Owing to their ease of administration and low burden on the subject, the assessment of dietary intake in epidemiological studies is often assessed by means of food frequency questionnaires (FFQs) 1 . The chief limitation of FFQs, however, is that serious errors can occur if foods governing inter-individual differences in intake of certain nutrients are omitted from the list. Folate is particularly prone to such error, because it is derived from a variety of foods of both animal and plant origin, not all of which can be included in an FFQ. Foods that contribute to inter-individual differences may fail to be included in food lists, potentially confounding estimation of folate intake by the FFQ.

Another implication of inter-individual variation in the intake of folate or any other nutrient is that such variation may be the underlying determinant of associations between food intake and disease. The preventive effect of certain foods on specific diseases is usually the effect of a particular nutrient contained in the food. Such associations are more likely to be detected when inter-individual differences in intake of the nutrient are larger. Conversely, even if a food contains high levels of a particular nutrient, association will be weak if consumption among individuals is similar. The identification of foods that contribute to inter-individual variation in nutrient intake among the population is therefore an important component of any investigation of nutrients responsible for associations between food intake and disease.

According to the National Nutrition Examination Survey of the United States Population 2 and the National Nutrition Survey in Japan 3 , folate intake between the countries is similar among both middle-aged and older age groups. Intake among younger groups, however, is much lower in Japan. This trend is presumably due to differences in the foods that contribute to folate intake. Specifically, folate-fortified food and supplements are large contributors to folate intake for individuals in some Western countries, and questionnaires especially designed to assess folate intake include these items 4 5 . Because folate-fortified foods are not available in Japan and supplement consumption is low, however, folate intake is almost exclusively from natural sources. Nevertheless, little is known about those foods that are the main sources of folate intake among Japanese.

A FFQ was developed and validated for the estimation of dietary intake for the JPHC study. Spearman's correlation coefficients (CCs) between serum folate and folate intake estimated by this FFQ using a biomarker as reference was 0.26 in men 6 . In the present report, we evaluated the validity of this FFQ in the population subgroup for which the FFQ was originally developed using dietary records (DRs) as standard. We then attempted to identify foods that most contributed to folate intake, and those responsible for the differences in intake between individuals. We subsequently repeated these analyses in a second subgroup that was independent of the population for which the FFQ was originally developed, i.e., an external population, to assess its generalizability in Japan. The objectives of this study were: 1) to validate this self-administered FFQ as a means of estimating folate intake, and identify foods that contribute to individual intake and inter-individual variation in folate intake in the population for which the FFQ was developed; and 2) to determine the validity of this FFQ, designed for a study cohort, in estimating the intake of folate and identifying foods that predict inter-individual variation in folate in the general Japanese population.

Daily folate intake as assessed by DRs and FFQ as well as the Spearman's rank CCs between the two measurements by cohort group and sex are shown in Table 1. Mean daily intake of folate based on the FFQ in Cohort I was significantly overestimated compared to the DR data. Both crude and adjusted CCs were higher in men than women in Cohort I but were similar in Cohort II (Table 1).

The 20 foods that made the greatest contribution to total folate according to the DR in Cohort I are listed in Table 2. The list consists of various foods, mainly from plant sources such as vegetables, with spinach making the highest contribution followed by rice, green tea, cabbage, eggs and beer. These 20 foods contributed 55.2% of total intake in men and 52.9% in women. The contribution of the same 20 foods to folate intake in Cohort II subjects was 44.9% for men and 43.2% for women. Gyokuro, a type of green tea, made the second highest contribution in Cohort II, but was not among the 20 in Cohort I. Other food items with the highest contribution in Cohort II but not Cohort I were kamairi-cha (pan-fried green tea), bread, tomatoes, and pumpkins in both sexes; purple laver in men; and sweet potatoes and komatsuna (a green leafy vegetable) in women. When both kinds of green tea were excluded, however, these foods accounted for only 4.6% of total intake in men and 6.2% in women.

The foods that best predicted inter-individual variation in dietary folate and the validity (correlation coefficients) of folate intake based on those foods are shown in Table 3. A total of 33 foods are listed with the cumulative R-square value of 0.59. The food that best predicted variation of intake was green tea, which contributed 12–15% of total intake. No other food predictive of variation contributed more than 1% of total folate intake. The cumulative folate intake from the 33 foods contributed approximately 30% of total intake in both men and women. The CC of intake from the 33 foods in the internal population (Cohort I) was 0.46 in men and 0.28 in women (indicated as "internal" in Table 3), and had approximately the same level of validity as the data from the full 138-food FFQ.

When the same 33-food list was used to compute intake in the external population (Cohort II), the cumulative folate intake contribution was again 30% (indicated as "external" in Table 3), and CC was 0.30 in men and 0.35 in women, showing approximately the same level of validity as in the internal population.

In this study, we evaluated the validity of a FFQ as a means of estimating dietary folate intake in the population for which the FFQ was originally developed. We also attempted to identify foods that differentiated the level of folate intake in individuals by stepwise regression, and tested the validity of assessing folate intake based on the intake of these foods. The results were also cross-validated in an independent population to evaluate generalizability.

Validity of the FFQ in estimating folate intake was moderate in both the internal and external populations. In previous studies, the validity of questionnaires in estimating energy-adjusted dietary folate intake varied from 0.2 to 0.6, depending on the study population 15 16 17 18 19 20 21 . In studies that reported intake from the diet and from dietary supplements separately, CC was 6–21% higher for folate intake that included supplements than for intake from diet alone. Although none of the subjects used supplements that contained folic acid, which had only just become available at the time of the study, the validity of our FFQ in estimating folate from the diet was relatively high, probably because the dietary folate intake of our subjects was as high as that of supplement users in some of the previously cited studies.

Although the largest proportion of folate intake was from vegetables, these did not necessarily explain the differences in intake between individuals. For example, spinach, which is very rich in folate and was one of the highest contributors to mean folate intake, could not explain inter-individual variation because it was consumed by almost every subject. In contrast, green tea contributed greatly to both individual intake and to inter-individual variation, probably because consumption was strongly dependent on individual preference. Although the analysis provided us with information about foods that predicted inter-individual variation in folate intake, some foods which had a moderate partial R-square value contributed less than 0.1% to total folate intake, such as luncheon meat, ham, and so on. These may have been selected by chance alone. In this kind of analysis, even unimportant contributors to the cumulative R-square value may be statistically significant, but can nevertheless be ignored 1 . In any case, it is noteworthy that individuals could be ranked by folate intake based on only 33 foods, with the same level of validity as with the long FFQ.

Any analysis of the possible effects of folate intake on disease also requires analysis of the effect of food items which contribute to total intake and inter-individual variation of folate. It is of great research interest to determine whether the association between food intake and disease is the result of folate intake. For example, we might hypothesize that folate intake may help explain the protective effect of green tea on gastric cancer described in the recent report of the JPHC study 22 . The mechanism of carcinogenesis through DNA instability and methylation abnormalities as a result of folate deficiency has been studied in animal and in vitro studies 23 24 25 , and an association between folate and gastric cancer has been reported in a number of case-control studies 26 27 28 29 . Other prospective studies, however, have failed to show a consistent association between green tea and gastric cancer 30 31 32 . We speculate that the association was strong in the JPHC population owing to the large contribution of green tea to the variation in folate intake, which was not seen in the other populations.

One of the strengths of our study is the precision of the reference data. The ratio of intra- to inter-individual variation in our data was somewhat higher (1.9–4.8) than in several studies in Western countries 33 34 . When intra-individual variation is larger, an increased number of dietary assessment days is required to obtain a valid standard. Although intra-individual variation among our subjects was high, we had more than a sufficient numbers of days (28 days) of data to represent the true intake of the individuals, because the number of DR days needed to estimate true intake within 20% of the true mean with our data was only about 17 to 19 days according to our analysis. By comparison, the greatest number of days of dietary assessment used as standard in previous studies was 14 days 35 .

In addition, our analysis is unique because we used regression analysis to identify the foods most predictive for inter-individual variation in dietary folate, and then evaluated both the internal and external validity of the intake of those foods. It is important to test external validity, because there is no assurance that explanatory variables selected by regression analysis are valid in an external population 1 . To our knowledge, this is the first study to attempt to identify foods that contribute to inter-individual variation of folate intake in Japan, where folate intake is almost exclusively from natural sources. We developed a list of foods that contribute to inter-individual variation in folate intake in the population for which FFQ was originally developed, and tested the generalizability of the results to an external population. The two populations covered various geographic areas throughout Japan. Our results imply the possibility that a shorter questionnaire which specifically targets folate intake in Japanese populations can be developed.

One limitation of our study is that because the subjects needed to be highly motivated to complete the 28-day DRs, they were not a randomly selected sample. Mean folate intake based on DRs was slightly higher than in the entire cohort, probably because the validation study subjects were likely more health conscious and consumed more vegetables. If all subjects had consumed a similar amount of certain foods, inter-individual variation in the food might have been falsely low. The generalizability of questionnaire results needs to be established with care.

Our FFQ is valid for estimating folate intake and is generalizable to the Japanese population. Although some foods such as spinach, rice and cabbage contributed more to total intake of folate, they did not necessarily contribute to inter-individual variation. Validity of the estimation of folate intake based on intake of the 33 foods that most contributed to inter-individual variation was about the same as that based on the original 138-food FFQ. Although folate is contained in a wide variety of foods, 33 foods in the FFQ were sufficient to account for the inter-individual variation of intake and ranking of individuals' intake in Japan. We concluded that the FFQ with 138 foods was valid for the estimation of folate intake, and that the FFQ with 33 foods might be useful for estimating inter-individual variation and the ranking of individual folate intake.

The author(s) declare that they have no competing interests.

FFQ: food frequency questionnaire

DR: dietary record

CC: correlation coefficients

JPHC Study: Japan Public Health Center-based prospective Study

JI performed the data analysis and drafted the manuscript

SY participated in the design of the study, coordinated the study and helped with analysis and with the preparation of the manuscript

HI helped with analysis and with the preparation of the manuscript

MI helped with analysis and with the preparation of the manuscript

ST participated in the design of the study and helped to draft the manuscript. He was principal investigator of the JPHC Study

All authors have read and approved the final manuscript.