Validity and reliability of a food frequency questionnaire to estimate dietary intake among Lebanese children

Compared with other dietary assessment methods, FFQs are the most practical and cost-effective means for assessing diet in large-scale nutritional epidemiology studies [32]. However, due to the fact that food availability, accessibility and preferences can vary greatly between settings, it has been recommended that questionnaires be developed and validated specifically for the populations they are intended to be used for in order to produce valid and reliable data [9]. In this study, the validity and reproducibility of an Arabic FFQ to be used amongst school-aged children were examined, with the reference method being four repeated 24-HRs. Ideally, the use of a biochemical indicator or marker, such as doubly labeled water, should be adopted in validation studies as it gives an independent measure of validity for dietary intake. However, biochemical measures are expensive, require the use of sophisticated laboratories and equipment, and/or do not provide information on all the nutrients of interest [33]. Hence, researchers tend to rely on robust dietary assessment methods in the validation of nutritional tools. The dietary record approach, typically considered as the gold standard [10, 33], is a burdensome method requiring a high level of motivation for the participant to write down all the foods and their corresponding portions throughout the study period [33]. In the present study, the 24-HR was chosen as the reference method given that this approach is not limited by literacy or motivation level [9, 33]. Through bypassing these two drawbacks, the 24-HR was found to be particularly helpful in investigating dietary intakes amongst youth [33]. The 24-HR has been validated against more lengthy dietary assessment methods and in most cases found to represent participants’ dietary intakes [33–35]. Mullenbach et al. [36] found that, compared to a 3-day food record, 24-HRs administered by phone to a group of adolescents reasonably assessed mean nutrients’ intakes. Moreover, a systematic review conducted by Burrows et al. [37] reported that when compared to doubly labeled water, the 24-HR MPR approach is the most accurate dietary assessment method to estimate total energy intake in children, when conducted for at least 3 days, including weekdays and weekends and using proxies for reporting. In the present study, we used 4 days of 24-HRs, which exceeds the generally accepted minimum number of three recalls needed to represent usual intake [38].

The results of this study showed that the estimated correlations of energy-adjusted data between the FFQ and the 24-HRs were low to moderate, ranging between 0.26 for MUFA to 0.45 for iron. These results are in agreement with those reported by previous studies [29, 39–42]. In a review of 227 studies examining the validity of FFQs, mean correlation estimates for macronutrient intakes as assessed by the FFQs and the respective reference methods ranged between 0.39 for vitamin A and 0.55 for calcium, a range similar in magnitude to what was obtained in the present study [14]. However higher correlation estimates were reported by other FFQ validation studies, reaching as high as 0.85 [43–45]. The observed discrepancies in validation results between the various studies could be due to differences in study protocols (type of the FFQ, sample size, specific nutrients examined, use of reference method, recall period or number of recorded days); study populations as well as between-person variations [46].

The observed moderate agreement between the FFQ and the 24-HRs in this study could be due to the fact that dietary intake, among children, is characterized by high day-to-day variability [47, 48]. Furthermore, errors in portion size estimation and limitations of recall ability could also have contributed to the moderate agreement between the two methods. Increasing the number of recall days has been suggested to enhance this agreement. However, long recall periods may reduce the accuracy of assessment, owing to increasing fatigue and boredom, potential alterations of dietary habits and increasing likelihood of drop-outs [49].

Our results showed an overestimation of dietary intake by the FFQ as compared to the 24-HRs (mean difference was positive for energy and all the nutrients examined). This tendency of the FFQ to overestimate dietary intake was also reported by previous studies in both the adult and pediatric populations [7, 38, 41, 45, 47–49]. Such an overestimation could be due to the large number of foods listed under each food group in the FFQ, thus providing wider selection options, as compared to 24-HRs [50]. Similarly, potential inaccurate subject reporting of frequency of consumption and/or the amount of commonly consumed foods could be an additional source for this overestimation [51].

The overestimation of energy and nutrient intakes by the FFQ as compared to the 24-HRs and the moderate agreement between the two methods were further confirmed by the results of the Bland-Altman analysis, which showed a positive mean difference for energy and macronutrients. Interestingly, the Bland-Altman plots showed that mean differences between the two methods were greater at higher levels of intake. These findings are in line with those reported by previous studies [43, 52] whereby over-reporting of energy intakes were described at higher intake levels, while under-reporting was observed at lower intakes. These results suggest that the FFQ is able to estimate total energy intake better on a group level as compared to the individual level.

The derivation of the calibration coefficients in this study lead to the correction of the mean nutrient intake values that were estimated by the FFQ. In fact, the calibrated values were close to the estimates obtained by the 24 HRs. This finding is in agreement with Araujo and colleagues [29] who also found that the calibrated values were similar to the means estimated by the reference method. The range of calibration factors obtained in this study is similar in magnitude to those reported by other studies. For instance, calibration factors estimated by Araujo and colleagues [29] were found to range between 0.15 for energy and 0.48 for protein intakes; while those estimated by Voci and colleagues were found to range between 0.07 for iron and 0.40 for vitamin C [53]. Slater et al. reported higher calibration factors, varying between 0.89 for energy, 0.41 for carbohydrates, 0.22 for total fat and 0.20 for protein [54]. Calibration factors with values closer to one indicate that dietary estimates obtained by the FFQ are closer to those estimated by the reference method [55]. It is important to note that the calibrated values displayed a considerable reduction in the data dispersion as compared to the original estimates, a phenomenon also observed in other studies and which could be related to the linear relationship between the FFQ and the dietary recalls [53, 54]. Nevertheless, calibration factors are useful in correcting biases in food intake estimates, particularly when dietary intake is the exposure variable in a study of the association between diet and disease [56, 57].

In order to evaluate reproducibility, the FFQ was administered twice, four weeks apart, minimizing potential temporal changes in children’s dietary intake. The Intraclass correlation coefficients between the two FFQ administrations were between 0.31 for trans-fatty acids to 0.73 for calcium, a range similar to the results other reproducibility studies of FFQs in this age group [7, 58, 59]. According to Cade et al. [12] and Willet [10], such a range is considered adequate.

A major strength in the present study is the extensive process adopted to tailor the FFQ items specifically to Lebanese children and their food culture. Furthermore, the portion size section in the FFQ was designed to minimize burden on the participants and provided visual assistance with real size photos. The FFQ was interviewer-administered to ensure adequate completion [9], provide immediate feedback and checking, and minimize possible bias due to lack of understanding of the process or misinterpretation of portion sizes and intake frequencies [14]. However, the results of this study ought to be considered in light of a few limitations. First, the use of proxy in assessment of dietary intake may have introduced a source of uncertainty or error. For example, the mother may not be fully aware of all the food items eaten by her child, especially foods consumed outside the house (e.g., in school) [49]. However, during all data collection (FFQs and 24-HR), the research dietitians made sure that the child was present with his/her mother, which gave the opportunity for the mother to consult with the child regarding foods not consumed at home. Furthermore, the social desirability bias could not be ruled out as parents usually tend to overestimate intakes of food considered ‘healthy’ and underestimate less ‘healthy’ foods [14, 60]. In order to minimize such a bias, interviewers were trained to limit any judgmental verbal and non-verbal communication during the completion of the 24-HRs and FFQs. Second, many FFQ validation studies conducted among children and adolescents have used the food record method as the gold standard [45, 61, 62]. However, among populations with low schooling levels and consequently a high illiteracy rate, the use of the food record method becomes challenging and may lead to a selection bias [63]. In this study 26 % of the participating mothers had less than high school level of which 23.4 % had intermediate school level or less than. Therefore, in the context of this study, the 24-HR was the method of choice to be used as reference. The MPR approach used in this study to collect the 24-HRs minimized memory bias and standardized the interviews by using five probing stages [64]. Another limitation that ought to be considered is the use of the USDA database, as opposed to a locally developed database. However, in the context of the study, there exists no Lebanese food composition database. For traditional Lebanese dishes, a specific food composition table developed for Middle Eastern foods was used [26]. Though the use of USDA database could have led to miscalculation and errors of nutrients intakes estimations, it is less likely to have affected the reliability and validity measures of the developed FFQ since this database was used in the analysis of nutrient intakes of both the FFQ as well as the 24-HRs. Related to the dietary assessment method, is the limitation of seasonal variation, which poses the possibility of a potential error due to seasonal inconsistency of food intake [65]. In this study, in order to minimize the errors of seasonal variation, mothers were encouraged to indicate the consumption of fruits and seasonal foods and as such adjustments in calculation of nutrients intake was carried out. In addition, data collection took place in various seasons of the year to capture a wider spectrum of dietary intake across seasons. It remains important to note that a self-selection bias could have led to the over representation of the overweight and obese children in the study sample. In fact, the latest national estimate of overweight and obesity prevalence among children 6–9 years old was 36 % [22] versus 52 % obtained in this study. Such a self-selection bias could be explained by the fact that mothers of heavier children may be keener to participate in nutrition-related studies in the hope to learn more about nutrition and good dietary practices.