The present study was conducted to assess validity of a dietary questionnaire and test its ability to rank individuals according to the level of intake of specific foods and food groups.
It has been suggested that when validating a questionnaire on present diet using a reference method, correlation coefficients should be ≥ 0.3 preferably over 0.4 and optimally in the range of 0.5-0.7 [11, 12, 25]. Of the 21 questions assessed here, 13 questions for the men and 14 for the women had a correlation ≥ 0.3 thereof 12 questions for the men and 10 for the women had a correlation ≥ 0.4 The foods showing the highest correlation were not in all cases identical for both genders, and men generally had higher correlations than women. The questions that had a correlation above 0.3 for both genders were on fresh fruits, oatmeal/muesli, raw vegetables, candy, dairy products, milk, pure fruit juice, cod liver oil, coffee, tea, and sugar used in coffee/tea.
The correlation between the AGES-FFQ and the reference method was not significant for fish, meat, cooked vegetables and soft drinks/sweetened juices. Part of the explanation for low or no correlation in general may be the inability of a 3-day food record to adequately reflect individual intake of foods that are consumed infrequently. Soft drinks are an example of this possible limitation of the reference method. For such food the 3-day food record may not be the ideal reference method, as the food in question may not show up on the food record. The AGES-FFQ data might even be closer to true intake in these cases. However, fish, meat and cooked vegetables were not consumed infrequently (2-4 times per week on average), and food items less frequently consumed had acceptable correlation between the two methods. A possible explanation for no correlation for meat and fish consumption might be the lack of distribution for answers to the AGES-FFQ, as almost 90% of participants marked either of two options - 1-2 times a week or 3-4 times a week - reflecting the uniform consumption of both fish and meat in this age group. Answers to the question on cooked vegetables were slightly better distributed even though almost 70% of participants answered either of the two previously mentioned options. In such cases, results from the food records may be better suited to rank individuals' intake. The validity of global questions with narrow distribution of answers, such as for meat, fish and vegetables, could presumably be improved by increasing frequency options to improve distribution, as well as by splitting them up into separate questions on types of meat, fish, etc. It is known that global questions may underestimate consumption , and affect validity. Global questions, chosen for the sake of simplicity, may thus limit the validity of the AGES-FFQ.
The results from the Jonckheere-Terpstra trend test, Kendall's tau-b and Pearson Chi-Square were mostly in agreement with the results from the Spearman's rank correlation, with few deviations. Limited and/or skewed distribution of answers from the AGES-FFQ may contribute to these differences between the methods.
The validity of questions on midlife diet (40-50 y) in the AGES-FFQ has previously been assessed . Retrospective food intake was estimated, where elderly individuals answered the AGES-FFQ on midlife diet, and data were compared to a detailed dietary history, obtained from the same individuals 18-19 years previously, i.e., in midlife. Questions and frequency options were mostly similar for the two periods of life in the AGES-FFQ. In the validation study for midlife diet, the reference method may have been better able to detect correlation for food consumed < 2-3 times a week than the 3-day weighed food recording used in the present study. In the previous study a significant correlation was found for fish and meat consumption (r = 0.25-0.30), along with a stronger correlation for cod liver oil . Part of the explanation for higher correlation in some cases might be linked to the detailed dietary history used as a reference method in the previous validation study, reflecting long-term diet. Another possible explanation for higher correlation for midlife diet might be that both dietary assessment methods, i.e., dietary history and FFQ, can be subject to similar sources of error, such as bias to overestimate foods considered healthy, and to underestimate foods considered unhealthy.
Looking at the distribution of intake according to the two methods, there was a tendency for higher consumption in grams from the food records than would be expected for certain foods/food groups, considering frequency of consumption according to the AGES-FFQ and the calculated consumption using gender-specific portions. This could partly be explained by exceptionally large portions consumed by a few individuals according to their food records. The largest single meat portion was 600 g; the largest portion of soda was 900 ml, and a few individuals had a daily consumption of milk ≥ 1000 ml, while their reported frequency of intake was 3-4 times per week to once a day. This discrepancy emphasizes the limitation of using an FFQ without portion sizes.
In an attempt to evaluate possible over-/underestimation of intake, frequency of intake was compared between the two methods, using actual eating occasions from the food records (data not shown). There was no clear sign of over-/underestimation related to gender or foods/food groups considered healthy/unhealthy. However, foods consumed infrequently according to the AGES-FFQ may not have shown up in the 3-day food records and lead to the perception of overestimation according to the AGES-FFQ. Reported frequency of milk intake according to the AGES-FFQ was generally lower than according to food records. One possible explanation may be that milk used in coffee/tea, or milk poured on breakfast cereals/porridge was not included when answering the AGES-FFQ.
In order to evaluate the representativeness of our study group, general characteristics of the group were compared to the participants of the AGES-Reykjavik study, for which the AGES-FFQ was designed. The AGES-Reykjavik study originates from the Reykjavik study established in 1967, which consisted of 30,795 randomly sampled men and women born 1907-1935. This large cohort equalled roughly 35% of this age-specific population in Iceland. The AGES-Reykjavik cohort was randomly sampled from the 11,549 individuals still alive when examinations began and is thought to represent the study population fairly well. The participants in the present study were heavier, had less fat free (FFM) mass, more fat mass (FAT) and a higher BMI. The weight and amount of FAT may possibly be related to our study group being slightly younger than the average participant in the AGES study, as aging is commonly accompanied by weight loss [27, 28]. With respect to the lower FFM of our participants, the fact that they signed up voluntarily to participate in the IceProQualita study, which included supervised exercise three times per week, may indicate that they themselves felt their physical fitness needed improvement, and that their weight should be better managed. This is further emphasized by the fact that our participants spent less time walking than the AGES participants, indicating that they were less physically active.
In spite of statistical significance for selected variables between participants of the present study and the AGES study, these differences are not extensive. Therefore, our study sample is still thought to represent the AGES group adequately for the purposes of validation.
Weighed food records are generally perceived as a good measure of food intake [11, 12], and have the least correlated errors with food frequency questionnaires . However, day-to-day variation can be great and even greater for individual food items than for nutrient intake . Hence, a longer period of food recording, or repeated recordings, would have been needed in the present study to find correlation to certain answers of the AGES-FFQ.
Nonetheless, a majority of the questions in the AGES-FFQ had an acceptable correlation (r = 0.3-0.7) and may therefore be used to rank individuals according to intake. Questions with lower or insignificant correlation, such as on fish and meat consumption, should not be ruled out or considered invalid without further assessment, as the validity of certain questions is likely to be underestimated rather than exaggerated. However, the same applies here as in the previous study on the AGES-FFQ on midlife diet, that is, that even though the AGES-FFQ on present diet is able to rank individuals according to their intake of several important food groups, one should always be aware of the limitations of the method and the different results seen for different food items. It should also be noted that the AGES-FFQ is only appropriate for ranking individuals according to level of intake of selected foods and food groups, and not for assessing total food intake, energy or nutrients.