The valid assessment of dietary intake is crucial for nutrition-related research. However, comprehensive dietary measurement methods are expensive and time-consuming and require a high commitment from participants. Self-administered food frequency questionnaires (FFQs) ask study participants about the eating frequency and portion size of a number of foods they may consume. They are suitable for estimating habitual intake over longer periods of time and are commonly used in large-scale epidemiological studies [1
]. In comparison to other dietary intake assessment methods, FFQs are relatively inexpensive, easy, and quick to administer. Disadvantages of using FFQs are that they are subject to socially desirable answers and rely on memory, which is often reported to be a specific problem for evaluations in older adults [3
]. However, no strong evidence exists that FFQs provide less valid information in community-dwelling older adults compared to younger adults [4
Many Dutch FFQs have been developed and validated for use in the general, adult population excluding young (<20 years) and old (>70 years) individuals [1
]. However, it is important to establish whether FFQs can be validly used in specific population subgroups, such as older adults. Validated tools to estimate the habitual diet of older adults are essential, as incorrect dietary information may lead to false associations of dietary factors with age-related changes in physical, emotional, and cognitive functioning.
In the Netherlands, ethnic-specific FFQs were developed for the HEalthy LIfe in an Urban Setting (HELIUS) study, with a focus on measuring the habitual diet of adults aged 18–70 years of Dutch, Surinamese, Turkish, and Moroccan populations [9
]. The four questionnaires include food items reflecting those commonly consumed by the populations studied and were developed to study the association between diet and (cardiovascular) health. More information is needed on how the Dutch questionnaire performs in an older population.
The aim of this study was to assess the relative validity of the HELIUS FFQ to determine the dietary intake of energy, nutrients, and food groups in a population-based sample of older men and women.
The characteristics of the validation sample as well as those of all LASA participants with FFQ data are shown in Table 1
. The mean age of the validation sample was 71.9 (SD 8.6) years and ranged from 58 to 88 years, 52.7% were female, and all participants had an MMSE score of 24 or higher. The characteristics of the validation sample (N = 88) did not differ from those of the total sample with FFQ data (N = 1399). There were also no relevant differences between the two samples regarding energy, macronutrient, mono/di-saccharides, and dietary fiber intake as assessed by the FFQ (Table S1
shows the mean energy, nutrient, and food group intake from the FFQ and the three 24-hour dietary recalls and the group-level bias between the two methods. Group-level bias was small for the intake of energy, protein, fat, carbohydrates, and alcohol, and for the intake of dietary fiber, (haem and non-haem) iron, and water. Relative to the 24-hour dietary recalls, the FFQ overestimated the intake of animal protein, polyunsaturated fatty acids (PUFA), monounsaturated fatty acids (MUFA), n-3 fatty acids, n-6 fatty acids, alpha linolenic acid (ALA), eicosapentaenoic acid (EPA), docosahexaenoic acid (DHA), mono/di-saccharides, and all other minerals and all vitamins compared to the 24-hour dietary recalls. The FFQ did not underestimate the intake of any of the nutrients. For the food groups, group-level bias was small for vegetables, grains, bread, and sugar/sweets. The FFQ overestimated the intakes of fish, eggs, dairy products, fruit, nuts/seeds, and legumes, whereas it underestimated the intakes of meat, potatoes, alcoholic beverages, non-alcoholic beverages, cakes/cookies, fast food/snacks, and soups compared to the 24-hour dietary recalls.
shows the Pearson’s correlation coefficient, the attenuation factor, and the de-attenuated correlation coefficient between the FFQ and the 24-hour dietary recall for the intake of energy, nutrients, and food groups. For energy, macronutrients, and micronutrients, most Pearson’s correlation coefficients were between 0.3 and 0.5. For the macronutrients, the Pearson’s correlation coefficients were the lowest for MUFA (0.20), EPA (0.21), and DHA (0.21) and the highest for mono/di-saccharides (0.53) and alcohol (0.72). For the micronutrients, the Pearson’s correlation coefficients were the lowest for β-carotene (0.08) and vitamin B1 (0.19) and the highest for haem iron (0.50) and vitamin B12 (0.50). For the food groups, most Pearson’s correlation coefficients were generally between 0.3 and 0.4. They were the lowest for fish (0.14) and grains (0.24) and the highest for non-alcoholic beverages (0.70) and alcoholic beverages (0.78). The attenuation factors varied from 0.03 for β-carotene and 0.06 for DHA to 0.56 for vitamin C and 0.65 for alcohol.
The agreement between the FFQ and the three 24-hour dietary recalls, based on the quintile distributions of the intakes of energy, nutrients, and food groups, is also shown in Table 3
. Agreement was high (i.e., <3% in extreme quintiles) for energy, vegetable protein, ALA, total carbohydrates, polysaccharides, mono/di-saccharides, and alcohol, as well as for calcium, iron, non-haem iron, vitamin B2, vitamin B6, vitamin B12, magnesium, and water. Regarding the food groups, agreement was high for vegetables, bread, alcoholic beverages, non-alcoholic beverages, and sugar/sweets. Agreement was poor (i.e., >10% in extreme quintiles) for fish and legumes only.
Using the Bland–Altman method, the level of discrepancy was estimated with 95% confidence intervals of agreement for the intake of energy and macronutrients only (Figure 1
). The mean difference in intake between the FFQ and the three 24-hour dietary recalls (with 95% confidence interval) was 76 (−1255;1408) for energy (kcal), 3.6 (−47.6;54.8) for total protein (g), 4.9 (−62.7;72.5) for total fat (g), −0.1 (−27.9;27.7) for saturated fatty acids (g), 4.7 (−145.9;155.3) for total carbohydrates (g), and 10.3 (−82.5;103.1) for mono/di-saccharides (g). Based on visual inspection of the Bland–Altman plots, there was no indication for an increase or decrease in the intake difference between the two methods with higher mean intake values.
In this study, we investigated the relative validity of the HELIUS FFQ to estimate the intake of energy, nutrients, and food groups in participants of the LASA study aged 58 to 88 years. For energy and macronutrients, the group-level bias was small, the Pearson’s correlation coefficient was moderate to good, and the agreement based on quintile intakes was moderate to high, implying that the FFQ is able to rank older adults according to their dietary intake of energy and macronutrients. For most micronutrients and most food groups, but not for all, the relative validity was moderate (Pearson’s correlation coefficient between 0.3 and 0.5). For all micronutrients and most food groups (except for fish and legumes), the agreement based on quintile intakes was moderate to high, indication that the FFQ can also be used to rank older adults according to their micronutrient and food group intake. These results suggest that the HELIUS FFQ can be used to estimate the intake of energy and macronutrients and of most micronutrients and food groups in older adults.
Compared to three 24-hour dietary recalls, the FFQ overestimated the intake of fish, eggs, dairy products, fruit, nuts/seeds, and legumes. The poor estimation of fish intake was also reflected in the relatively low Pearson’s correlation coefficients for fish (0.14), EPA, and DHA (0.21 and 0.21) and the poor quintile agreement for EPA (9% in extreme quintile). It is striking that overestimation was observed in particular for food groups (i.e., fish, eggs, nuts/seeds, legumes) for which the average daily intake is generally low in older Dutch adults [17
] and consumption is infrequent [18
]. Using the mean of three 24-hour dietary recalls may not have been appropriate to capture the usual intake of these food groups and nutrients such as EPA, DHA, and beta-carotene, and therefore, no clear conclusions can yet be drawn regarding the validity of the FFQ to estimate the intake of these food groups and nutrients. Future studies using for example nutrient blood concentrations or a higher number of 24-hour dietary recalls per person are needed to validate the estimated intake of the above-mentioned food groups and nutrients by the FFQ in older adults. Until then, the estimated intakes of these specific food groups and nutrients by the FFQ should be carefully interpreted.
It has been suggested that the FFQ method is less suitable for older adults, as it heavily relies on memory. We therefore compared the relative validity of the HELIUS FFQ in older adults as observed in this study to the relative validity of other Dutch FFQs observed in younger adults. The correlation coefficients for the intake of macronutrients for the FFQ and three 24-hour dietary recalls (protein 0.39, fat 0.26, carbohydrates 0.41, and alcohol 0.72) observed in our study were comparable to or only slightly lower than those from two recent Dutch FFQs validation studies conducted in younger adults (protein 0.38–0.51, fat 0.30–0.39, carbohydrates 0.41–0.70, and alcohol 0.77–0.78) [7
]. These comparisons indicate that the relative validity of the HELIUS FFQ in older adults is comparable to the relative validity of other Dutch FFQs in younger adults, which suggests that the FFQ in general is as valid for application in older adults as it is for younger adults to estimate dietary intake.
The observed Pearson’s correlation coefficients for nutrients are very similar to those of FFQ validation studies conducted in older adults that also used multiple 24-hour dietary recalls as the reference method [19
]. For example, their coefficients for energy, protein, fat, calcium, and fiber, respectively, ranged from 0.22 to 0.39, 0.30 to 0.41, 0.25 to 0.58, 0.38 to 0.54, and 0.29 to 0.50. FFQ validation studies in older adults using multiple dietary records as the reference method also showed similar results (0.19–0.40, 0.19–0.45, −0.01–0.47, 0.35–0.49, 0.49–0.55) [21
]. For food groups, a Spanish study conducted in adults aged 55–80 years and using three-day dietary records as the reference method, reported higher correlation coefficients for meat (0.61), fish (0.42), fruit (0.57), and vegetables (0.70) compared to the current study, which could potentially be explained by their much higher intakes of these food groups [24
]. Overall, these comparisons show that the validity of the FFQ is fairly similar to that of FFQs used for older adults from other countries.
The strength of our study is that we specifically included a sample of older men and women. Previous studies investigating the relative validity of Dutch FFQs excluded participants older than 70 years of age [1
] or had a much lower mean age of the study sample [8
]. Our study fills an important knowledge gap, showing that the HELIUS FFQ can also be used to estimate dietary intake in older persons. A second strength is that our validation sample was representative of the total LASA sample with complete FFQs with regard to general demographic, lifestyle, and health characteristics, as well as with regard to the intakes of energy, macronutrients, and dietary fiber. This observation suggests that the results of this validation study can be extrapolated to all LASA participants. Thirdly, participants in the validation study completed a total of three 24-hour dietary recalls in order to take day-to-day variation into account and because using the mean of three recalls allows obtaining results more highly correlated with the true usual intakes of energy and protein compared to using just one or two 24-hour dietary recall(s) [25
]. Finally, the average consumption of energy, nutrients, and food groups from the three 24-hour dietary recalls was used as a reference for dietary intake. In contrast to the FFQ, this method used a telephone interview and was thereby less affected by visual or physical limitations of older adults compared to the completion of the FFQ (on paper or online), and it does not depend on long-term memory and the capability of older participants to calculate their average intake over a time period of four weeks. This will lower the correlation between measurements errors of both methods and decrease the likelihood of inflating the relative validity.
Some limitations of our study should also be addressed. The FFQ was obtained first (with a reference period of the past 4 weeks), followed by three 24-hour recalls obtained within 6 weeks after completion of the FFQ. Even though the intake data of both methods were obtained within a relatively small time frame, the time periods during which the food intake was assessed by both methods did not overlap, likely reducing the reported relative validity. In addition, no objective methods such as urinary or serum nutrient concentrations or total energy expenditure using the doubly labelled water method were available to validate the FFQ. However, the repeated 24-hour dietary recall method is considered the next best alternative reference method when biomarkers are not available [26
]. The FFQ was originally developed for adults aged 18–69 years. Portion sizes might be smaller in older adults compared to younger adults, and the nutritional values assigned to the food lines may be different between older and younger adults [27
], which could potentially impact the validity of the FFQ in older adults. Because the FFQ used relevant units, household measures, as well as pictures to indicate portion size and because portion size estimation is similar in younger and older adults [28
] and using pictures is a valid method to estimate portion size by older persons [29
], this impact is expected to be limited. Furthermore, previous studies conducted in the Netherlands showed that the selected foods for the food lists of an FFQ for older adults were similar to those identified for younger adults [5
]. Although using a telephone interview for the 24-hour dietary recalls has some advantages as discussed above, a face-to-face interview might have been better for older adults suffering from hearing problems. The telephone interview also did not allow the use of food models; however, participants were instructed to measure the content of frequently used kitchenware before their first recall and were sent a picture book to facilitate the estimation of portion sizes. A final limitation is that the number of participants in the current study was lower in comparison to those of other FFQ validation studies (median 110 subjects) [30
] and the recommended number [12
In conclusion, the results of this validation study show that the relative validity of the HELIUS food frequency questionnaire to assess dietary intake in older adults was acceptable to good for energy and macronutrients and for most micronutrients and most food groups. The relative validity is comparable to the reported relative validity of other Dutch FFQs versus 24-hour dietary recalls in younger adults.