Research evidence shows that there are differences between current dietary practices and nutrition recommendations for children and adolescents. Excessive energy intake and sedentary lifestyle are strongly linked to obesity(Reference Gidding and Dennison1). There is ample evidence that prevention of obesity in school-age children and adolescents requires changes in family and school environments, as well as improvement in their dietary pattern(Reference Daniels, Arnett and Eckel2).
It is more difficult to measure children's dietary intake compared with that of adults; children present greater within-person variability in intake than adults; they have limited skills in recording or recalling what they have eaten; besides, they have limited understanding about what they eat and how the food has been cooked(Reference Rockett and Colditz3).
In both clinical and research settings, having accurate dietary data is of fundamental importance when dealing with health conditions that are associated with nutritional factors. Several instruments have been validated, mostly for adults.
The 24 h dietary recall (24-HDR) provides detailed quantitative information on foods and beverages consumed on the day before the interview(Reference Rutishauser4). It is the method most often used to obtain quantitative data, because it allows the researcher to investigate the mean intake of energy and nutrients in populations of different cultural backgrounds(Reference Fisberg, Martini and Slater5); its application is fast and depends on the subject recalling recent intake. The 24-HDR does not depend on literacy and is the method that is less likely to interfere in dietary behaviour. However, the quality of information will depend on the interviewer's ability to obtain complete and accurate responses(Reference Buzzard6). Information about intake over several days is necessary in order to measure the usual intake. Consequently, it is more expensive(Reference Cullen, Watson and Zakeri7) and implies greater respondent burden. In the case of children <10 years of age, it is preferable that the parents or guardians answer the questions(Reference Biró, Hulshof and Ovesen8).
The underlying principle of the FFQ approach is that the average long-term diet (intake over weeks, months or years) is the conceptually important exposure rather than the intake on a few specific days(Reference Willett9). Instead of gathering information over several days, it provides a global view of the subject's diet over a longer period of time, requiring only one interview(Reference Slater, Philippi and Fisberg10). The FFQ does not require the interviewer to be highly trained, demands fairly less time for the interview and can even be self-administered; it is relatively inexpensive, and the respondents’ burden is less(Reference Biró, Hulshof and Ovesen8, Reference Cade, Thompson and Burley11). The FFQ has been used extensively over the last decades in epidemiological studies. It is most commonly used to obtain estimates of an individual's food intake in relation to the development of various diseases, enabling to rank subjects according to dietary intake(Reference Cade, Thompson and Burley11, Reference Thompson and Byers12). The accuracy of information obtained from an FFQ depends largely on the ability of respondents to give precise information about their usual food intake(Reference Rutishauser4). When the studied population comprises children <12 years of age, the intervention of an interviewer is indispensable(Reference Biró, Hulshof and Ovesen8, Reference Moore, Tapper and Murphy13).
For any new questionnaire, it is important to determine whether its results are reliable and valid. For those instruments that have already been tested, it is necessary to measure their performance in different populations(Reference Willett and Lenart14). To study its relative validity, the agreement between the FFQ and a more accurate method must be evaluated(Reference Slater and Lima15). Ideally, the errors of both methods must be as independent, or uncorrelated, as possible(Reference Willett and Lenart14). A good option would be to use biomarkers as the reference method, but they are often expensive, invasive and nutrient specific(Reference Lopes, Caiaffa and Mingoti16). The diet record, particularly when food is weighed, is likely to have the least correlated errors to the FFQ. Although the 24-HDR presents some correlated errors to the FFQ, the choice of multiple 24-HDR as a standard method is well established, because it is easier to apply and demands less from the subjects than does the food record(Reference Willett and Lenart14).
The purposes of the present study were to adapt an FFQ to measure the usual food intake of 6–10-year-old children, living in Porto Alegre, south of Brazil, and to test its relative validity, comparing it with an average of three 24-HDR. The validated FFQ will then be used to rank children according to their dietary intake.
Materials and methods
Subjects and study area
The present study was conducted on 6–10-year-old children attending grades 1–4 of primary school in Porto Alegre. From a total of 2300 children studied by our research group(Reference Alves17), 103 were selected, by convenience, to take part in this survey. The children had to be living with their parents or caregivers, who would answer the research instruments. Only those children who did not have any gastrointestinal, neurological or psychiatric disorder, were not on a diet for the last 6 months and were not taking any systemic drug were included in the study.
Written consent was obtained from parents or guardians, and oral consent was obtained from the children. The present study was approved by the Ethics Committee of Hospital de Clínicas de Porto Alegre.
Anthropometry
Weight and height were measured in accordance with standard methods for the collection of anthropometric measures(18) by a trained researcher. All measures were taken with light clothing and without shoes. Weight was measured by a digital scale (Plenna®, São Paulo, Brazil). Height was measured to the nearest 0·1 cm using a portable stadiometer (Sanny®, São Paulo, Brazil). BMI was calculated as weight in kilograms divided by the square of height in metres (kg/m2), and the Centers for Disease Control and Prevention BMI cut-offs were used to classify children's weight status(Reference Kuczmarski, Ogden and Grummer-Strawn19). Children classified as overweight and obese were grouped together, because of sample size.
Questionnaires
Information on family's socio-economic status (SES) was collected using a questionnaire(20), which included questions about schooling of parents and characteristics of the house (number of bathrooms and of durable goods). Families were classified into three categories from the highest to the lowest SES (A, B and C).
The 24 h dietary recall (reference method)
The 24-HDR was chosen as the reference method. The questionnaires were administered face-to-face by trained interviewers, according to the protocol developed for the present study, and were based on the multiple-pass 24-HDR technique(Reference Jonnalagadda, Mitchell and Smiciklas-Wright21). When it was not possible to gather details of food consumed, cooking measurement tables and regional recipe manuals were used(Reference Pinheiro, Lacerda and Benzecry22–Reference Ben24). All dietary recall questionnaires were reviewed by the research team before data were transformed into grams, and only then was the nutrient calculation carried out. The children's guardians answered three 24-HDR in non-consecutive days (two of them on working days, and one after a weekend or holiday). After the interview with the parents/guardians, the child was questioned about meals at school.
FFQ for schoolchildren
The semi-quantitative FFQ was delineated to describe the dietary habits of 6–10-year-old children. An interviewer asked the parents or guardians about the frequency with which the child had consumed each item on the food list over the past 6 months. The respondent had to report the amount of food consumed by the child by choosing the portion size from reference pictures that was the closest match to what the child ate. The portion sizes of food were household measures (e.g. spoons, glasses, bowls, etc.) or the usual serving portion (e.g. slice, pack, standard glass, etc.). The total intake of a nutrient was calculated as the sum of the products of the intake frequency, usual serving size and the nutrient content of each food, i.e. Σ(frequency × weight of the usual serving size × nutrient content)(Reference Willett9). The response categories for each food item were ‘never’, ‘less than once a month’, ‘1–3 times a month’, ‘once a week’, ‘2–4 times a week’, ‘once a day, every day’ and ‘every day, more than once’, which assign frequencies of 0, 0·02, 0·07, 0·14, 0·43, 1·0 and 2·0, respectively.
For certain foods, some extra information was required, such as the fat content of milk and the kind of carbonated soft drink or fruit drink (diet or light). At the end of the questionnaire, the respondent could add any food usually consumed by the children, if it was not on the food list. Owing to the tendency of subjects to overestimate the vegetable and fruit intake, at the end of the vegetable and fruit list the parent/guardian had to answer the question: ‘How many times a week does the child eat vegetable/fruit?’ The cross-checking of the two answers resulted in a weighting factor for each group (vegetable and fruit). The cross-check information helps in identifying potential misreporting, and, if appropriate, in adjusting FFQ intake(Reference Calvert, Cade and Barrett25).
A photographic album(Reference Zaboto26) of foods was used to help identify the portion sizes; some photographs of regional foods were added to the album.
Intakes of dietary nutrients were obtained from the FFQ for schoolchildren (FFQSC) and 24-HDR using the US Department of Agriculture nutrient database(27) and analysed using Nutribase 7 Clinical Nutritional Manager Software version 17·0 (CyberSoft Inc., Phoenix, AZ, USA). Nutrient data on frequently consumed foods were updated if necessary and/or complemented with data obtained from local manufacturers of specific industrialized foods.
Adaptation of the FFQ for schoolchildren
In the present study, the FFQSC under evaluation was adapted from a validated FFQ for adolescents in the city of São Paulo, Brazil(Reference Slater, Philippi and Fisberg10). The original distribution of food groups was maintained (candies and pastry; milk and dairy products; fat; grains, breads and tubers; vegetables; fruits; legumes; meat and eggs; and beverages), as well as the frequency of food intake. For the adaptation phase, twenty mothers of eligible children were invited to answer a 24-HDR, answering questions about the food intake of their children on the day before. According to the answers, some food items were excluded, whereas others, especially regional foods, were included. The usual serving size was established as the mean of the study population. In order to better quantify the amount of food eaten, the portion sizes of foods were stratified into small (75 % of the mean), mean, large (125 % of the mean) and extra large (200 % of the mean). The participants of the adaptation phase did not take part in the validation phase.
The resulting FFQSC was analysed by five experts with experience in child nutrition. They suggested some additional changes in the food list, the portion size and the presentation of foods in order to improve the questionnaire.
The final version of the FFQSC comprises ninety food items and was designed to assess the amounts of food consumed over the preceding 6-month period. It was developed to be individually applied by a trained interviewer and requires around 42 min to be applied.
The relative validity of the FFQ for schoolchildren
The relative validity phase was from July 2007 to June 2008. The reference method for comparison was the average of three 24-HDR. Subjects included ninety-one caregivers, who answered the FFQSC and the three 24-HDR. The interviews were mostly carried out at the children's school. On the first day of interview, guardians answered the FFQSC, followed by the first 24-HDR. The next two 24-HDR questionnaires were administered over the next 30 d, when the socio-economic questionnaire was answered and anthropometric measures were obtained.
After collection, the data were converted into energy and nutrients (carbohydrate, sucrose, protein, fat, saturated fat, trans fat, monounsaturated fat, polyunsaturated fat, cholesterol, fibre, vitamins A, D and C, folic acid, Fe, Na, K and Zn).
Statistical analyses
Statistical analyses were performed using the Statistical Package for the Social Sciences statistical software package version 17·0 (SPSS Inc., Chicago, IL, USA). A two-sided P value of 0·05 was used as the threshold for significance. The difference between mean intakes of energy and nutrients was tested using the paired t test. Nutrients that were not normally distributed were described as medians, 25th and 75th percentiles, and compared with the Wilcoxon signed rank test. Adjustment for total energy intake was carried out by the nutrient-density method for nutrients with non-symmetric distribution, and by regression analyses when nutrients were normally distributed. The residuals from the regression represent the differences between each individual's actual intake and the intake predicted by their total energy intake(Reference Willett, Howe and Kushi28).
The association between the two methods was described by the Pearson and Spearman correlation coefficients for nutrients presenting symmetric and non-symmetric distribution, respectively. Crude and adjusted values were analysed. Data from the 24-HDR were adjusted for intra-individual variability in order to obtain the deattenuation of Pearson's correlation(Reference Willett9).
The κ statistic was used to compare categories of nutrient intakes measured by the two methods, informing the per cent agreement. It permits to distinguish the proportion of subjects who, by chance, show good association. The subjects were ranked into the same quartile, or into adjacent or opposite quartiles. The first quartile represents subjects with the lower intake of a nutrient and the fourth quartile represents those in the upper intake of the same nutrient.
The agreement between observations was also analysed as proposed by Bland and Altman(Reference Bland and Altman29) to verify how much the FFQSC probably differs from the 24-HDR (relative bias). The 95 % limits of agreement, estimated as the mean difference plus or minus 1·96 times the standard deviation of the difference, shows the interval between which 95 % of the differences between the measures obtained by the two methods are expected to be.
Results
From a total of 103 children, twelve were excluded from the analysis because of incomplete data. These children did not differ from the ninety-one children included in the validation study (forty-four boys (48·4 %) and forty-seven girls (51·6 %)). Table 1 shows the characteristics of the ninety-one children.
Descriptive characteristics of the children included in the study (n 91)
Table 2A presents average intakes of energy and nutrients as measured by the FFQSC and 24-HDR. The crude estimations of energy and nutrient intakes obtained with the FFQSC were statistically greater than the average of the 24-HDR. After adjusting for total energy (Table 2B), the differences between FFQSC and 24-HDR fell for most of the nutrients. Nevertheless, they were roughly unchanged for carbohydrate, fat and monounsaturated fat, and increased by approximately 1·0 % for sucrose.
Daily intakes of energy and nutrients as estimated by the FFQSC and the 24-HDR (n 91)
FFQSC, FFQ for schoolchildren; 24-HDR, 24 h dietary recall; IQR, interquartile range; RE, retinol equivalents.
†Difference % = (FFQSC − 24-HDR)/FFQSC × 100.
‡Values are median and IQR.
Nutrient intakes, adjusted by energy, as estimated by the FFQSC and the 24-HDR (n 91)
FFQSC, FFQ for schoolchildren; 24-HDR, 24 h dietary recall; IQR, interquartile range; RE, retinol equivalents.
†Difference % = (FFQSC − 24-HDR)/FFQSC × 100.
‡Values are median and IQR.
There were no significant differences between the averages obtained from the FFQSC and the FFQSC after being adjusted for the weighting factor. It was obtained by cross-checking the information of fruit and vegetables consumed over the week and the FFQSC list for those foods (data not shown).
Pearson's correlation coefficients between the two methods are presented in Table 3. Correlations above 0·50 were found for energy, carbohydrate, fat, saturated fat, monounsaturated fat, polyunsaturated fat, fibre, Ca, K, vitamins A, C and D. The correlations ranged from 0·24 to 0·47 for protein, cholesterol, Fe, Na, Zn and folic acid. There were no significant correlations for sucrose and trans fat. When adjusted for energy intake, the correlations for protein, carbohydrate, fat, Na, Zn, Fe, monounsaturated fat, polyunsaturated fat and cholesterol decreased. For saturated fat and vitamin D, the correlations did not change; for fibre, Ca, K, folic acid and vitamins A and C, the correlations increased. Adjustment for within-person variability increased all the correlation coefficients.
Correlation coefficients of energy and nutrient intakes between the 24-HDR and the FFQSC (n 91)
24-HDR, 24 h dietary recall; FFQSC, FFQ for schoolchildren; RE, retinol equivalents.
*P < 0·01, **P < 0·05.
†The energy and nutrient values were log-transformed to normalize the distribution.
‡The deattenuated correlation was calculated according to the equation: r c = r o (1 + (S 2w/S 2b)n)0,5; r c, corrected correlation; r o, observed correlation; S 2w, within-person variance; S 2b, between-person variance; n, number of observations.
§Pearson correlation coefficient.
||Spearman correlation coefficient.
Table 4A shows the ability of the FFQSC to classify individuals into the same quartile of intake as estimated from the 24-HDR. In terms of crude values, the proportion of subjects appearing in the same quartile ranged from 34 % for Fe to 70 % for energy. It was 44 % for nutrients in general. Otherwise, the proportion of children classified into opposite quartiles was 18 %, ranging from 12 % for fat to 24 % for sucrose. The strength of agreement according to the κ statistic ranged from 0·12 (slight) for Fe to 0·36 (fair) for vitamin C. Zn, folic acid and trans fat did not present significant agreement.
Assessment of agreement between the FFQSC and 24-HDR
FFQSC, FFQ for schoolchildren; 24-HDR, 24 h dietary recall; RE, retinol equivalents.
Results are classified into quartiles for energy and nutrient intakes (n 91).
†1 kcal = 4·184 kJ.
After adjusting for energy (Table 4B), the proportion of subjects classified into the same quartile ranged from 32 % for sucrose and trans fat to 53 % for vitamin D. The average was 40·5 % for nutrients in general.
Assessment of agreement between the FFQSC and 24-HDR after adjusting for energy
FFQSC, FFQ for schoolchildren; 24-HDR, 24 h dietary recall; RE, retinol equivalents.
Results are classified into quartiles, for nutrient intake, adjusted by energy (n 91).
The proportion of children classified into opposite quartiles ranged from 8 % for vitamin D to 23 % for sucrose (average 17 %).
After adjustment for energy, polyunsaturated fat and Fe did not show significant agreement, unlike Zn, which presented significant agreement, although very low. The nutrients showing better agreement were saturated fat, K, vitamins C and D. The nutrients that presented worse subject classification were Na, protein, monounsaturated fat, trans fat, sucrose and Zn. The κ values continued to be low, ranging from 0·12 (slight) for Na to 0·44 (fair) for vitamin D.
Figure 1 shows Bland and Altman's(Reference Bland and Altman29) agreement analysis for protein. It illustrates what was observed for most nutrients. It seems that there were no differences in agreement as the intake increased. The large scatter of results indicated a bias between the two methods for most nutrients. This bias was not systematic; it was positive for some nutrients (overestimation) and negative for other nutrients (underestimation).
Agreement between FFQ for schoolchildren (FFQSC) and 24 h dietary recall (24-HDR); middle solid line indicates mean difference; upper and lower dashed lines indicate mean difference ± 1·96 × sd of the difference limits of agreement (i.e. 95 %; n 91)
Discussion
The FFQ analysed in the present study lacks relative validity and does not allow the examiner to confidently classify subjects according to food intake levels. Therefore, it is far from ideal as a tool for epidemiological studies. The number of children presenting any health disorder, such as hypertension, dyslipidemia or obesity, is increasing. Nevertheless, there are only very few instruments that analyse the usual diet of children in this age range. Therefore, we chose to adapt an FFQ validated for adolescents(Reference Slater, Philippi and Fisberg10), thus bypassing the cumbersome development of a food list for the instrument. This adaptation was carried out with the cooperation of experts in children's diet, which oversaw the adaptation of the original food list(Reference Willett9). The protocol of the present study was similar to that used to validate the original questionnaire. The number of days used to calculate the average of 24-HDR was maintained, as increasing the day of observations, in general, does not improve the relative validity of FFQ(Reference Cade, Thompson and Burley11); the method under evaluation (FFQSC) was applied before the reference method in order to avoid any influence from the 24-HDR(Reference Willett and Lenart14). Although the FFQ for adolescents (FFQA) was designed to be self-administered, the FFQSC was designed to be answered by the child's guardian since many children are not able to self-report their intake(Reference Biró, Hulshof and Ovesen8).
Only two other studies have tried to validate an FFQ for this age range in Brazil. Assis et al. (Reference Assis, Guimarães and Calvo30) tested the validity of a different FFQ, designed to be applied in this group, comparing it with the direct observation of three school meals. The respondents were 131 children aged 8–10 years, living in Camboriú, Santa Catarina, south of Brazil, attending a public school. In another study(Reference Fumagalli, Monteiro and Sartorelli31) with 151 children (5–10-year-olds) living in São Paulo, Brazil, an FFQ, designed to be applied in adults in reference to the previous month, was tested against the average of three food records. The authors concluded that other studies would be necessary to validate the instrument.
The FFQ in the present study overestimated all nutrients, whereas the FFQA(Reference Slater, Philippi and Fisberg10) gave similar values for energy, fat, vitamin C and Ca, overestimated values of carbohydrate and fibre and underestimated protein, polyunsaturated fat, cholesterol, vitamin A and Fe. The FFQA was not validated for sucrose, monounsaturated fat, trans fat, Na, K, Zn, vitamin D and folic acid. The FFQSC showed higher average intakes than the FFQA. The finding of overestimation is in agreement with other studies(Reference Fumagalli, Monteiro and Sartorelli31–Reference Marshall, Gilmore and Broffitt34). Fumagalli et al.(Reference Fumagalli, Monteiro and Sartorelli31) did not show significant differences for carbohydrate, protein, Ca and Fe, but their FFQ overestimated fibre, fat, saturated fat, cholesterol and Zn to a greater extent.
Compared with the FFQA (0·46–0·87), the FFQSC (0·22–0·68) presented lower crude correlations with the 24-HDR for energy, protein, carbohydrate, fat and Fe; similar correlations for polyunsaturated fat, fibre and cholesterol; and higher correlations for vitamins A, C and Ca. Fumagalli et al. (Reference Fumagalli, Monteiro and Sartorelli31) (0·21–0·68), Field et al.(Reference Field, Peterson and Gortmaker33) (0·19–0·31) and Marshal et al.(Reference Marshall, Gilmore and Broffitt34) (0·20–0·52) found poorer correlations, whereas the results of Wilson et al.(Reference Wilson and Lewis35) (0·69, for energy) and Bertoli et al.(Reference Bertoli, Petroni and Pagliato32) (0·5–0·7) were similar to ours.
After adjusting for total energy intake, the correlation decreased for some of the nutrients, instead of improving, as could be expected, since the intake variability of these nutrients is related to energy intake. The decrease may be related to an over- or underestimation of the nutrient intake(Reference Willett and Stampfer36).
When the adjusted correlations were corrected for intra-individual variability, all correlation values improved, as occurred with Slater et al.(Reference Slater, Philippi and Fisberg10), Fumagalli et al.(Reference Fumagalli, Monteiro and Sartorelli31) and Field et al.(Reference Field, Peterson and Gortmaker33), pointing to a greater variability in the diet of these subjects.
The agreement between the two methods is at best fair, reaching a κ value of 0·4. Our results are similar to those of some studies(Reference Wilson and Lewis35, Reference Andersen, Lande and Trygg37–Reference Vereecken and Maes39), and superior to others(Reference Slater, Philippi and Fisberg10, Reference Fumagalli, Monteiro and Sartorelli31, Reference Kiwanuka, Astrom and Trovik40). Nevertheless, the present study observed that many children were classified into opposite quartiles (12–24 %), resulting in lower κ values. The comparison with other studies is difficult, because some of them do not inform the agreement into opposite quartiles, or even the κ coefficient. Our results are better than those of Fumaggali et al.(Reference Fumagalli, Monteiro and Sartorelli31), but classified more subjects into opposite quartiles than did Slater et al.(Reference Slater, Philippi and Fisberg10). The difference between the FFQA and the FFQSC may be due to the age range of the subjects, which is characterized by a gradually increasing independence of children in terms of diet. Although the children were asked about their meals outside home, or in the absence of their parents, children <12 years of age have limited ability to recall the portion size of foods they have consumed(Reference Livingstone, Robson and Wallace41, Reference Baxter, Smith and Hardin42).
The FFQSC evaluates the food eaten by children over the previous 6 months, whereas the 24-HDR estimates the average of food intake of the previous 45 days. This may limit the agreement between these two methods.
Comparison data were collected over a 1-year period, but the majority of the data was collected from July to December, representing the food consumption over winter and spring. Therefore, seasonal effects may not have been completely accounted for.
The choice of 24-HDR as the reference method could in itself be a limitation. It depends, as in the FFQSC, on the subjects’ memory(Reference Willett9, Reference Cade, Thompson and Burley11). Nevertheless, this choice was carefully weighed, since the food record method would be operationally difficult, more costly and less well accepted by participants.
Although the adaptation of the FFQ is theoretically possible, desirable and should ideally be carried out before it is used for each different population(Reference Cade, Thompson and Burley11), the procedure appears to be very laborious and bias prone. The FFQ presented good correlation with the reference method, tending to overestimate the nutrients. However, it lacks relative validity, and does not provide the necessary confidence to rank subjects by levels of food intake in epidemiological studies, as far as the comparison with 24-HDR is concerned.
Although our results are similar to those of others, our conclusion is different. Other studies tended to consider that the FFQ is valid in the face of very similar results to ours. We would challenge this conclusion in as far as our results show that the adapted FFQ is inaccurate.
Therefore, we do not recommend the use of this particular FFQ in the current stage, and will wait until further studies ensure that this instrument – or an alternative FFQ – is accurate enough to allow researchers and/or practitioners to use it with confidence.
Acknowledgements
The present study was funded by Fundo de Incentivo à Pesquisa (the Research Fund of Hospital de Clinicas de Porto Alegre). The authors declare that they have no conflict of interest regarding the submission of the present paper. D.L.D.P. carried out the interviews and the initial data analysis; R.F. supervised the statistical analysis and was responsible for the English version of the text. Both authors designed the study protocol. The authors thank Hospital de Clinicas de Porto Alegre for the financial support and for providing statistical advice through Grupo de Pesquisa e Pos-Graduacao (the Research and Post-Graduation Division of the Hospital).
