Validity and Reproducibility of a Self-administered Food Frequency Questionnaire in the JPHC Study Cohort II: Study Design, Participant Profile and Results in Comparison with Cohort I

The objective of this study was to evaluate the validity and reproducibility of a self-administered food frequency questionnaire (FFQ) to estimate nutrient and food intake in the subjects of the Japan Public Health Center-based prospective Study on Cancer and Cardiovascular Diseases (JPHC Study Cohort II). The FFQ was originally developed to estimate intake in the JPHC Study Cohort I. A total of 392 subjects were recruited from the entire cohort participants in the 6 areas of Cohort II on a voluntary basis. The subjects completed the FFQ used for the 5-year follow-up survey twice at approximately a 1-year interval. Seven-day dietary records (DR) and blood samples were collected 4 times at 3-month intervals over a year. Daily nutrient and food intakes from FFQ and DR were estimated. The Spearman correlation coefficients for estimated intakes were calculated between FFQ and DR for validity, and between 2 identical FFQs for reproducibility. Correlation coefficients for the validity ranged from 0.09 to 0.82 among various nutrients and food groups. The correlation coefficients for most of the nutrients and food groups were improved to a level comparable to that of Cohort I by energy-adjustment. Correlation coefficients for reproducibility ranged from 0.42 to 0.82, similar to those of Cohort I.

The same FFQ was used for the 5-year follow-up survey in Cohort II in 1998.To examine the validity of the FFQ in the external population, a validation study was conducted among the subsample of 6 areas in Cohort II.The objective of this study was to evaluate validity and reproducibility of the FFQ to estimate nutrient and food intakes in the JPHC Cohort II.

Study Subjects
The subjects of the validation study were a subsample of the participants in the JPHC Study Cohort II.The location of the study area is shown in Figure 1.This study was designed according to the study design of the JPHC Cohort I validation study for the FFQ. 3 The detailed study design and participants of the entire cohort were reported elsewhere.4   married couples, 60 from Iwase and Tomobe Towns, Mito PHC area; 60 from Oguni Town, Kashiwazaki PHC area; 76 from Noichi and Kagami Towns, Chuo-higashi PHC area; 66 from Shin-uonome, Arikawa and Kamigoto Towns, Kamigoto PHC area; 66 from Gusukube Town and Hirara City, Miyako PHC area; and 64 from Suita City, Suita PHC area) were recruited on a voluntary basis.Subjects were without any diet restriction, and all of them provided written informed consent.
Although our attempt was to select the subjects from as many cohort participants as possible, some spouses (5 from Kamigoto, 2 from Miyako) were not in the age range of Cohort II participants because we recruited married couples as a unit for the subjects.From Suita PHC area, 30 subjects that included 1 married couple were not cohort participants because the cohort from the Suita PHC area was a random sample not covering the entire population.Among the 30 subjects in Suita, only one was not within the age-range of Cohort II participants (40 to 69 years old at the baseline year).
For the analysis in this report, we excluded 7 non-cohort participants from the Kamigoto and Miyako PHC areas.Because 29 non-cohort participants from the Suita PHC area were possible candidates for the random sample, we excluded only one subject whose age was not within the range of Cohort II participants.For the validity analysis, we also excluded 34 subjects who did not complete the DR for 28 days and/or the second FFQ (FFQ_V).For the reproducibility analysis, we excluded 61 subjects without the first FFQ (FFQ_R) from subjects in the validity analysis (Table 1, 2).

Data Collection
The sequence of data collection was shown in Figure 2.

Dietary Records (DR)
The subjects provided 7-day dietary records (DR) in 4 seasons (a total of 28 days): spring (May), summer (August), autumn (November) and winter (February).In Mito the PHC area, the study was launched in the spring of 1996.Half of the subjects from Chuo-higashi (n=32) joined the study in the summer of 1996, and the other half (n=44) in the winter of 1997.In other areas, the study began in winter of 1997.
Weighed DRs were collected over 7 consecutive days in each of the 4 seasons.Dietitians from the PHC, the cities or towns in each area instructed the subjects to weigh all foods and beverages using the measuring spoons, cups and an electronic scale provided, and to record them in a booklet especially designed for the purpose.The subjects gave detailed descriptions of each food, the method of preparation and names of the recipes.The dietitians checked the records at subjects' homes at least once during the survey.
At the end of each season, the dietitians from the PHC reviewed the records in a standardized way, and coded all the foods recorded according to the Standardized Tables of Food Composition, 4th edition.5If codes were not available for certain local foods, the dietitians substituted the food considered to be most similar by asking subjects for details on the food.When ingredients were not obtained for any already prepared recipes, the standard recipes developed by the authors were used.Nutrient and food calculation was done by the method used in the Cohort I validation study.6 The mean daily intake of energy and 16 nutrients was calculated from the records using the Standardized Tables of Food Composition, 4th edition.5For cholesterol, and additional nutrients and compounds such as fatty acids (saturated, monounsaturated, n-3 polyunsaturated, n-6 polyunsaturated)7, dietary fiber (water-soluble, -insoluble),8 sele-nium9 and carotenoids (alpha-carotene, beta-carotene, lycopene),10 the original food composition tables were developed by filling in the missing values for the Japanese composition tables.For isoflavones (daidzein and genistein), the values in the specially developed food composition table for isoflavones in Japanese foods were used.11,12 Additional information about the diet, the standard portions/units for rice and green tea, and brand names for usually used cooking oil, sugar, soy sauce and miso (fermented soybeans) were reported.The frequency of eating out and dietary supplement use for the week was also recorded.Name, age, sex and occupation of all members in the family, self-reported physical activity level, and the number of steps counted by pedometer for one arbitrary day in each season were reported for information on demographic status and physical activity.
Food Frequency Questionnaire (FFQ) Subjects filled in two types of self-administered FFQs used for surveys in the JPHC Study; the FFQ used for the baseline survey (FFQ_B) and the semi-quantitative FFQ used for the 5-year follow-up survey.The latter was repeated twice.The first FFQ (FFQ_R) was to compare with the second FFQ for the evaluation of reproducibility, and the second FFQ (FFQ_V) was to compare with the dietary records for the evaluation of validity.The FFQ_Rs were completed at the same time as the first DR in all but the Chuo-higashi PHC area, where no FFQ_Rs were completed.The FFQ_Vs were completed after a year from the FFQ_R (3 months after completion of last DR) except for the Mito PHC area, where it was nine months after the FFQ_R (same time as last DR).In the Chuo-higashi PHC area, the FFQ was completed in February of 1998.

Blood
A total of 22 ml of blood (10 ml in a heparinized tube and the rest in tubes without anticoagulant) was collected by venipuncture from all participants during or after the 7-day in DR each season .
All tubes were centrifuged for 10 minutes at 3,500-4,000 r .p. m. within the day of blood collection to obtain plasma, red blood cells and a buffy coat from the heparinized tube, and serum from the other tubes.The plasma, red blood cells, buffy coat and serum were divided into several tubes and frozen in the freezer at each PHC.After the temperature reached -80 , the tubes , placed in several small boxes, were packed with sufficient dry ice and sent to the National Cancer Center (NCC) by temperature-controlled deliveries.Boxes were immediately put in the freezer at the NCC and stored at -80 .One of the tubes with serum of each subject was brought to the laboratory of SRL, Inc. (Tokyo, Japan) for the analysis of HDL-cholesterol, total protein, albumin-globulin ratio, total cholesterol, triglyceride, uric acid, gamma-GTP, GOT (AST) and GPT (ALT).
At the time of blood collection, subjects were asked about their health status, medication, and whether they had consumed any foods or beverages within 5 hours.Female subjects were asked about their most recent menstrual period or their age of menopause.Height, weight and blood pressure of the subjects were measured.

Statistical Analysis
Means and standard deviations of energy, nutrient and food intakes from DRs and FFQ_Vs were calculated by sex.Percent differences between DR and FFQ_V for energy, each nutrient and food were calculated by dividing the difference in mean intake from DR and FFQ by the DR mean, i.e., (FFQ mean-DR mean)/DR mean.Correlation coefficients were calculated using both crude and energy-adjusted values.Nutrient and food group intakes were adjusted for energy using the residual model.
For validity, the correlation coefficients of estimated nutrient and food intakes between DRs and FFQ_Vs were computed.For reproducibility, the correlation coefficient of estimated nutrient and food intakes between FFQ_Rs and FFQ_Vs were computed.Spearman rank correlation coefficients were used for both validity and reproducibility.

Profiles of Subjects Completing DRs
The background profile of subjects who completed DRs for 4 seasons and FFQ _V was shown in Table 3.The average age of the subjects was 58 years in men and 55 years in women.The average body mass index (BMI) was 23.7 for men and 23.7 in women.The most frequent occupation indicated was "employee" in men and "housewives" in women, followed by agricultural workers in men and employed workers in women.

Mean Nutrient and Food Intake and FFQ Validity
Daily intakes of energy and nutrients assessed by DR and FFQ and their correlation coefficients were shown in Table 4.The percent difference in nutrient intake between DR and FFQ varied from -33% for selenium to +53% for retinol in men, and from -65% for alcohol to +71% for retinol in women .The correlation coefficients of the crude values varied from 0 .17for polyunsaturated fatty acid (PUFA) to 0.82 for alcohol in men , and 0.21 for n-3 PUFA to 0.52 for alpha-carotene in women.Medians were 0 .37 in men and 0.40 in women.After the energy adjustment , the correlation coefficients varied from 0.26 for selenium to 0 .65 for calcium in men, and 0.18 for selenium to 0.64 for calcium in women.Medians were 0.49 in men and 0.45 in women.
Daily food intakes by food groups assessed by DR and FFQ and their correlation coefficients were shown in Table 5.The percent difference between DR and FFQ varied from -78% for seasoning and spices to +48% for non-alcoholic beverages in men, and from -75% to +60% for the same food groups in women.The correlation coefficients of the crude values varied from 0.10 for fungi and algae to 0.77 for alcoholic beverages in men, and 0.09 for fungi to 0.64 for milk and dairy products in women.Medians were 0.35 in men and 0.30 in women.After the energy adjustment, the correlation coefficients varied from 0.11 for algae to 0.69 for milk and dairy products in men, and 0.12 for fungi to 0.65 for pickled vegetables in women.Medians were 0.41 in men and 0.30 in women.

Reproducibility
Reproducibility of nutrient intake estimated by 2 identical FFQs (FFQ_R and FFQ_V) was shown in Table 6.The correlation coefficient of the crude values varied from 0.49 for lycopene to 0.82 for alcohol in men, and from 0.54 for carbohydrate and genistein to 0.70 for alcohol in women.Medians were 0.61 in men and 0.63 in women.After the energy adjustment, the correlation coefficient varied from 0.46 for alpha-carotene and lycopene to 0.77 for alcohol in men, and from 0.33 for MUFA to 0.72 for alcohol in women.Medians were 0.56 in men and 0.51 in women.Reproducibility was highest for alcohol intake among all the nutrients.
Reproducibility of food intakes estimated by 2 identical FFQs (FFQ_R and FFQ_V) were shown in Table 7.The correlation coefficient of the crude values varied from 0.44 for nuts and seeds to 0.78 for alcoholic beverages in men, and from 0.42 for cereals to 0.74 for milk and dairy products in women.After the energy adjustment, the correlation coefficient varied from 0.38 for algae to 0.70 for alcoholic beverages in men, and 0.40 for fish and shellfish and algae to 0.80 for alcoholic beverages in women.

DISCUSSION
In this study, we evaluated the validity and reproducibility of the FFQ in the external population for which the FFQ was originally developed.From 392 subjects, we collected 2 types of FFQ; the FFQ used for the baseline survey (FFQ_B) and two identical FFQs used for the 5-year follow-up survey (FFQ_R and FFQ_V).We also collected 7-day DRs and blood samples from each of the 4 seasons.In this report, we estimated daily nutrient and food intakes from the DR, FFQ_R and FFQ_V.The correlation coeffi-  1 Nutrient intakes were adjusted for energy intake by residual model. 2 Saturated fatty acid 3Monounsaturated fatty acid 4Polyunsaturated fatty acid.5 Median of correlation coefficients for crude intakes of energy and 31 nutrients, and for energy-adjusted intake of 31 nutrients.1 Food intakes were adjusted for energy intake by residual model.
2 Pickled plum (umeboshi) was included in pickled vegetables , and not in total vegetables but rather in fruits.3 Median of correlation coefficients for 19 food groups.cients of nutrient and food group intakes between DR and FFQ_V were computed to evaluate validity of the FFQ.The correlation coefficients of nutrient intake between FFQ_R and FFQ_V were computed to evaluate reproducibility of the FFQ.
The validity of the FFQ to estimate energy and nutrient intake in Cohort II was slightly lower for most of the nutrients compared to Cohort I.13 Correlation coefficients for energy and 16 crude nutrient intakes from this study were compared to those from the validation study of Cohort I in Figure 3. Except for alcohol in men, the correlation coefficients for each nutrient in Cohort II did not agree with the corresponding nutrients in Cohort I (r=0.30for men and 0.11 for women after excluding alcohol).However, the validity for estimated nutrients was improved to a comparable level of Cohort I by energy adjustment, except for alcohol in men (Figure 4).The correlation coefficients became similar for the same nutrients between Cohort I and Cohort II (r=0.61 for men and 0.57 for women after excluding alcohol).We assumed that the error associated with the estimation of dietary intake was offset by adjusting nutrient intake by total energy intake.The alcohol intake in men, however, might not have included the same error as the others because alcohol consumption did not depend on diet in men.
The validity of the FFQ to estimate food intake in Cohort II was slightly lower than that in the Cohort I for most of the food groups.14Correlation coefficients for food group intakes (crude values) from this study were compared to those from the validation study of Cohort I in Figure 5. Food groups with low correlation coefficients in Cohort I such as potatoes and starches , fats and oils, algae, and seasonings and spices were also low in Cohort II, while food groups with high correlation coefficients in Cohort I such as milk and dairy products , and pickled vegetables were also high in Cohort II.Correlation coefficients for most of the food groups improved after the energy adjustment , especially in men (Figure 6).
Reproducibility of the FFQ to estimate nutrient and food intakes in Cohort II was similar to that in Cohort I (0 .41-0.83 in men and 0.52-0.77 in women for crude nutrient intake , 0.42-0.80 in men and 0.45-0.74 in women for crude food intake) .15 The correlation coefficients were slightly lower after energy adjustment , reflecting the same tendency in Cohort I .We initially speculated that there would be lower reproducibility in Ibaraki because of the FFQ interval of 9 months, against a year in other areas .However, the results of subgroup analysis did not show any difference between Ibaraki and other areas.This might be a result of the design of our FFQs which assess the usual dietary intake over a year by asking subjects their highest seasonal consumption of certain foods in light of the high seasonal variety.The result contradicted the result from a previous study which indicated lower reproducibility when FFQs were repeated in different seasons compared to when repeated at a year's interval.16 The reproducibility of our FFQ was comparable to findings in the literature in terms of the dietary intake estimated by the FFQ.Correlation coefficients between repeated measurements for estimating dietary intake are usually on the order of 0.5-0.7.17This level of reproducibility is comparable to that of many biological measurements such as serum cholesterol and blood pressure, which are strong and consistent predictors of disease in epidemio-logic studies 17 Except for lycopene, nuts and seeds, algae in men and cereal in women, the reproducibility of our FFQ for estimating intake of nutrients and food groups was r>0.5, which is sufficiently reasonable for epidemiologic use.This reasonable degree of reproducibility ensures that our FFQ can repeatedly assess the intake of individuals over a year.Given its reasonable validity, we can be sure that our FFQ at one point in time could obtain the usual intake of individuals over the period of a year.
In the present report, we did not discuss the validity of the FFQ in relation to biochemical markers because the ongoing analysis is still in the early stages.Those biochemical indicators, however, are an important resource for comparison of dietary intake when used in conjunction with DR17 for further analysis.Biochemical measurements such as blood levels of vitamin C, carotenoid, phospholipid, selenium, folate, vitamin B6, vitamin B12 and isoflavones have been used for analysis to validate the FFQ in Cohort 8,18-21 Analysis for validation and the seasonal variation using these biochemical measurements and dietary data of DR from each of the seasons is currently in progress in Cohort II.
In conclusion, the validity of our FFQ for use with Cohort II was comparable to the validity in Cohort I after the energy adjustment.Reproducibility of the FFQ in Cohort II was similar to the reproducibility in Cohort I.Both validity and reproducibility were at a reasonable level, showing that the FFQ was accurately estimating dietary intake of individuals for epidemiologic use.The results of our study thus indicated the possible use of FFQs in the external population for which it was originally developed .

Figure 1 .
Figure 1.Location map for six areas of JPHC Study Cohort II.

Figure 2 .
Figure 2. Sequence of data collection for FFQ validation study of JPHC Study Cohort II.

Table 4 .
Nutrient intakes calculated by DR for 28 days and FFQ_V used for 5 -year survey of JPHC Study Cohort II and their correlations 1 (FFQ mean -DR mean)/DR mean.2Nutrient intakes were adjusted for energy intake by residual model. of energy and 31 nutrients , and for energy-adjusted intake of 31 nutrients.Table 5. Food intakes calculated by DR for 28 days and FFQ2 used for 5-year survey of JPHC Study Cohort II and their correlations 1 (FFQ mean -DR mean)/DR mean. 2 Food intakes were adjusted for energy intake by residual model.3 Pickled plum (umeboshi) was included in pickled vegetables, and not in total vegetables but rather in fruits .4 Median of correlation coefficients for 19 food groups.

Figure 3 .Figure 4 .
Figure 3.Comparison of correlation coefficients for selected nutrients (crude) between Cohort I and II.

Figure 5 . 11 Figure 6 .
Figure 5.Comparison of correlation coefficients for selected food groups (crude) between Cohort I and II.

Table 6 .
Spearman rank correlation coefficients between nutrient intakes assessed with two FFQs one year apart (JPHC Study Cohort II)

Table 7 .
Spearman rank correlation coefficients between food intakes assessed with two FFQs one year apart (JPHC Study Cohort II)