Impact of the revision of a nutrient database on the validity of a self-administered food frequency questionnaire (FFQ)

BACKGROUND Revision of the national nutrient database in 2000 had a strong impact on the absolute level of estimated nutrient intake in dietary assessments. However, whether it influenced the ranking of individuals by estimated intake, a more important function in epidemiologic studies, has not been investigated. Here, we investigated the effect of this revision of the nutrient database on the validity of a food frequency questionnaire (FFQ) used to estimate nutrient intake in the Japan Public Health Center-based prospective Study (JPHC Study). METHODS Subjects were a subsample of the JPHC Study who volunteered to participate in the validation study of the FFQ. Validity of the FFQ was evaluated by reference to the 28-day weighed dietary records as a gold standard. Nutrient intake according to the FFQ was recalculated using the revised database, and the results were compared to those using the previous database. Spearman’s rank correlation coefficients (CCs) between intakes estimated by the FFQ and dietary records were computed using the revised database, and were compared to CCs computed using the previous database. RESULTS For most of the nutrients, mean intake increased or decreased significantly using the revised database. However, no notable change was seen for the CC between estimated intake according to dietary records and FFQ when the revised database was used for calculation. Differences in the point estimates of the CCs ranged from -0.14 to 0.15. Likewise, CCs between biomarkers and estimated intake according to FFQ were similar for the two databases. CONCLUSION Despite changes in intake levels for many nutrients, the validity of our FFQ using rank correlation by nutrient intake was not influenced by revision of the nutrient database in Japan.

estimated intake is therefore important.
When calculating the individual nutrient intake from foods estimated by an FFQ, food composition databases are used as a source of nutrient contents. Given the variation among databases, database selection would affect the results of individual nutrient intakes greatly. The Standard Tables of Food Composition in Japan, published by the Ministry of Education, Culture, Sports, Science and Technology, is the most commonly used food composition database in Japan. It lists the nutrient contents of various foods per 100g which are average and representative values among those foods available in Japan. The database has been revised on an irregular basis. The Fifth Revised Edition was released in 2000, almost 20 years after the Revised Fourth Edition, 2 and an Enlarged Edition covering additional nutrients was released in 2005. 3 The database was revised to update the nutrient content of a greater variety of food items commonly eaten by Japanese, which have changed over time with changes in manufacture and distribution in the food industry. 3 Further, the revised database is more comprehensive, including additional nutrients not listed in the previous database. This is greatly beneficial when associations with disease are investigated because it allows the estimation of exposure to specific nutrients of interest.
This revision of the nutrient database, however, has greatly influenced the estimation of intakes in the National Nutrition Survey (NNS) in Japan. 4 A decline assumed to be attributable to the revision was observed in average intake for a number of nutrients including iron, vitamin B1, vitamin B2, and vitamin C. Other studies have reported that the degree of difference between the previous and current editions varies by age group. 5,6 Nevertheless, it remains unknown whether the revision of the food composition tables has had an effect on the validity of any of the various FFQs, and the validity of the intake of nutrients newly added in the Enlarged Edition of the Fifth Revised Edition has never been evaluated. Indeed, we are unaware of any previous study which has evaluated the impact of a revision of a nutrient database on the validity of an FFQ.
Here, to investigate the effect of the revision of the food composition tables on the validity of an FFQ, we compared the ranking of individuals by estimated nutrient intake calculated using the revised database (Fifth Edition) to that using the previous database (Fourth Edition) in a subgroup of the Japan Public Health Center-based Prospective Study (JPHC Study) using dietary records (DRs) and biomarkers as references. Additionally, we also evaluated the validity of the FFQ in estimating the intake of nutrients newly included in the Enlarged Edition of the Fifth Revised Edition.

Study Setting
The JPHC Study is a population-based prospective cohort study which consists of two cohorts, the first established in 1990 in the Ninohe, Yokote, Saku, and Chubu (previously named Ishikawa) individual. The intake of each food item was calculated by multiplying the frequency of consumption (never, 1-3 times/months, 1-2 times/week, 3-4 times/week, 5-6 times/week, once/day, 1-2 times/day, 4-6 times/day, 7+ times/day) by relative portion size (small, medium, and large). The food item code in the Standardized Tables of Food Composition, 4th ed. 2 was also assigned for each food item in the FFQ, 14 and daily intake of energy and nutrients according to the FFQs for each individual were calculated by summing the product of the intake of each food multiplied by the nutrient content of that food for the same nutrients which were calculated for dietary records. In addition, folate, vitamin B6, and vitamin B12 intake were calculated using the database developed for the food items which appeared on the FFQ. 15 Because a database of dietary supplements was not available, intake from dietary supplements was not included in calculations for both DR and FFQ.
Energy and nutrient intake according to the FFQ and DR were then recalculated using the Standardized Tables of Food Composition, 5th ed. (revised database). 3 The 4th edition (previ-dish in detail. They also reported all dietary supplements used, if any. At the end of each season, the DRs were reviewed in a standardized manner, and each food was coded using the food item code in the Standardized Tables of Food Composition, 4th ed. 2 by local dietitians. Energy and nutrient intake were calculated by summing the product of the intake of each food multiplied by the nutrient content of that food. The nutrients listed in the Standardized Tables of Food Composition, 4th ed. were protein, total fat, carbohydrate, sodium, potassium, calcium, phosphorus, iron, retinol, vitamin B1, vitamin B2, niacin, and vitamin C. Additionally, for those nutrients with missing values for some foods, i.e., carotenes (alpha-and beta-), 11 fatty acids (saturated, monounsaturated and saturated), 12 cholesterol, and dietary fiber (soluble, insoluble and total), 13 a comprehensive database was developed by substitution methods.
The self-administered semi-quantitative FFQ consisted of 138 food items and 14 supplementary questions concerning the use of dietary supplements, dietary habits, and others. Results were used to assess the usual dietary intake of the preceding year for each DR: 28-day dietary records FFQV: food frequency questionnaire for validity FFQR: food frequency questionnaire for reproducibility BLD: blood collection; URN: urine collection trients. Intakes of all minerals were estimated to be lower with the revised database, most evidently for iron (-8.3% to -12.5%). The impact of the database revision was more obvious for vitamins; among these, intake of carotenes and retinol was 55% and 12.5% higher, respectively, whereas that of B group vitamins was lower.
For nutrients for which we supplemented missing values in the database, intake of monounsaturated fatty acid was lower after the database revision, while that of water-soluble fiber was drastically higher.
In contrast, revision of the database did not have a substantial effect on the validity of intake levels by FFQ compared to those by DR ( Table 2). A greater than 0.1 decline in point estimates of Spearman's CCs was seen only for the crude intake of vitamin B1 and water-soluble fiber in the Cohort I males; in energy-adjusted intake of vitamin B2 in Cohort I females; and in crude intake of sodium in Cohort II females. On the other hand, a greater than 0.1 increased point estimate of Spearman's CCs was observed for the crude intake of crude retinol and polyunsaturated fatty acid in Cohort II females. Confidence intervals of CCs between the previous and revised database overlapped for all nutrients.
Likewise, the validity of the FFQ was not influenced by the database revision when compared to biomarker data (Table 3). For those nutrients for which biomarkers are a good indicator of dietary intake, such as serum polyunsaturated fatty acid, carotenoids, and urinary sodium and potassium, CCs for the estimated intake calculated by the previous and revised databases were similar. As with comparison by DR, confidence intervals of CCs between the previous and revised database overlapped for all nutrients. Moreover, reproducibility (FFQV vs. FFQR) was also not altered by the database revision (data not shown).
Estimated intake according to DRs and FFQ, as well as Spearman's CC, for nutrients which were newly included in the revised database and never previously evaluated for validity and reproducibility are presented in Table 4. Spearman's CC for the estimation of most of these nutrients by FFQ indicated moderate validity (Spearman's CC=0.3-0.6), except for vitamins D and E, which indicated slightly lower validity.
We evaluated the impact of revision of the food composition database on the estimation of energy and nutrient intake by the FFQ in the JPHC Study, and its validity. The results of recalculation using the revised food composition table showed that, notwithstanding a significant impact on the estimation of individual intake levels for some nutrients, the revision had little substantial influence on the validity of individual rankings by estimated nutrient intake.
We observed major decreases in the intake of iron, vitamin B1, and monounsaturated fatty acid, and increases in that of carotene, retinol, niacin and water-soluble fiber as a result of revision of the food composition database. These results are in agreement with several previous studies which investigated changes in nutrient Nutrient Database and Validity of a Food Frequency Questionnaire (FFQ) ous database), which was published in 1982, included values for energy, protein, fat, carbohydrate, sodium, potassium, calcium, phosphorus, iron, retinol, carotene, vitamin B1, vitamin B2, niacin, and vitamin C of 1621 food items. Continuously thereafter, values for amino acids, fatty acids, cholesterol, vitamin E, magnesium, zinc, copper, dietary fiber, and vitamins D, K, B6, and B12 were published, but only for some major food items, rather than all 1621 food items. The various databases were integrated in the revised database, published in 2000, which also included a greater variety of food items (1,878 foods). This database provided food composition values for some nutrients which were not presented in the previous database, such as retinol equivalents, betacarotene equivalents, cryptoxanthin, pantothenic acid, and NaCl deducted from sodium content. It also provided food composition values for all 1,878 food items for those nutrients for which values were only available for some foods in the previous database, such as magnesium, zinc, copper, vitamins D, E, K, B6, and B12, and folate. For all food item codes in the previous database that appeared in the DR and FFQs, equivalent food item codes in the revised database were assigned. When an exactly equivalent food item was not available, an alternative item of close botanical or zoological relevance was taken as a surrogate.

Statistical Analysis
The mean intakes of energy and nutrients according to the FFQs were calculated by sex for Cohorts I and II using the previous and revised databases. Intake levels based on the revised database were compared with those based on the previous database by means of mean difference (in which intake calculated with the previous database is subtracted from that with revised database), and percentage of changes (in which mean difference is divided by intake calculated using the previous database). Statistical differences between intake levels based on the two databases were tested by Student's paired t-tests.
Validity of the FFQ in the estimation of crude and energyadjusted intake (residual method) was evaluated by Spearman's rank correlation coefficients (CCs) using mean intake from the 28-day DR and biomarkers as references. In addition, reproducibility of the FFQ for the estimation of crude and energyadjusted intake (residual method) was evaluated by the Spearman's rank CCs between intake levels according to the two FFQs administered at different times. These CCs were compared to the respective CCs calculated using the previous database using the point estimate and its 95% confidence interval of each CC. All analyses were performed using SAS ® Version 9.1 (SAS Institute Inc., Cary, NC).
Mean nutrient intakes by the FFQ calculated using the previous and revised databases are shown in Table 1. Differences in estimated intake as a result of the revision were not particularly apparent for macronutrients, but were more apparent for micronu-   Female Male intake estimates in Japan, 5,6 with the exception of the estimated intake of fatty acids and cholesterol, which showed a drastic decrease in one study but no decrease or radical change in the present study. This difference is likely due to our supplementation of missing values in the food composition database. Influence on nutrient intake did not differ among age groups because, unlike the previous study, the age range of our subjects did not include subjects aged below 45 years. 6 In addition, changes in intake by database revision were also computed for nutrient intake according to the DRs (data not shown). The percentages of differences between nutrient intake calculated using the two databases according to the DR were closely similar to those assessed by the FFQ; in other words, the degree of over-or underestimation of intake by the FFQ was not modified by revision of the database.
Validity levels of the FFQ were moderate to high for the estimation of energy and of most nutrient intakes. These levels were not changed by revision of the food composition table in subjects of either Cohort I or II. Similar results between the two cohorts suggested the possibility that the results could be generalized; that is, revision of the nutrient database might not have affected the validity of the FFQ as assessed in an external population.
In general, DRs provide the best available comparison method, 1 and are often used as the gold standard in validation studies of FFQs. However, nutrient intake calculated using an FFQ is not completely independent from that using a DR because the same food composition table is used to calculate nutrient intakes for both methods. The present results therefore appear unsurprising, given that the reference method was also calculated using the revised food composition tables. To compensate for this limitation, validity was also tested using biomarkers as references, which are totally independent of dietary assessments. Results for these also indicated that validity was only little influenced by revision of the nutrient database. Validity of the estimation of fatty acids, B group vitamins, and vitamin C was markedly low when biomarkers were used as references with either database, however, because biomarkers are not good indicators of the longterm habitual intake of these nutrients. 15,16 The second limitation of this study was that we did not conduct equivalence testing for two correlation coefficients. Although this is required to show equivalence, it is generally not done because of the complexity of estimating variance components and constituting a confidence interval from the statistics for the ratio and difference between two correlation coefficients. Comparison using the point estimates and confidence intervals of each correlation coefficient revealed relatively low differences for each, and on this albeit informal basis we evaluated the two correlation coefficients as being similar.
In conclusion, the validity of the FFQ used in the JPHC Study to estimate nutrient intake was not influenced by revision of the Standard Tables of Food Composition in Japan. Associations between disease and nutrients would therefore be consistent between the databases as long as nutrient intake was used for ranking.