A data-based approach for designing a semiquantitative food frequency questionnaire for a population-based prospective study in Japan.

A self-administered food frequency questionnaire (FFQ) has been used commonly in epidemiologic studies of diet and chronic diseases. The analysis of dietary data in the target population is useful for designing a new questionnaire. The authors studied the major food sources of nutrient intakes in the Japanese population in order to develop a FFQ for a population-based prospective study. Subjects were 180 men aged 40-49 years and their 155 spouses who were sampled from four Public Health Center districts. Weighed food records over three consecutive weekdays were collected during the winters of 1989-1991. All foods reported were grouped into 154 items, and the percent contribution of each food to the total population intake of 15 nutrients was computed. Based on these data, a FFQ with 138 items was developed. The food list covered well the population intake of most nutrients (median = 83%, range = 55-91%), except for lipid (70%) and sodium (55%). The questionnaire was applied to the prospective study conducted in the same areas where the dietary data were collected, and an investigation to assess its reproducibility and validity is currently underway.

Large-scale prospective studies of diet and chronic diseases have been conducted increasingly throughout the world1). A key issue in planning these studies is the development of a dietary assessment method that is inexpensive and accurate, and a self-administered food frequency questionnaire (FFQ) has been the most common choice. A number of questionnaires have been developed which could not only evaluate the average long-term frequency of individual food intakes but also provide semiquantitative estimates of nutrient intakes, and the reproducibility and validity of their nutrient estimates have been examined2-6).
The most important process in designing a new questionnaire is the selection of food items to cover a large proportion of foods and nutrients consumed by the target subjects. Although judgment of nutritional experts based on experience is no doubt useful, quantitative data on diet in the study population would further help the selection of adequate food items7).
In Japan, several large-scale prospective studies of cancer have been ongoing8, 9). There have also been a few attempts to develop FFQ for the Japanese populations 10.11) However, none of the prospective studies have used the dietary questionnaires that are based on the quantitative analysis of diets in the target populations. We have recently developed a FFQ in which dietary data drawn from the random samples of target subjects were used to select food items, and applied it for a populationbased prospective study. This paper reports the major food sources of nutrient intakes in these populations and the procedures for questionnaire development.

Study design and subjects
The subjects were participants in a cross-sectional study conducted in five Public Health Center districts in Japan with varying rates of mortality from gastric cancer. Details of the study were described elsewhere12,13). Briefly, in the winters of 1989-1991, 880 men aged 40-49 years were sampled randomly from the five areas and asked to participate. The response rate was 72% (n=634). Questionnaire interview, anthropometric measurements, and blood collection were conducted for all subjects. The dietary survey was conducted among a subgroup of the participants and their spouses (207 men and 183 spouses). Since diets in the area of Okinawa island differed substantially from those in the other four areas in mainland Japan, data from Okinawa were analyzed separately. This study only reports the results of analysis for the subjects in the four mainland areas (180 men and 155 spouses).

Food records
Weighed food records over three consecutive weekdays were collected by a method used in the National Nutrition Survey14) with some modifications. Four research dietitians instructed the subjects to record all foods and beverages prepared and consumed in a specially designed booklet. The participants were asked to provide detailed descriptions of each food, including the method of preparation and recipes whenever possible. A dietetic scale was provided for weighing food servings. For all foods eaten at home, the weights prepared and the proportions consumed by each man and the spouse were specified (e.g., 100g of rice was boiled and the man and the spouse consumed 50% and 30% of them, respectively). The dietitians checked the records at the participants' home during the survey, and reviewed them in a standardized way after completion.
Nutrient intakes were computed from these records using the Standard Food Composition Tables published by the Science and Technology Agency of Japan15). Foods were originally coded into 1624 different codes based on the classification of the Standard Food Composition Tables. They were combined into 154 food items for this analysis, each representing several food codes. For instance, 89 different codes originally assigned for various type of beef were grouped into a single category as "beef ".

Calculation of contribution rates of food items for the total population intake of nutrients
The percent contribution of each food to the total population intake of 15 nutrient variables was computed according to the procedures described by Block  The selection of food items for the questionnaire The selection was based on percent contribution of a food to the consumption of 15 nutrients, trying to cover 80% of the total population intake of each nutrient. Although beef, pork and chicken (other than livers) were classified into single categories to calculate the contribution rates, they were listed as separate items based on the methods of cooking (such as steak, roasted and stew for beef). The food contributions were also computed using the data from Okinawa area, and several foods with substantial contribution to nutrient intake of the district were added. Some items with potential interest (for instance, oolong tea, black tea, and beta carotene beverages) were also included regardless of their contribution rates.
Preliminary versions of the questionnaire were developed twice and tested for wording of questions and adequacy of the food list in February and August, 1994. These pilot studies revealed that, although two preliminary questionnaires asked the average intake frequencies during the previous year, responses on fruits and vegetables differed substantially between the two questionnaires reflecting the seasonal differences in availability of these foods. The time frame for fruits and vegetables was therefore modified as average frequencies during the seasons when they were available in the market, and several items with large seasonal variation in consumption were added to the food list. Table 1 presents the contribution rates of 154 food items for the total population intake of 15 nutrient variables among the subjects in four mainland areas of Japan. With regards to macronutrients, rice was the largest source of energy, followed by vegetable oils, pork, egg, and breads. For protein, animal products and vegetable sources consisted of 53.7% and 46.3% of total intake, respectively. The major contributors of protein were rice, pork, egg, chicken and salted fish (pollack, salmon and atka mackerel). For dietary lipid, animal products and other sources consisted of 40.9% and 59.1% of the total intake, respectively. Lipid was consumed mainly from vegetable oils, pork, egg, milk and rice. More than half of carbohydrates came from rice, followed by breads, Chinese noodles, sugar, and wheat flour.

RESULTS
Based on these data, a self-administered FFQ was developed for a population-based prospective study conducted in the same Public Health Center districts where the dietary data were collected. As listed in Table 2, the developed questionnaire included a total of 138 food items and 14 supplementary questions on several aspects of diet. Respondents were asked about the average frequency of consumption of each food during the previous year. For fruits and vegetables, the frequencies during the seasons when they were available in the market was queried. Nine frequency categories ranged from not eating to consuming seven or more times per day. Standard portion size at one serving was specified for each food and respondents were asked to choose if their usual portion was larger (more than 1.5 times), the same, or smaller (less than half) relative to the specified standard. The standard portion was determined primarily from the observed median values. For 18 non-alcohol beverages, only frequency was questioned. Intakes of rice, bean paste (miso) soup, and 5 alcohol beverages were asked in different formats with more detailed questions.
As shown in Table 3, the developed food list covered the population intake of 15 nutrients generally well (median=83%, range=55-91%). However, it covered only 70% of lipid and 55% of sodium consumption.

DISCUSSION
Investigation of the major food sources of nutrient consumption is useful to understand dietary patterns of a population and to develop a food frequency questionnaire to be applied to them. These data have been reported from a nationally representative sample in the U.S.16.17) Hispanic-American women and their children18), U.S. women 19 through 50 years old"), and elderly Italians20,21). In Japan, the Reports of the National Nutrition Survey have continuously reported the contribution rates of foods to the consumption of selected nutrients using the nationwide samples 14). Although the contributions of foods were generally comparable between this study and the national survey, several differences were noted. For instance, the contribution of bread to energy intake was lower in this study than in the national survey (2.9% vs 4.2%), while the contribution of bean paste (miso) to sodium consumption was higher (20.4% vs 12.9%). These differences imply that our subjects maintain the traditional Japanese diet to a greater extent than do the subjects in the national survey.
The developed food list generally covered the large proportion of the nutrient consumption by the subjects. However, the coverage for lipid and sodium was relatively lower. Vegetable oils used for cooking were the largest source of lipid intake, and seasonings such as soy sauce and salts were the major contributors of sodium consumption. Since these items could not be listed in the same manner as other foods by asking simply the frequencies and usual portions, substantial proportions of lipid and sodium intake were uncovered by the food list. We therefore supplemented several questions that focused on the use of cooking oils, soy sauce and salts, such as a type of cooking oil most frequently used, consumption frequency of fried foods, frequency of soy sauce use at table. If we assume that the supplemented questions could fully cover the consumption of vegetable oils, soy saurce and salts, then the questionnaire would cover up to 86% of lipids and 85% of sodium consumption. The incorporation of these items to the regular food list should therefore increase the capability of questionnaire in estimating lipid and sodium intakes.
Several issues on our procedures of the questionnaire development deserve comments. First, since the dietary survey was conducted in winter, contributions of foods were underestimated if they were consumed mainly in other seasons. This would particularly be the case with fruits and vegetables, since their consumption varied substantially among seasons in Japan. We therefore included some fruits and vegetables (such as pumpkin, pears, grapes, melons) based on the judgment that they might be important sources of several nutrients, although they were rarely consumed and made no substantial contributions in the observed data.
Second, although we listed the standard portion of each food and asked about the usual portion size, the utility of such question has been controversial 22). Addition of portion size items automatically doubled the number of questions for the food list and imposes greater burden on respondents. However, studies in western populations were inconsistent if such addition could improve the validity of nutrient estimates 7.23,24) Since the importance of portion size information relative to frequency information may be culturally based1) and no data have been available to date from the Japanese population, further studies addressing this issue are necessary Finally, the selection of a food list that is based on quantitative data from the target population does not necessarily imply validity of the questionnaire. An independent study is required to compare nutrient intakes estimated from the questionnaire and those calculated from more extensive method of dietary assessment (such as food records). Such an investigation is currently underway and a future study will report the validity and reproducibility of the questionnaire developed by the authors.

ACKNOWLEDGMENTS
This study was supported in part by Grants-in-Aid for Cancer Research from the Ministry of Health and Welfare, for Scientific Research from the Ministry of Education, Science and Culture, and from Daiwa Health Foundation. The authors are grateful to Dr. Walter C. Willett for reviewing the manuscript. We also appreciate the assistance from the Japan Population-based Prospective Study Group.