Diet in childhood plays vital role on the risk of chronic processes such as obesity, type II diabetes, hypertension, and cardiovascular diseases in adult life [1
]. Thus, an accurate assessment of the child dietary intake is crucial to monitor nutritional status, its relationship with other lifestyles, and to elucidate the consequences on health and disease later in life.
In assessing a complex exposure like long-term dietary intake in epidemiological studies, food frequency questionnaires (FFQs) have been suggested as the best option because this tool saves time and money compared to other dietary methods [3
]. At present, many FFQs have been developed and validated in different populations around the world, mostly in adult populations [4
] and, to a lesser extent, in other specific populations such as pregnant women [7
], children or infants [8
], adolescents [10
] or elderly people [12
To date, few validation studies of FFQs have been published for children aged 6–12 years, of which 11 studies were summarized in a systematic review conducted by Orti-Andrelluchi in 2009 [9
] and 14 studies were developed in more recent years [13
]. Briefly, these studies were conducted in Europe (n
= 7), South-America (n
= 6), Australia (n
= 5), United States (n
= 3), Canada (n
= 2), and Asia (n
= 3); the majority of FFQs were used to assess macro and micro nutrients, and the length of FFQs encompasses a range between 50 and 160 items. The food records and dietary recalls were the preferred dietary assessment methods used as the reference for comparison with FFQs; and average correlation coefficients between nutrient intake in the validation studies was low–moderate (0.3–0.5).
In Spain, there is no validated FFQs to assess dietary intake of school-aged children (6–12 years). Therefore, the purpose of the present study was to examine if the short semi-quantitative FFQ is reliable and valid against three 24-hour dietary recalls (24hDRs) and several vitamins in blood (carotenoids, vitamin D and α-tocopherol) tool to assess the diet in Spanish children aged 7–9 years.
This validation study supports the argument that the FFQ may be an acceptable method for dietary assessment in Spanish children aged 7 to 9 years. Our results suggest that the FFQ was moderately reproducible for most nutrients and food group intakes, with correlations ranging from 0.25 to 0.70. The FFQ also showed a reasonable relative validity for most nutrient intakes in comparison to three 24hDRs, and plasma concentration (carotenoids, vitamin E and D), since correlations were in general higher than 0.20 for most of the nutrients.
Regarding the correlation coefficients to assess the reproducibility of the FFQ for nutrients intakes after 9–12 months’ period, they were consistent with that reported in previous studies conducted in children with similar ages and using shorter periods of time [17
]. The results of fruit and vegetable intake reported in our study are similar to those of another study conducted in children living in the Mediterranean area [38
]. A previous study conducted with 111 Lebanese children aged 5–10 years reported a higher average of correlation coefficients than observed in our study, 0.62 vs. 0.41 [14
], although the better reproducibility could be related to the shorter period of time used between the administration of the two FFQs (one month). Moreover, the number of studies exploring the reproducibility for food groups is very limited [20
]. Field et al. observed coefficients of 0.28 for vegetables and of 0.29 for fruit and vegetables intake [37
], considerably lower than those observed in our study (r
= 0.60 for vegetables, and r
= 0.67 for fruits). However, Buch-Andersen et al. [39
= 0.30 to 0.84) and Saeedi et al [20
= 0.40 to 0.82) reported higher correlations than those observed in our study which may be related to the shorter period of time elapse to complete the two FFQ (2 or 4 weeks), much lower than the 9–12 months’ period used in our study. According to Cade et al. [40
] and Willett [25
], the range of reproducibility correlation coefficients for nutrient and food groups observed in our study may be considered as adequate, and would suggest that our FFQ is a satisfactorily reproducible tool to assess the nutrient and food intake in Spanish children aged 7–9.
Dietary record has been proposed as a “gold standard” method to validate other dietary assessment methods such as FFQs [3
], although they require a high level of motivation and collaboration from participants. Thus, they may become too burdensome and generate losses to follow-up among other problems (e.g., changes in usual diet). In the present study, we considered it more feasible to use a combination of methods, multiple 24hDRs and biomarkers (for specific nutrients), similar to a previous validation study we performed in children aged 4–5 years from the same project [24
]. In accordance with other studies in children with similar age range [13
], we used three non-consecutive 24hDRs as the reference method to validate our FFQ, which has been considered adequate to estimate energy and nutrient intakes [41
]. We did not use a higher number of 24hDRs because this could increase the losses of follow-up and because of the limited resources. In addition, it should be noted that a higher number of recalls can decrease the quality of information collected due to a training effect or the weariness of participants [41
]. The mean nutrient intake estimated by 24hDRs observed in our study were around 15%–25% lower than the estimated by the FFQ. Lower estimates for nutrient intakes have been also reported in other studies when comparing 24hDR and FFQ [13
]. A possible explanation for this lower estimation could be that caregivers did not report the whole children’s food intake, particularly for foods eaten out-of-home, since 62.4% of children in our study reported to use the school lunchroom once a week at least. Despite these disparities in absolute intakes between 24hDR and FFQ, the average correlation coefficients did not change when children attending school lunchroom were excluded from the analysis, which may suggest that the FFQ can be appropriate for ranking participants’ intakes.
The range for correlation coefficients between 24hDR and FFQ observed for crude nutrient estimates in our study (0.14 to 0.48) was similar to those reported by Watson et al. [26
] (0.17 to 0.46), Scaglius et al. [22
] (−0.04 to 0.59) and Moghames et al. [16
] (0.26 to 0.54), and lower than the observed by del Pino et al [13
] (0.18 to 0.74). When energy-adjusted nutrient intakes were used, the correlation coefficients tended to increase although not substantially, which may be in part due to the wide variability of the study population characteristics since our study was population-based [34
]. Similar findings have been also reported for energy-adjusted correlation coefficients in other studies [16
]. Finally, the use of de-attenuated correlations further increased correlations due to correction of the day to day variation in nutrient intakes
The Bland–Altman diagrams showed no evidence of bias for any of the 31 nutrients examined, except for TFA and iodine, where there was some evidence that the FFQ overestimated intake compared with the 24hDR and that this overestimation increased with increasing average intake. The Bland–Altman diagrams only allow a graphical interpretation and the results should be considered together with other results (e.g., correlation coefficients, cross-classification into quantiles). Thus, our study shows a satisfactory level of agreement in intake ranking between the two dietary methods used.
Several validation studies have reported correlation coefficients between FFQ and serum carotenoids in children to additionally support the biochemical validity (calibration) of the FFQ. In one study carried out in 97 children aged 6–10 years [42
], the correlations reported between fruit and vegetables intakes and total serum carotenoids were higher than observed in our study (r = 0.30 vs. r = 0.19), although correlations with specific plasma carotenoids were not performed presented. Our results are comparable to other studies conducted in children [43
] reporting correlations of plasma concentration for α-carotene (r
= 0.25 and 0.27 vs. 0.23), β-carotene (0.14 vs. 0.13) and β-cryptoxanthin (0.36 vs. 0.38). In addition, our results for α-carotene, β-carotene and β-cryptoxanthin were similar to other study populations [24
]. However, we found lower correlations between dietary intake and serum lycopene than the reported in other studies in children with similar range of age [43
]. The low correlation observed in our study between the intake and serum concentration of lutein and zeaxanthin has been also reported in previous studies [24
], probably suggesting that food sources of these carotenoids were not commonly consumed by children and a much large sample would be required to study this association or that the plasma level for this nutrient is not very sensitive to the intake in children at this age range. Burrows et al. reported strong correlations between dietary and serum carotenoids after adjusting for BMI; however, the correlations did not materially change when we adjusted for BMI [43
]. In our study, we also observed very low correlations between dietary and serum for retinol, vitamin D and E which may be due to other non-dietary factors affecting circulation concentrations, efficiency of absorption and metabolism. In fact, retinol concentration is highly regulated by liver stores over a wide range of dietary intakes that can affect serum levels. Also, it has been suggested that plasma concentrations of α-tocopherol may not be a good marker of vitamin E for usually intake in children, and 25OHD may be a poor biomarker of dietary vitamin D since the absorption can be influenced for the sunlight exposure [45
A major strength of the present study may be the use of two reference methods (three 24hDRs and biomarkers) with different error sources to validate the FFQ. Another strength was the use of well-trained nutritionists to collect dietary information during the study period and standard protocols for collecting blood samples. Moreover, the study population was a subsample of the children with very similar characteristics to the whole participant children in the INMA study, a population-based birth cohort and, therefore, the results could be more generalizable to children of the same age range. On the other hand, a potential limitation would be that the dietary information collected by FFQs was self-reported by caregivers and they may be not fully aware of all the food items eaten by their children, especially foods eaten out-of-home. However, we observed that correlations did not change when we excluded the children who had school meals. In addition, the majority of the caregivers in this study had already participated in a prior validation study of a similar FFQ for dietary assessment at the age of 4–5 years and, therefore, they could provide more precise information in this new study. Children were present at the moment of interview and were able to provide additional details about the menu of school lunchroom and other foods and beverages eaten out-of-home. In order to minimize the possible influence of the nutritionists in the dietary report of caregivers, they were instructed in particular to collect the dietary information of the first and second FFQs avoiding subjective opinions. In addition, the 24hDRs were completed by telephone following the USDA Automated Multiple-Pass Method to minimize the forgotten food and beverages consumed by the children in the previous day.