chosen as the reference. Other validation studies of web-based methods in connection with children have used relative validation methods (7, 9) (15). We have used an objective method, which has proved useful with both children and adults (19, 44).

The results from the qualitative questionnaire showed that the WebDASC method was well accepted by participants, who also provided useful feedback for improving the interactive recording method. The present study population consisted of a higher proportion of children whose parents have a vocational education than is the case in the general population (data not shown). This suggests that WebDASC works well, irrespective of parental educational level.

Moreover, we found no difference in recorded EI between measurement periods in the present study. This may be a result of the standardized WebDASC interface guiding respondents through all meals, the use of questions with a conditional response option, which makes it difficult to skip responses, and the use of probing and internal checks to enhance memory and food recording.

In general, the recorded level of EI is low compared to EI recorded by the same age group in the DANSDA 2003–2008 and compared to international reference values, but TEE was also low. This could be a seasonal effect because data were collected during the winter, whereas DANSDA data are collected throughout the year (45). Kolle et al. (46) also found seasonal variations in physical activity level where they used accelerometers in a representative sample of 9-year-old Norwegian children, with their physical activity level being lowest during the winter. Furthermore, a large number of children reported illness during both periods, due to a flu epidemic. As a result of the seasonal low activity level and the illness among children, we measured less variation in AEE than expected and AEE accounted for 20% of TEE.

Accelerometers do not accurately capture certain forms of activity, such as arm movement, carrying loads, and cycling, due to the way the instrument is designed. Moreover, the accelerometer was removed during water activities, such as swimming. However, the average duration of these activities was very low (14 min/day) in the present study, and cannot explain the low TEE.

Errors can have been introduced in estimating the different parts of TEE, because the choice of cut-off for non-wear time to derive AEE may have affected the number of misreporters in either direction. There is no consensus about the most appropriate energy expenditure prediction equation to use with regard to accelerometer data, including how to distinguish between periods of non-wear and bouts of sedentary behaviour (47). The prediction equation used in the present study to derive AEE is based on a small number of participants. However, it is derived from double-labelled water measurements, and as far as the authors know, this is the only equation based on free-living European children of the same age-group as in the present study (27). The accelerometers used in this study have also been shown to have good agreement with the double-labelled water technique comparing TEE_dlw to AEE, TEE or accelerometer counts (48).