Self-administered Food Frequency Questionnaire Used in the 5-year Follow-up Survey of the JPHC Study: Questionnaire Structure, Computation Algorithms, and Area-based Mean Intake

In this section we described the structure of the self-administered semiquantitative food frequency questionnaire used in the 5-year follow-up survey of the JPHC study, the computation algorithms, and the area-based mean intakes of nutrients and food groups in the subjects of the validation study. The FFQ consists of five sections: 1) semiquantitative frequency questions for rice and miso (fermented soybean paste)-soup, 2) those for alcoholic beverages, 3) those for vitamin supplements, 4) those for foods and beverages, and 5) questions on dietary and cooking behaviors. From the questions, intakes of nutrients and foods by food groups were computed. Although most of them were computed from the frequency and relative portion size indicated in the replies, together with the fixed portion size, a seasonal coefficient was added in the computation of vegetables and fruits. Only frequency of intake and fixed portion size were used for computation of beverages. Sugar and cream added in coffee and tea were computed from the frequency of coffee and tea intake. The intakes of cooking oil, cooking salt (sodium), and salt in noodle-soup were estimated from the questions of relative preference of oil, salt, and noodle-soup.

Self-administered semiquantitative food frequency questionnaires (FFQ) have been used to assess nutrient and food intake levels of the subjects in large-scale epidemiologic studies from several countries. 1 The development of a FFQ most suited to the study objectives and subject characteristics is the key to obtaining fruitful results in a cohort study when this type of questionnaire is used for the assessment of dietary habits. The JPHC study group has developed a FFQ based on a data-based approach using 3-day dietary record data obtained from 335 subjects sampled from the 4 areas in the JPHC study, Cohort I. 2 In this section, the questionnaire design of the FFQ and the computation algorithm are briefly presented together with the intake levels of main nutrients and food groups among the subjects of the validation study. In the first section, questions on the relative size of the ricebowl were asked in terms of 3 categories; small, medium-size, and large. The intake of rice was asked in terms of the number of bowls consumed, from less than 1, to 1, 2, 3, 4, 5, 6, 7-9, or 10 per day. Two additional questions about rice were asked: "Do you eat vitamin-enriched rice?", and "Do you mix in other grains such as barley, foxtail millet, or Japanese barnyard millet?". For misosoup, the frequency of the intake was asked from almost never to 1-3 times/month, 1-2 times/week, 3-4 times/week, 5-6 times/week, or daily. The intake amount was asked in terms of the number of bowls from less than 1, to 1, 2, 3, 4, 5, 6, 7-9, to 10 per day. The relative salt content was also asked in terms of less salty, medium-salty, and salty.
In the second section, frequency of alcohol drinking was asked using 6 categories from almost never to 1-3 times/month, 1-2 times/week, 5-6 times/week, once/day, 2-3 times/day, 4-6 times/day, to > 7 times/day. The beverage intake frequency was from 1-2 times/week, 3-4 times/week, 5-6 times/week, to one cup/day, 2-3 cups/day, 4-6 cups/day, 7-9 cups/day, or > 10 cups/day. The standard portions/units for each food were stated for 108 foods, and both in writing and in photo form for 5 vegetables. The relative portions/units used were from small (50% smaller) to medium and large (50% larger). Questions on the amount of sugar and milk used for tea and coffee were asked by 5 categories; none, half a spoon, one spoonful, two spoonfuls, and 3 spoonfuls for tea and coffee separately.
In the fifth section on dietary and cooking behaviors, 13 questions were asked. These questions are described elsewhere in this supplement.2

Computation of Food and Nutrient Intakes
Intakes of rice and miso from miso-soup were calculated from bowl size, frequency, and the number of bowls consumed per day. Standard bowl sizes were 140 grams for rice and 150 grams for miso-soup for both sexes. The portion of rice for a small bowl and a large bowl was 110 and 170 grams, respectively. In the category of rice, rice boiled with cooking water was used for food and nutrient computations, while for miso-soup, miso and cooking water were used. Cooking water was not considered in other food computations.  Table 1). The composition values of 147 foods were multiplied by the frequencies and the relative portion sizes for the food items from the FFQ. For 18 beverages without unit volume questions, the standard unit was used for the computation. The coefficients for the categories of relative portion sizes were 0.5 for small, 1.0 for medium, and 1.5 for large. For frequencies, the median frequency was used, for example, 2.5 times per day instead of 2-3 times per day. For the most extreme category, slightly smaller and larger values were used than the limit for the lowest and highest categories, respectively (e.g., 8 times/week for 7 times/week). Sugar and milk intakes for tea and coffee were calculated from tea and coffee intake. Of the 147 foods in the composition table, the weighted mean was used for 16 which consisted of 2 or more foods (e.g., salted fish, salmon and trout, other oranges) ( Table 1). The weighting ratios for the composition of those foods were obtained from the DR data in this validation study, although values rounded off to 60% and 40% for food A and B, respectively, were used because of the limited value of the data.
The significant seasonal variation reported for the intake of some nutrients such as vitamin C, is probably due to the seasonal variation in fruits and vegetables.3 For those foods, subjects were asked in the FFQ to report the consumption at the time of year when each food was most available. Seasonal coefficients were used to calculate the average yearly intake of such foods. The coefficients were determined based on the intake reported by the dietary records by season. Area was not considered in this computation. The equation may be stated: SCi = Xi,ave / Xi.max, Where SCi is the seasonal coefficient of foodi , Xi,ave is the mean intake of the study subjects for the whole year average estimated from a 28-day (4 season) DR, and Xi,max is the mean intake of the population for the intake in a 7-day (1 season) DR of the season with the highest consumption. Table 1 indicates the SCi for each food.
The salt intake from cooking salt and salty seasonings such as soy sauce was estimated from the cooking methods most frequently used for meats, fish, and vegetables, the use of table salt and soy sauce, and the intakes of meats , fish, and vegetables. Intake of cooking oil and salt was estimated by the method used in the dietary history: the amounts of cooking oil and salt for 5 cooking methods (raw, stewed, grilled, deep-fried , stir-fried, and other) were estimated for 3 food groups (meats , fish, and vegetables), and multiplied by the individual intake of each food group according to the cooking methods most frequently used by the individual.4 All supplement use and some nutrition-fortified beverages were excluded from the computations in this study beca use their composition tables were not available.
The intakes of 16 nutrients for each food were calculated using the food composition

Statistical Analysis
The subjects of the present study were 215 persons (102 men and 113 women) with whom both the FFQ for the validation study and their complete dietary records (14-day records in Okinawa and 28-day records in the other 3 areas) were used for the analysis. The mean intakes for food groups and nutrients were calculated by sex and area, and compared by ANOVA. The definition of food groups was mainly based on the Standard Tables of Food Composition in Japan, the 4th revised edition.5 Green and yellow vegetables were defined as 44 vegetables with 600 micrograms of carotene per 100 gram and 10 frequently used vegetables that contribute to the intake of carotene among Japanese, according to the definition by the Ministry of Health, Labor and Welfare. Additionally, mugwort (leaves) with 3600 micrograms of carotene per 100-gram food portion was included in the green and yellow vegetables because it could not be overlooked as a carotene source among the subjects in Okinawa in a previous survey.2 Salted pickled vegetables were defined according to the food group used in the National Nutrition Survey.8 In FFQ, sugar intake was assessed only for those for coffee and/or tea. Therefore, sugar from cooking was not included in the analysis, but the energy and nutrients derived from sugar for coffee and/or tea were included in the corresponding computations. Table 1 shows the list of 147 foods used in the computation of intake in the FFQ, their standard portion sizes, and seasonal coefficients. The food codes of the Standard Tables of Food Composition in Japan, the 4th revised edition5 used in the computation and their weighting ratios were also described. Tables 2 and 3 show the intake levels of main nutrients by sex and area. The mean intake of energy was highest in the Ninohe PHC area for men, and in the Saku PHC area for women, and lowest in the Ishikawa PHC area for both men and women. Mean intakes were significantly different among areas for energy and most of the nutrients except alcohol and carotene in both men and women for crude values (p<0.05). After the adjustment for energy intake, the mean intakes were significantly different among areas for protein, total fat, carbohydrate, calcium, phosphorus, sodium, and carotene in both men and women (p<0.05). Tables 4 and 5 show the intake levels of food groups by sex and area. Crude mean intakes were significantly different among areas for food groups except for fats and oils, meats, green and yellow vegetables, algae, and non-alcoholic beverages in both sexes, pulses and alcoholic beverages in men, and milk and dairy products, and vegetables in women (p<0.05). After the adjustment for energy intake, the mean intakes were significantly different among areas for the food groups except for cereals, algae and non-alcoholic beverages in both sexes, pulse, fungi and alcoholic beverages in men, and meat, milk and dairy products and vegeta-bles in women (p<0.05).

DISCUSSION
Although it was nearly impossible to obtain enough data to develop the FFQ and the attached computation algorithm, a databased approach was used to the degree possible in developing the questionnaire. First, the food items were listed in order to cover 90% of total intake of energy and 13 nutrients, except for sodium (cover rate was 50%) using the previous data obtained from the subjects in the study areas. The standard portion sizes, food codes, and their weighting ratios were also basically determined from the data set. The seasonal coefficients were determined from the relative intakes observed in each season in the dietary records of this validation study. We consider this paper to be rather unique in that it presents the questionnaire makeup along with its algorithm for the computation of food intake; nevertheless the validity heavily depends on it. Only food items and the standard portion sizes were described in some reports on the FFQ development (see reference 9).
The most outstanding characteristic of the FFQ may be its use of a seasonal coefficient. Most of the reports from Western countries concluded that seasonal variation in food and nutrient intake is negligible. On the other hand, some Japanese studies suggested possible seasonal variation in intakes of some foods and, as a result, in some nutrients such as oranges, vitamin C and carotene.10 Also in the data of this validation study, a significant seasonal variation was observed for these two nutrients, and their main food sources (i.e., vegetables and fruit).11 Therefore, we asked the "frequency of intake for the most available season" for 48 vegetables and fruit. Then we estimated the intakes over the year using the seasonal coefficients: [intake of the most available season]/[intake over the year] obtained from the dietary records in this validation study. This method of using seasonal coefficients to improve the validity of the FFQ was efficient and feasible in our study because the DR data from 4 seasons were available. However, a more precise evaluation of its efficacy is needed to generalize the method. We also estimated intake of cooking oil and cooking salt, and intake of noodle-soup. They were estimated from relatively simple qualitative, rather than quantitative, questions. Because reliability may be compromised by this type of question (see reference 12), further study is needed to assess the possibility and limitations of these questions for inclusion in the computation of intake levels. In estimating the oil and salt used in food preparation, we used answers to questions regarding dietary behaviors, similar to the method employed in the diet history questionnaire. 4 Cooking oils are major sources of n-6 polyunsaturated fatty acids," and cooking salts including salty seasonings have been major sources of sodium in a Japanese population.14 Therefore, the consideration of these foods seems important when these nutrients are investigated. However, the validity of the method remains to be examined, because the validity for sodium, for example, was relatively low when the 24-hour urinary excretion was used as the gold standard both in the present and previous studies. 15,16 The validity and the reproducibility of the FFQ are reported in the other papers in this Supplement.17,18 1 The food items with portion sizes were indicated as "yes". 2 The definition of food group was based on the Standard Tables of Food Composition in Japan, the fourth revised edition, 3 g=green and yellow vegetable, p=salted and pickled vegetable. Seasonal coefficient used to estimate the average yearly intake year for each food. 4 Food code in the Standard Tables of Food Composition in Japan, the fourth revised edition. 5 Estimated from combination of food intakes and answers to dietary behavior questions.