The objective of the present study was to examine the relative validity of the FFQ used to assess dietary intake in the Leiden Longevity Study (LLS). The FFQ overestimated as well as underestimated the absolute intake of various nutrients and foods. We observed that the agreement between the two dietary assessment methods in estimating total energy intake was dependent of the intake level. Pearson correlation coefficients ranged from 0.21 (ALA) to 0.78 (ethanol) for nutrients and from -0.02 (NS, legumes) to 0.79 (alcoholic beverages) for foods. Adjustment for total energy intake slightly lowered the correlation coefficients for nutrients, while adjustment for within-subject variation in the 24-hour recalls resulted in higher correlation coefficients for both nutrients and foods. The subjects included in the present validation study were a representative
sample of the total LLS population that provided FFQ data with respect to age, gender, BMI, and lifestyle factors, like smoking habits, alcohol use and being on a prescribed diet. Therefore, the findings of the validation study can be extrapolated to the total LLS population.
Although the estimated mean energy intake did not differ between the FFQ and 24-hour recalls, we did observe that the agreement worsened as total energy intake increased. Siebelink et al.
assessed how accurately participants report their energy intake by a comparable FFQ; they compared reported energy intake with actual energy intake needed to maintain a stable body weight during controlled dietary trials [12
]. Just like the present study, Siebelink et al.
observed a general trend of under-reporting of energy intake at lower intakes and over-reporting at higher intakes [12
]. These results suggest the FFQ is able to estimate total energy intake on a group level, but not on an individual level.
To study diet-disease relationships, ranking of subjects according to their dietary intake is more important than estimating their absolute intake level. Therefore, we examined the relative validity of the FFQ as compared to the 24-hour recalls by calculating Pearson’s correlation coefficients. Adjustment for measurement error in the 24-hour recalls resulted in higher correlation coefficients. For most nutrients and foods, we observed acceptable to (very) good correlations between the FFQ and 24-hour recalls. For the macronutrients, vitamins and minerals, the correlations were comparable to other FFQs [2
]. With regard to SFA, MUFA, and PUFA, the correlations were lower than found for the original VetExpress questionnaire [5
]. However, the correlations observed by Feunekes et al.
] were probably overestimated because they used a dietary history as reference method which is based on the same principle as the FFQ. Although the FFQ in the LLS was designed to assess habitual ALA intake, we observed a poor correlation (r
<0.30) between the FFQ and 24-hour recalls, even after adjustment for measurement error in the 24-hour recalls. In the Netherlands, average ALA intake is about 1.7 g per day in men and 1.2 gram per day in women [15
]. This is lower than the intake estimated using the mean of three 24-hour recalls in the present validation study. As ALA occurs in foods that are consumed infrequently, our reference method may not be suitable to validate the FFQ with regard to ALA intake. Other FFQ validation studies used biomarkers or weighted food records that assess the intake >7 days. In these studies, the summarized correlation for ALA was poor for studies that used biomarkers as reference and acceptable for studies that used weighted food records as reference [16
In addition to nutrients, we also examined the relative validity of the FFQ in estimating habitual consumption of foods. The crude correlation between the FFQ and 24-hour recalls was low for potatoes, vegetables, and nuts and seeds. These correlations were also lower than those observed for the Dutch EPIC questionnaire [17
]. However, adjustment for measurement error in the 24-hour recalls resulted in acceptable correlations (r
>0.40) for these foods. For non-alcoholic beverages, legumes, and composite dishes, we observed no correlation between the FFQ and the 24-hour recalls. In the Netherlands, legumes are consumed infrequently and using the mean of three 24-hour recalls as a reference method may not be not appropriate. With regard to non-alcoholic beverages, our FFQ seems to be unsuitable to rank subjects according to their intake. This is in contrast with the correlation for the Dutch EPIC questionnaire, which was good (r
=0.67 for men and 0.49 for women). Our FFQ was not designed to estimate the intake of liquids and –contrary to the EPIC questionnaire– did not include specific questions about tap water. This may explain the large difference between the intake of non-alcoholic beverages estimated by the 24-hour recalls and the intake estimated by the FFQ found in the present study.
To evaluate underreporting, we calculated EI to BMR ratios and compared them to predetermined cut-off values [8
]. The calculated EI to BMR ratios indicated underreporting of energy intake on group level. Previous studies have suggested that the probability of underreporting increases with increasing BMI [18
]. In the present validation study, we indeed observed an inverse association between the EI to BMR ratio and BMI, indicating that the magnitude of underreporting increases with increasing BMI. This may affect diet-disease relationships. On individual level, ~30% of the subjects had an EI to BMR ratio below the cut-off value. According to Black [9
], data on physical activity are needed to identify diet reports of poor validity. Unfortunately, this information was not available in the present study. To set the cut-off values for underreporting, we assumed a PAL of 1.55, which is the estimated average for a sedentary lifestyle. When the actual PAL is higher, the magnitude of underreporting on group and individual level in the present study is higher. However, as a FFQ is in general not suitable to estimate an individual’s absolute energy intake, and one should be careful when excluding subjects with an EI to BMR ratio below the cut-off value.
A vital component in validating a FFQ is the selection of the appropriate reference method. We used 24-hour recalls as the reference method. 24-Hour recalls are suitable to assess dietary intake on group levels but repeated recalls are needed to estimate usual intake, i.e. capture daily variation at an individual level [1
]. The mean of three 24-hour recalls, as was used in the present study, may not be sufficient to capture the daily variation of foods that are consumed infrequently. As a result, the reference method will perform worse in estimating usual consumption of those foods –and thus the intake of specific nutrients from those foods– than a FFQ. In general, using another dietary assessment method as reference has its limitations. In their literature review, Poslusna et al.
found that misreporting also occurs when using 24-hour recalls or food records to estimate dietary intake [19
]. Thus, measurement errors –both systematic as well as random errors– exist in every dietary assessment method. For validation, especially the random errors need to be uncorrelated; however, this is usually not the case when using another dietary assessment method as reference. A FFQ and a 24-hour recall share common errors as they are both methods based on memory and the same food composition table is used to convert the foods to energy and nutrient intake. Random variation in the 24-hour recalls and the correlated errors in the repeated recalls may underestimate the correlation between the intake assessed with the FFQ and the true intake. On the other hand, correlated errors between the FFQ and the 24-hour recalls may overestimate this correlation. Nowadays, dietary biomarkers, i.e. biochemical indicators of dietary intake or nutritional status; indexes of nutrient metabolism; or markers of the biological consequences of dietary intake [20
], are more often being used as reference method as they are an objective measure of dietary intake and are independent of all the biases and errors associated with dietary assessment methods [21
]. Unfortunately, no data on dietary biomarkers were available in this validation study.