Abstract

Background: The aim of the present study was to assess the relative validity of a self-administered qualitative food frequency questionnaire (FFQ) applied in the Belgian food consumption survey. Methods: Comparison of food consumption data from an FFQ with 7-day estimated diet records (EDR) was made in a sample of 100 participants (aged 15–90 years). The FFQ comprised a total of 50 foods. Both FFQ and EDR foods were categorized into 15 conventional food groups. Results: De-attenuated Spearman rank correlation coefficients between the FFQ and the EDR ranged from −0.16 for potatoes and grains to 0.83 for alcoholic beverages, with a median of 0.40 for all 15 food groups. The proportion of participants classified in the same tertile of intake by the FFQ and EDR ranged from 32% for potatoes and grains to 76% for alcoholic beverages. Extreme classification into opposite tertiles was <10% for milk and soy products, alcoholic beverages, fried restgroup foods and fats. Conclusion: Notwithstanding the short nature and the absence of portion size questions, the FFQ appears to be reasonably valid in both genders and across different age categories for most food groups. However, for the food groups bread and cereals, potatoes and grains, and sauces, estimates should be interpreted with caution because of poor ranking agreement.

Introduction

In 2004, the first food consumption survey was performed in Belgium.1 During this survey, a representative sample of the Belgian population aged 15 years and over was recruited from the national register. The individual level of food and nutrient intake was assessed by two non-consecutive 24-h recalls using EPIC-Soft. A quantitative food frequency questionnaire (FFQ) was used to study the adequacy of food intake in different subgroups of the population. Furthermore, subgroups at risk for a deficient or excessive intake of specific foods or nutrients were identified. An extensive overview of the methods used in this first Belgian food consumption survey is given elsewhere.2

Short FFQs satisfy many conditions to be used as dietary assessment instrument in the context of epidemiological studies because of their inexpensiveness and low burden for participants.3 Also, in the context of nutritional surveillance, they have potential to serve as a quick measure for long-term usual food intake and identification of non-consumers both in adults and children.47 For both purposes (epidemiological and surveillance), however, it is paramount that validity of the instrument is assessed and taken into account during interpretation of results in future use.

Data from the Belgian food consumption survey indicated that the response rate of a short FFQ was higher compared with the 24-h recall interviews, which is very likely to be due to the lower respondent burden. Because of these advantages, the FFQ is being used as a quick screening tool to assess different aspects in the diet of our Belgian population. Therefore, it is important to evaluate the validity of this short FFQ. Hence, the aim of the present study is to assess its validity compared with 7-day estimated diet records (EDR). For 15 food groups, the performance of the FFQ to rank individuals according to intake and agreement with a 7-day EDR will be evaluated. In addition, more extensive analyses will be performed to investigate associated measurement error structures.

Methods

Study design

Using a cross-sectional study design, food intake assessed with a 7-day EDR was compared with food intake assessed with the short FFQ. During a first visit, participants were provided with a general questionnaire comprising socio-demographic and anthropometric questions and a paper-based FFQ. After 2–6 weeks, a second visit was planned, during which both questionnaires were returned to the researchers. Furthermore, a 7-day EDR was provided, and instructions were given for completion. During a final visit, the EDR was collected and checked for completeness by a dietitian. Any remaining quality issues were discussed with the participant and clarified or corrected.

Data collection was performed in Flanders from October 2005 to April 2006. This study was conducted according to the guidelines laid down in the Declaration of Helsinki, and all procedures involving human subjects were approved by the regional Ethics Committee of Ghent University Hospital. A written informed consent was obtained from all participants.

Participants

To resemble best the target population of the food consumption survey (i.e. nationally representative), different age categories were included and equality in gender was pursued. In total, three age categories were recruited: adolescents and young adults (15–29 years), adults (30–59 years) and elderly (60+ years). For those categories, a different approach for recruiting participants was performed. However, a convenient sample was drawn from the population. (i) In adolescents, a multi-stage sampling was performed. Firstly, five secondary schools providing both general education and vocational training were contacted in the region of Ghent. Four schools agreed to participate in the study. Subsequently, parent’s permission was asked by written request. Because selection of classes and communication with parents were performed by the school’s administration, the number of invited participants is unknown. (ii) Young adults and adults invited for participation were acquaintances and family of students and researchers. (iii) Elderly were recruited via social service centres. Elderly living in a residential care setting were excluded given the more limited freedom in food choices. A total of 233 adults (>18 years) were invited accordingly. The participants did not receive any incentive for their participation.

The food frequency questionnaire

The FFQ under study was a self-administered qualitative questionnaire comprising 50 food items. These food items were either individual foods (e.g. soft drink) or an aggregation of similar foods (soy products containing soy milk, soy drinks and soy desserts). The FFQ is a data- and experience-based version, adopted from the Health interview survey in Belgium,8 to be used alongside a repeated 24-h recall using EPIC-Soft in the Belgian national food consumption survey.9 The frequency categories used in the FFQ were as follows: never, less than 1 day per month, 1–3 days per month, 1 day per week, 2–4 days per week, 5–6 days per week, 1 time a day, 2–3 times a day and more than 3 times a day. The usual food intakes derived from the FFQ were calculated by multiplying the frequency of consumption with a standard portion size for each food item.10 The same standard units were used to calculate portion sizes of estimated foods during data entry of the EDR.

Estimated dietary record

Structured open-ended diaries containing pre-defined food groups (including the option ‘other food items’) at six food occasions (breakfast, lunch, dinner and three snacks) were provided to all participants. All participants were informed on how to complete the food record. The diary also contained a written example for future reference. During a 7-day period, all consumed foods and drinks had to be reported with notification of date and place of consumption, estimated consumed quantity expressed as a household measure, unit or weight, specification and if present a brand name. Separate forms were included to report home-made recipes, so name of dish, total quantities of all ingredients and fraction of dish consumed could be stated. Only participants with zero missing dietary records were included in the analysis.

Data and statistical analysis

For almost all food groups, consumption data were not normally distributed. Therefore, only non-parametric tests were used during analysis. First, Spearman rank order correlations of food group intakes between the 7-day EDR and FFQ were calculated for all participants and stratified by gender and age category. In addition, because day-to-day variation in intake of most specific foods is generally high, de-attenuated correlation coefficients were calculated to correct for within-person variation in the EDR.11 Secondly, to assess measurement error of the FFQ, ‘actual values for surrogate categories’ were calculated12 as follows: participants were assigned to tertiles according to food consumption estimated by the FFQ, then the mean food intake in each tertile was calculated using data from the EDR. This gives an indication of the ‘true’ (EDR) values that are indicated by the FFQ tertiles. Kruskal–Wallis one-way analysis of variance was used to determine whether differences of means between tertiles were statistically significant. Third, intakes assessed with the EDR were classified into tertiles, and agreement between both methods was assessed using the weighted κ statistic, calculated with a linear set of weights and CI of 95%.13 This analysis was not performed for food groups for which more than 33.3% of the participants had a zero consumption either for the FFQ or EDR. Fourth, the percentage classified into the correct or adjacent tertile and the percentage grossly misclassified (lowest tertile for one method and highest tertile for the other) was calculated. Finally, agreement between the FFQ and the EDR at the individual level was assessed using mean difference and standard deviation of the mean difference, visually represented by a Bland and Altman plot.14 To correct for non-normal distributions, a Box-Cox transformation was performed on food group intake data from both instruments (EDR and FFQ) before plotting. The grand mean between both methods is plotted (dotted lines) including the 95% CI around the mean (error bars at the right). The fitted regression line is also plotted with the 95% CI (dash-dot lines) and 95% CI for new observations (dashed line). To quantify the error measurement structure, two additional tests were performed on the Bland and Altman plots. The first one tests whether the zero difference line (solid line) is outside of the grand mean 95% CI. The second one tests whether the fitted regression line’s slope is significantly different from zero. Statistical tests were performed using IBM PASW Statistics program version 18.0.0 (SPSS Inc., an IBM company, IL, USA), two-tailed and P < 0.05 was considered as statistically significant. The construction of the Bland and Altman plots including the tests for difference and slope were created using the S-PLUS statistics program.

Results

In total, 156 participants agreed to participate in the study representing a response rate of 55% among adults and elderly. The response rate for the adolescents could not be calculated because cluster sampling was used and the total number of adolescents eligible for inclusion from all schools was unknown. Almost all participants completed the FFQ (n = 155); however, only 100 (64%) were able to complete all 7 days of the food record. Characteristics of participants are shown in table 1.

Table 1

Distribution of age and body mass index categories of participants

 All
 
15–59 years
 
≥60 years
 
 Malea (n = 44) Female (n = 56) Malea (n = 25) Female (n = 23) Male (n = 19) Female (n = 33) 
Mean age (SD) 50 (23.0) 56 (23.4) 32 (12.9) 30 (12.7) 73 (7.6) 74 (6.1) 
Mean BMI (SD) 25.6 (4.0) 24.7 (4.1) 24.7 (4.1) 22.3 (3.0) 26.8 (3.6) 26.4 (4.0) 
BMI category (n, %)       
    Under-/ normal weight (BMI < 25.0 kg/m224 (56) 36 (64) 16 (67) 21 (92) 8 (42) 15 (46) 
    Overweight (BMI: 25.0–29.9 kg/m214 (33) 13 (23) 6 (25) 1 (4) 8 (42) 12 (36) 
    Obesity (BMI ≥ 30.0 kg/m25 (11) 7 (13) 2 (8) 1 (4) 3 (16) 6 (18) 
 All
 
15–59 years
 
≥60 years
 
 Malea (n = 44) Female (n = 56) Malea (n = 25) Female (n = 23) Male (n = 19) Female (n = 33) 
Mean age (SD) 50 (23.0) 56 (23.4) 32 (12.9) 30 (12.7) 73 (7.6) 74 (6.1) 
Mean BMI (SD) 25.6 (4.0) 24.7 (4.1) 24.7 (4.1) 22.3 (3.0) 26.8 (3.6) 26.4 (4.0) 
BMI category (n, %)       
    Under-/ normal weight (BMI < 25.0 kg/m224 (56) 36 (64) 16 (67) 21 (92) 8 (42) 15 (46) 
    Overweight (BMI: 25.0–29.9 kg/m214 (33) 13 (23) 6 (25) 1 (4) 8 (42) 12 (36) 
    Obesity (BMI ≥ 30.0 kg/m25 (11) 7 (13) 2 (8) 1 (4) 3 (16) 6 (18) 

a: For one participant body length was missing so BMI could not be calculated.

Twenty-six percent of the participants were between 15 and 29 years old and 22% were 30–59 years old. These percentages agree with those from the food consumption survey in 2004, where 33% from the sample was between 15 and 29 years, 19% between 30–59 years and 48% >60 years old.9 Forty-four percent of men and 36% of women were overweight or obese (BMI ≥ 25.0 kg/m2). These body characteristics also agree with those from the Belgian food consumption survey where mean BMI was 25.0 kg/m2 for men and 24.2 kg/m2 for women and percentage of men and women being overweight or obese was 47% and 35%, respectively.9

In a previous version, crude correlations were also presented. De-attenuated correlations of consumed foods between both methods are presented in table 2. For the food groups fish, alcoholic beverages and fried restgroup foods, correlation coefficients are calculated on consumers only given the high number of zero consumptions for these foods during the 7-day recording of dietary intake. For the total sample, a strong de-attenuated correlation coefficient (≥0.70) was found for alcoholic beverages. The de-attenuated correlation coefficients were moderate (0.40–0.69) for beverages, fruit, milk and soy products, cheese, restgroup drinks, fried restgroup foods and fats. Fair (0.20–0.39) de-attenuated correlation coefficients were found for vegetables, fish, meat and eggs, restgroup foods and sauces. Finally, weak de-attenuated correlations were found for bread and cereals, and potatoes and grains.

Table 2

De-attenuated Spearman rank correlation coefficients of food group intakes between the FFQ and 7-day EDR

 All Gender
 
Age category (years)
 
Food group (n = 100) Male (n = 44) Female (n = 56) 15–59 (n = 48) 60+(n = 52) 
Beveragesa 0.40 0.57 0.24 0.30 0.49 
Bread and cereals 0.16 0.07 0.29 0.36 −0.08 
Potatoes and grainsb −0.16 −0.41 0.25 0.06 −0.08 
Vegetablesc 0.36 0.59 0.22 0.23 0.42 
Fruit 0.52 0.60 0.48 0.56 0.41 
Milk and soy productsd 0.65 0.61 0.71 0.58 0.77 
Cheese 0.51 0.56 0.49 0.66 0.33 
Fishe 0.22 0.34 0.38 0.09 0.46 
Meat and eggsf 0.35 0.29 0.36 0.33 0.30 
Alcoholic beveragesg 0.83 0.81 0.71 0.94 0.82 
Restgroup foodsh 0.37 0.32 0.48 0.30 0.60 
Restgroup drinksi 0.59 0.65 0.43 0.67 0.18 
Fried restgroup foodsj 0.65 0.80 0.58 0.60 0.71 
Saucesk 0.29 0.30 0.26 0.26 0.33 
Fatsl 0.68 0.60 0.65 0.65 0.56 
 All Gender
 
Age category (years)
 
Food group (n = 100) Male (n = 44) Female (n = 56) 15–59 (n = 48) 60+(n = 52) 
Beveragesa 0.40 0.57 0.24 0.30 0.49 
Bread and cereals 0.16 0.07 0.29 0.36 −0.08 
Potatoes and grainsb −0.16 −0.41 0.25 0.06 −0.08 
Vegetablesc 0.36 0.59 0.22 0.23 0.42 
Fruit 0.52 0.60 0.48 0.56 0.41 
Milk and soy productsd 0.65 0.61 0.71 0.58 0.77 
Cheese 0.51 0.56 0.49 0.66 0.33 
Fishe 0.22 0.34 0.38 0.09 0.46 
Meat and eggsf 0.35 0.29 0.36 0.33 0.30 
Alcoholic beveragesg 0.83 0.81 0.71 0.94 0.82 
Restgroup foodsh 0.37 0.32 0.48 0.30 0.60 
Restgroup drinksi 0.59 0.65 0.43 0.67 0.18 
Fried restgroup foodsj 0.65 0.80 0.58 0.60 0.71 
Saucesk 0.29 0.30 0.26 0.26 0.33 
Fatsl 0.68 0.60 0.65 0.65 0.56 

For fish, alcoholic beverages and fried restgroup foods, correlations are based on consumers only.

a: All drinks (including fruit and vegetable juices and non-sugared soft drinks, excluding milk, soy drinks and drinks from restgroup).

b: Potatoes (excluding fried potatoes and fries), rice and pasta.

c: Raw and cooked vegetables including legumes.

d: Milk, buttermilk, chocolate milk, milk added to coffee or tea, yoghurt, soy drinks and desserts.

e: Fish, shellfish and fish products.

f: Meat, meat products, poultry, game, offal, eggs and vegetarian products (tofu, Quorn, tempeh).

g: Wine, beer and spirits.

h: Sweets and candy bars, chocolate, biscuits and pastry.

i: Sugared soft drinks, sports drinks and energy drinks.

j: Fries, baked potatoes and crisps.

k: Cold sauces like mayonnaise and ketchup.

l: Butter, margarine, low-fat margarine and lard.

Large differences in de-attenuated correlation coefficients were found between men and women. For both, strongest correlations were present for alcoholic beverages and milk and soy products in women. The weakest correlation was present for potatoes and grains in men and vegetables in women. For all age categories, alcoholic beverages yielded strongest de-attenuated correlation coefficients. In the youngest age category, the weakest correlation coefficient was present for beverages, whereas for the middle and oldest age category, there was a negative correlation for potatoes and grains, and bread and cereals in the oldest age category only.

Table 3 includes mean food group intakes based on the 7-day EDR for the FFQ tertiles. For all food groups, significant differences, indicating good ranking of participants, were found between ranks of intake, except for bread and cereals, potatoes and grains, restgroup foods and sauces.

Table 3

Mean food group intakes from EDR for categories based on FFQ tertiles with agreement of tertiles for both methods

 Agreement of tertiles
 
 
 Mean intake EDR (g)
 
 Same tertile (%) Adjacent tertile (%) Opposite tertile (%) Weighted kappa
 
Food groupa T1 T2 T3 pb κ (95% CI) 
Beveragesc 814.0 1017.5 1211.6 0.002 50.0 38.3 11.7 0.30 (0.15–0.44) 
Bread and cereals 120.8 163.9 151.9 0.064 39.2 39.2 21.6 0.13 (−0.01–0.27) 
Potatoes and grainsd 105.3 105.4 105.3 0.894 32.0 42.3 25.8 −0.06 (−0.20–0.08) 
Vegetablese 90.2 107.2 132.1 0.024 46.9 38.8 14.3 0.23 (0.09–0.37) 
Fruit 48.8 115.0 149.3 <0.001 45.5 42.4 12.1 0.26 (0.12–0.40) 
Milk and soyf 71.6 117.5 227.1 <0.001 57.4 36.2 6.4 0.45 (0.31–0.60) 
Cheese 22.1 25.4 37.2 0.001 50.0 37.5 12.5 0.30 (0.16–0.44) 
Fishg 14.0 25.7 33.3 0.003 44.8 44.8 10.4 0.23 (0.09–0.38) 
Meat and eggsh 108.4 122.8 163.5 0.002 44.7 42.6 12.8 0.23 (0.09–0.37) 
Alcoholic beveragesi 18.7 120.3 437.5 <0.001 76.0 21.9 2.1 0.71 (0.56–0.85) 
Restgroup foodsj 48.9 56.6 74.2 0.076 43.8 41.7 14.6 0.20 (0.05–0.34) 
Fried restgroup foodsk 26.9 35.7 72.9 <0.001 52.0 39.8 8.2 0.37 (0.23–0.51) 
Saucesl 6.3 6.2 11.0 0.137 38.1 42.3 19.6 0.10 (−0.04–0.24) 
Fatsm 4.1 15.8 22.4 <0.001 57.7 39.2 3.1 0.50 (0.36–0.64) 
 Agreement of tertiles
 
 
 Mean intake EDR (g)
 
 Same tertile (%) Adjacent tertile (%) Opposite tertile (%) Weighted kappa
 
Food groupa T1 T2 T3 pb κ (95% CI) 
Beveragesc 814.0 1017.5 1211.6 0.002 50.0 38.3 11.7 0.30 (0.15–0.44) 
Bread and cereals 120.8 163.9 151.9 0.064 39.2 39.2 21.6 0.13 (−0.01–0.27) 
Potatoes and grainsd 105.3 105.4 105.3 0.894 32.0 42.3 25.8 −0.06 (−0.20–0.08) 
Vegetablese 90.2 107.2 132.1 0.024 46.9 38.8 14.3 0.23 (0.09–0.37) 
Fruit 48.8 115.0 149.3 <0.001 45.5 42.4 12.1 0.26 (0.12–0.40) 
Milk and soyf 71.6 117.5 227.1 <0.001 57.4 36.2 6.4 0.45 (0.31–0.60) 
Cheese 22.1 25.4 37.2 0.001 50.0 37.5 12.5 0.30 (0.16–0.44) 
Fishg 14.0 25.7 33.3 0.003 44.8 44.8 10.4 0.23 (0.09–0.38) 
Meat and eggsh 108.4 122.8 163.5 0.002 44.7 42.6 12.8 0.23 (0.09–0.37) 
Alcoholic beveragesi 18.7 120.3 437.5 <0.001 76.0 21.9 2.1 0.71 (0.56–0.85) 
Restgroup foodsj 48.9 56.6 74.2 0.076 43.8 41.7 14.6 0.20 (0.05–0.34) 
Fried restgroup foodsk 26.9 35.7 72.9 <0.001 52.0 39.8 8.2 0.37 (0.23–0.51) 
Saucesl 6.3 6.2 11.0 0.137 38.1 42.3 19.6 0.10 (−0.04–0.24) 
Fatsm 4.1 15.8 22.4 <0.001 57.7 39.2 3.1 0.50 (0.36–0.64) 

a: For the food group restgroup drinks, no tertiles could be calculated because >33.3% of the participants did not consume any food from this food group during the 7-day EDR period.

b: Kruskal–Wallis one-way ANOVA.

c: All drinks (including fruit and vegetable juices and non-sugared soft drinks, excluding milk, soy drinks and drinks from restgroup).

d: Potatoes (excluding fried potato products), rice and pasta.

e: Raw and cooked vegetables including legumes.

f: Milk, buttermilk, chocolate milk, milk added to coffee or tea, yoghurt and soy drinks and desserts.

g: Fish, shellfish and fish products.

h: Meat, meat products, poultry, game, offal, eggs and vegetarian products (tofu, Quorn, tempeh).

i: Wine, beer and spirits.

j: Sweets and candy bars, chocolate, biscuits and pastry.

k: Fries, baked potatoes and crisps.

l: Cold sauces like mayonnaise and ketchup.

m: Butter, margarine, low-fat margarine and lard.

The degree of misclassification associated with categorized intakes assessed by the FFQ compared with the 7-day EDR was examined as the proportion of participants classified in the same, adjacent or opposite tertile (table 3). For restgroup drinks ranking into tertiles was not possible because more than 33.3% of participants did not consume any food from this food group during the 7-day EDR. The proportion of participants classified in the same tertile was 50% or higher for beverages, milk and soy products, cheese, alcoholic beverages, fried restgroup foods and fats. Extreme misclassification into the opposite tertile did not exceed 10% for milk and soy products, alcoholic beverages, fried restgroup foods and fats. Results from the weighted κ statistic showed good agreement (0.61–0.80) for alcoholic beverages; moderate agreement (0.41–0.60) for milk and soy products, and fats; fair agreement (0.21–0.40) for beverages, vegetables, fruit, cheese, fish, meat and eggs, and fried restgroup foods. For bread and cereals, potatoes and grains, restgroup foods and sauces, κ coefficients were poor (≤0.20).

Observation of the Bland and Altman plots and associated tests suggested intake-related bias for beverages, vegetables, milk and soy products, meat and eggs, and restgroup foods (table 4 and Supplementary Annex 1). Constant additive error was present for bread and cereals, potatoes and grains, cheese, alcoholic beverages, and restgroup drinks. For bread and cereals, overestimation by the FFQ was observed, whereas for all other foods listed underestimation errors were present.

Table 4

Test statistics of mean differences and slopes of Bland and Altman data after Box-Cox transformation

 Mean difference
 
Regression
 
 
Food group FFQ-EDR Pdifference Intercept Slope Pslope n 
Beveragesa −24.3 0.046 185.4 −0.695 <0.001 94 
Bread and cereals −7.7 <0.001 −2.5 −0.296 0.08 97 
Potatoes and grainsb 4.4 <0.001 10.7 −0.281 0.14 95 
Vegetablesc −1.5 0.01 9.6 −0.636 <0.001 97 
Fruit 0.56 0.21 −1.3 0.179 0.19 87 
Milk and soy productsd −0.3 0.16 −3.04 0.371 <0.001 87 
Cheese −1.43 <0.001 −1.36 −0.018 0.89 81 
Fishe 0.02 0.88 1.58 −0.366 0.06 77 
Meat and eggsf −5.1 <0.001 4.95 −0.314 0.04 94 
Alcoholic beveragesg −1.2 <0.001 −1.20 −0.004 0.98 68 
Restgroup foodsh −0.8 0.03 −3.04 0.276 0.02 96 
Restgroup drinksi −0.47 0.046 −1.38 0.167 0.25 37 
Fried restgroup foodsj 0.043 0.75 1.06 −0.196 0.15 83 
Fatsk −0.095 0.70 −0.43 0.101 0.30 97 
 Mean difference
 
Regression
 
 
Food group FFQ-EDR Pdifference Intercept Slope Pslope n 
Beveragesa −24.3 0.046 185.4 −0.695 <0.001 94 
Bread and cereals −7.7 <0.001 −2.5 −0.296 0.08 97 
Potatoes and grainsb 4.4 <0.001 10.7 −0.281 0.14 95 
Vegetablesc −1.5 0.01 9.6 −0.636 <0.001 97 
Fruit 0.56 0.21 −1.3 0.179 0.19 87 
Milk and soy productsd −0.3 0.16 −3.04 0.371 <0.001 87 
Cheese −1.43 <0.001 −1.36 −0.018 0.89 81 
Fishe 0.02 0.88 1.58 −0.366 0.06 77 
Meat and eggsf −5.1 <0.001 4.95 −0.314 0.04 94 
Alcoholic beveragesg −1.2 <0.001 −1.20 −0.004 0.98 68 
Restgroup foodsh −0.8 0.03 −3.04 0.276 0.02 96 
Restgroup drinksi −0.47 0.046 −1.38 0.167 0.25 37 
Fried restgroup foodsj 0.043 0.75 1.06 −0.196 0.15 83 
Fatsk −0.095 0.70 −0.43 0.101 0.30 97 

Note: values for mean difference, intercept and slopes are in transformed scale.

n: number of participants with positive consumptions of food groups during both collections (FFQ and EDR).

a: All drinks (including fruit and vegetable juices and non-sugared soft drinks, excluding milk, soy drinks and drinks from restgroup).

b: Potatoes (excluding fried potatoes and fries), rice and pasta.

c: Raw and cooked vegetables including legumes.

d: Milk, buttermilk, chocolate milk, milk added to coffee or tea, yoghurt, soy drinks and desserts.

e: Fish, shellfish and fish products.

f: Meat, meat products, poultry, game, offal, eggs and vegetarian products (tofu, Quorn, tempeh).

g: Wine, beer and spirits.

h: Sweets and candy bars, chocolate, biscuits and pastry.

i: Sugared soft drinks, sports drinks and energy drinks.

j: Fries, baked potatoes and crisps.

k: Butter, margarine, low-fat margarine and lard.

Discussion

The purpose of the current FFQ is to be used in the context of nutritional surveillance (i.e. food consumption surveys), complementary to a repeated 24-h recall interview, providing policy makers with a quick instrument to screen consumption behaviour of the population and to detect zero consumers necessary for proper estimation of food intake distributions from more detailed dietary interviews. The present qualitative FFQ showed a fair-to-moderate agreement in ranking participants towards their food intake compared with a 7-day EDR. Considerable differences were found across gender and age categories with respect to the FFQ’s ability to correctly rank participants according to their usual intake. Acceptable ranking of participants by the FFQ was demonstrated for beverages, fruit, milk and soy products, cheese, alcoholic beverages, fried restgroup foods and fats. According to results reported in other studies,15, 16 correlation coefficients between FFQ and EDR-based estimates tended to be lower for vegetables compared with fruit. According to Wakai,16 a possible explanation could be that frequency of fruit consumption is easier to report then vegetables because fruits are more often consumed as raw foods whereas vegetables are more frequently part of cooked dishes and therefore not integrally recalled. In addition, fruit is frequently consumed as a single food item and comes in natural or typical units, whereas vegetables are often sliced or cut which makes them more difficult to quantify. On the other hand, in a Mediterranean population, a higher correlation was found for vegetables compared with fruit by Fernandez-Ballart et al.17

Food groups for which relative validity turned out rather low in the current study were typical carbohydrate-containing food groups like bread and cereals, and potatoes and grains. A possible explanation for bread and cereals might be that bread is likely to be consumed more than once a day with large differences in portion sizes between participants, which is not reflected by the FFQ, and, especially in the older age category, breakfast cereals are consumed less frequently. Also, for the food group potatoes and grains, it was found that in men, the FFQ largely overestimated potato consumption. For food groups with low validity, more detailed questionnaires containing more food items may be needed to accurately assess actual food consumption. On the other hand, the trade-off between adding items for improvement of validity and longer questionnaires, which in turn can affect participation rate, should be kept in mind.

The correlation coefficients for cheese (0.33) and restgroup drinks (0.18) in the oldest age category were lower than those in the other age categories, whereas for fish the correlation coefficient was higher (0.46). In elderly, data from both 7-day EDR and FFQ indicate very low consumption of restgroup drinks (data not shown).

We are aware that the participants included in the different strata are a selected group of the population and not a population-based random sample due to the convenience sampling approach. No upper age limitation was set for inclusion into the study; therefore, some participants were older than 80 years (15% of the oldest age category, data not shown). It was suggested by Rothenberg18 that elderly up to the age of 80 perform well in reporting their food habits retrospectively and that from the age of 80, elderly tend to report food habits earlier in life.

Although extensively used in other validation studies, the current methodology has some drawbacks. When designing a validation protocol, a key step lies in identifying a method that will serve as a reference for the instrument to be validated. Given that no method is perfect to serve as gold standard, it is of vital importance that errors of both methods are as independent as possible.12 Failing to select a reference method of which errors are uncorrelated with those from the instrument under study will inevitably lead to higher estimates of validity. Major sources of error associated with FFQs are caused by restrictions resulting from a fixed list of foods, memory and interpretation of questions. Among the available and feasible comparison methods for validating an FFQ, diet records, with their open-ended format and independency towards memory, are likely to have the least correlated errors.12 In addition, the reference method used in the present study (7-day EDR) is the best obtainable instrument that takes into account within-person variation in food intake, and therefore better reflects usual food intake. Also, replicate measures by EDR allow to correct for within-person variability in food intake by calculating ratios of within- over between-person variances suitable for estimating de-attenuated coefficients of correlation.

A limitation of the current study is the different time frame over which food intake was assessed by both methods. Typically, FFQs are designed to measure long-term dietary intake, whereas EDRs measure short-term intake when not repeated over time. Ideally, replicate 7-day periods, in which food intake is recorded over a time span of one year, is advised to include all seasonal variations. For instance, higher consumption of soups during winter months as opposed to summertime can have considerable impact on total beverages intake. The same will count for consumption of raw vegetables during summer.

The current study compared a quantitative EDR with a qualitative FFQ. To do so, standard portions sizes were used to calculate food intakes assessed by the FFQ. Therefore, it is very likely that two persons with identical frequencies of consumption have a different true consumption of a particular food due to differences in portion sizes consumed. This loss of detail, inherent to qualitative FFQs, will certainly attenuate ranking of individuals because variability in amounts of intake is reduced. On the other hand, it was demonstrated by Noethlings et al.19 that portion size adds only limited information on variance of food intake in a large European sample, suggesting that assignment of standard portions to frequencies of intake seems to be adequate. This finding was also documented earlier in an American sample where it was concluded that due to a smaller contribution of between-person variance to the total variance in portion size, specification of a standard portion size may not introduce a large error in the estimation of food intake.20

Conclusion

In general, the FFQ tends to underestimate food intake compared with EDR. For fruit, fish, fried restgroup foods and fats, no systematic bias was present. Considering the short character and the absence of portion size questions, the FFQ appears to be reasonably valid in both genders and across different age categories for assessment of group level intakes. However, for the food groups bread and cereals, potatoes and grains, and sauces, estimates should be interpreted with caution because of poor ranking agreement.

Supplementary data

Supplementary data are available at EURPUB online.

Acknowledgements

I.H. and H.V.O. were responsible for the study concept and design. I.H. was responsible for the collection of data. W.D.K. was responsible for the statistical analysis and writing of the manuscript. A.D. is acknowledged for his statistical advice and programming of the Bland and Altman plots and associated tests. All authors were involved in the interpretation of data and reviewing of the manuscript. The authors acknowledge the students who conducted the fieldwork of this study in the frame of their thesis: Karen Bresseleers, Josefien Bisschop, Kim Den Blauwen-Van Driessche and Ann Devos. We also would like to thank the respondents who voluntary participated. The University College Ghent is also acknowledged for the Doctoral Research Support Grant (W.D.K.). This work was orally presented at the 11th European Nutrition Conference—FENS Madrid, Spain, October 2011.

Conflicts of interest: None declared.

Key points

  • Large-scale dietary intake assessments should preferably have low burden for participants. Short FFQs meet this condition and are therefore suitable instruments for nutrition research in the context of public health monitoring.

  • To correctly interpret results resulting from the use of these instruments, it is necessary to validate them and investigate their measurement error structure.

  • The present study shows that for some, but not all, food groups, the short FFQ is able to correctly rank participants according to their level of intake compared with a 7-day EDR. For the food groups fruit, fish, fried foods and fats, no systematic group level bias was found.

References

1
De Vriese
S
Huybrechts
I
Moreau
M
Van Oyen
H
The Belgian Food Consumption Survey 1—2004: Report (Enquête de consommation alimentaire Belge 1—2004: Rapport)
2006
2
De Vriese
S
De Backer
G
De Henauw
S
, et al.  . 
The Belgian food consumption survey: aims, design and methods
Arch Pub Health
 , 
2005
, vol. 
63
 (pg. 
1
-
16
)
3
Willett
W
Willett
W
Food-frequency methods
Nutritional Epidemiology
 , 
1998
2nd edn
New York / Oxford
Oxford University Press
(pg. 
74
-
100
)
4
Andersen
LF
Johansson
L
Solvoll
K
Usefulness of a short food frequency questionnaire for screening of low intake of fruit and vegetable and for intake of fat
Eur J Public Health
 , 
2002
, vol. 
12
 (pg. 
208
-
13
)
5
Osler
M
Heitmann
BL
The validity of a short food frequency questionnaire and its ability to measure changes in food intake: a longitudinal study
Int J Epidemiol
 , 
1996
, vol. 
25
 (pg. 
1023
-
9
)
6
Thompson
FE
Subar
AF
Smith
AF
, et al.  . 
Fruit and vegetable assessment: performance of 2 new short instruments and a food frequency questionnaire
J Am Diet Assoc
 , 
2002
, vol. 
102
 (pg. 
1764
-
72
)
7
Therese
L
Lillegaard
I
Evaluation of a short food frequency questionnaire used among Norwegian children
Food Nutr Res
 , 
2012
, vol. 
56
  
doi:10.3402/fnr.v56i0.6399
8
Van Oyen
H
Tafforeau
J
Hermans
H
, et al.  . 
The Belgian Health Interview Survey
Arch Public Health
 , 
1997
, vol. 
55
 (pg. 
1
-
13
)
9
De Vriese
S
Huybrechts
I
Moreau
M
Van Oyen
H
The Belgian national food consumption survey 1. Scientific Institute of Public Health
2004
 
[cited 2012 February]; available from: http://www.iph.fgov.be/epidemio/epien/index5.htm
10
Health Counsil Belgium
Household Weights and Measures. A Manual for a Standardised Quantification of Food Items in Belgium
 , 
1997
pg. 
280
 
11
Beaton
GH
Milner
J
Corey
P
, et al.  . 
Sources of variance in 24-hour dietary recall data: implications for nutrition study design and interpretation
Am J Clin Nutr
 , 
1979
, vol. 
32
 (pg. 
2546
-
59
)
12
Willett
W
Lenart
E
Willett
W
Reproducibility and validity of food-frequency questionnaires
Nutritional Epidemiology
 , 
1998
2nd edn
New York / Oxford
Oxford University Press
13
Altman
DG
Practical Statistics for Medical Research
 , 
1991
London
Chapman & Hall
14
Bland
JM
Altman
DG
Statistical methods for assessing agreement between two methods of clinical measurement
Int J Nurs Stud
 , 
2010
, vol. 
47
 (pg. 
931
-
6
)
15
Mikkelsen
TB
Olsen
SF
Rasmussen
SE
Osler
M
Relative validity of fruit and vegetable intake estimated by the food frequency questionnaire used in the Danish National Birth Cohort
Scand J Public Health
 , 
2007
, vol. 
35
 (pg. 
172
-
9
)
16
Wakai
K
A review of food frequency questionnaires developed and validated in Japan
J Epidemiol
 , 
2009
, vol. 
19
 (pg. 
1
-
11
)
17
Fernandez-Ballart
JD
Pinol
JL
Zazpe
I
, et al.  . 
Relative validity of a semi-quantitative food-frequency questionnaire in an elderly Mediterranean population of Spain
Br J Nutr
 , 
2010
, vol. 
103
 (pg. 
1808
-
16
)
18
Rothenberg
EM
Experience of dietary assessment and validation from three Swedish studies in the elderly
Eur J Clin Nutr
 , 
2009
, vol. 
63
 
Suppl 1
(pg. 
S64
-
8
)
19
Noethlings
U
Hoffmann
K
Bergmann
MM
Boeing
H
European Investigation into Cancer and Nutrition
Portion size adds limited information on variance in food intake of participants in the EPIC-Potsdam study
J Nutr
 , 
2003
, vol. 
133
 (pg. 
510
-
5
)
20
Hunter
DL
Sampson
L
Stampfer
MJ
, et al.  . 
Variability in portion sizes of commonly consumed foods among a population of women in the United States
Am J Epidemiol
 , 
1988
, vol. 
127
 (pg. 
1240
-
9
)

Comments

0 Comments