Validation of a food frequency questionnaire in the Hiroshima/Nagasaki Life Span Study.

We evaluated the performance of a 22-item food frequency questionnaire (FFQ) administered in 1980-81 to 3,005 members of the Adult Health Study cohort, part of the Life Span Study. The questionnaire was compared with the records of a 24-hour dietary survey that was performed in 1984-85. From the dietary records, food and nutrient intakes were estimated. The association between the two measures of dietary intake was assessed using Mantel-Haenszel chi-square test and the Spearman's rank correlation coefficient. The frequency of food intake as measured by the FFQ was linearly associated with food intake as measured by the 24-hour diary, with the exception of dry fish. The highest correlations were observed for beverages, including coffee (0.51), milk (0.32) and black tea (0.26). Foods such as fruit (0.27), confectionery (0.23), rice (0.34) and bread (0.28) were also moderately correlated. These results show that, with the exception of dry fish, the FFQ is moderately correlated with the 24-hour diary and can be used to assess diet intake in this cohort.

The development of methods to measure habitual, long-term dietary intake in epidemiologic studies has been a considerable challenge over the past two decades. The semi-quantitative food frequency questionnaire (FFQ) is the method most commonly used, and usually includes questions on the average frequency of consumption during the past year for a given number of food items. In order to evaluate the performance of a FFQ, dietary intake is often compared with a more objective or detailed measurement, such as biomarkers, an interviewer-assisted 24-hour recall, or self-reported dietary records.
The Radiation Effects Research Foundation in Hiroshima and Nagasaki, Japan is following the Life Span Study cohort, a large cohort of atomic-bomb survivors recruited in 1958.1 Several mail surveys have been carried out to examine lifestyle factors among the Life Span Study cohort members2-4 and a self-administered food-frequency questionnaire based on the frequency of consumption of 22 dietary items (FFQ22) was conducted in 1980-81. Since then, two decades have passed and the number of chronic diseases cases among the respondents has become sufficient to allow statistical analysis. Recently, three studies on diet and can-cer have been published, based on the questionnaire.5-7 However, the validity of the FFQ22 was not considered in these reports, since no study has been specifically designed to measure its per-

Subjects
A self-administered questionnaire including questions on diet (FFQ22), smoking and drinking habits, education, and medical history was sent to all the 55,650 participants of the Life Span Study who were alive as of September 1, 1978 .3 Responses were survey; 6,179 persons (91.6%) accepted and 3,728 persons (60.3%) returned a completed 24-hour diary of food intake. In total, 3,005 participants including 1,133 men (mean age: 60 years old, standard deviation: 11 years, range: 39-89 years old) and 1,872 women (mean age: 57 years old, standard deviation: 7 years, range: 38-80 years old) completed both the FFQ22 and the 24-hour diary, and were available for the present study.

The FFQ22
Dietary questions consisted of 22 food items: beef and pork, chicken, pork products, dairy products, milk, eggs, fish (except broiled or dry fish), broiled fish, dry fish, salted foods (vegetables and fish gut), green-yellow vegetables (such as pumpkin, carrot, or spinach), fruit, seaweeds, confectionery, tofu, miso soup, rice, bread, non-alcoholic fizzy drinks (such as coca-cola, or soda), black tea, coffee, and green tea. Subjects were asked to state how frequently they ate each of these foods over the past year, based on 'never', 'once or less per day', 'twice per day', and 'three times or more per day' for bread and rice, and 'never', 'once or less per day', 'two to four times per day', and 'five times or more per day' for green tea. For the remaining foods, the frequency of consumption was 'never', 'once or less per week', 'two to four times per week', and 'almost everyday'.
The 24-hour diary A trained nurse gave detailed instructions for completing the unstructured 24-hour diary and provided participants with examples and a measuring spoon for weighing each food. Each subject chose a day to record his/her usual average meals (breakfast, midmorning snack, lunch, tea, dinner, after-dinner snack), giving names of dishes and foods, and the amount ingested or drunk at each meal during a 24-hour period. Participants were given a prepaid envelope to return the completed diary. A dietician checked and coded the diaries and contacted the subjects if entries needed clarifying. Nutrient consumption was calculated by multiplying the content of each food, derived from the Fourth Edition of the Japan Food Composition Table,' by the frequency of consumption and according to the portion size, as stated in the 24-hour diary.

Statistical analysis
The food codes from the 24-hour diary were grouped according to the corresponding food items on the FFQ22. The proportion of subjects who recorded whether they had consumed a particular food item in the 24-hour diary was compared with the frequency of consumption as stated in the FFQ22; a test for linear trend was performed using a Mantel-Haenszel chi-square test. The association between the mean intake (g/day) of each food item on the FFQ22 and the 24-hour diary was analyzed using Spearman's rank correlations. Correlation coefficients were also stratified by sex and age (<60 years, 60+ years). In addition, the mean intake of vitamin C and carotene were estimated from the 24-hour diary data and compared with the intake of green-yellow vegetables and fruit as stated in the FFQ22 as these two nutrients are specific to vegetable and fruit intake.

RESULTS
Three thousand and five 24-hour diaries were completed and were equally represented throughout the week and the year, with the exception of Sunday which was slightly less represented (8.2%) than the other days (14-15%). Eighty-two percent of the diaries were completed by the subjects themselves, 13% were completed by the spouse and 5% were completed by the daughter or the daughter-in-law. Table 1 shows the distribution of subjects according to the frequency of consumption as classified by the FFQ22. The majority of subjects reported eating rice twice or more per day and bread once or less per day. More than half reported eating beef/pork, fish and tofu at least 2-4 times per week, and 40-50% ate chicken, pork products, broiled fish, and dry fish once or less per week. Fifty-six percent of subjects consumed dairy products and 37% consumed milk once per week or less, although 75% ate eggs at least 2-4 times per week. Approximately 70-80% of subjects consumed fruit and vegetables at least 2-4 times per week, and nearly half ate salted foods almost daily. Green tea was the most popular beverage, followed by coffee, fizzy drinks and black tea. The proportion of subjects who recorded consuming a particular food item in a given 24-hour period according to each FFQ category is shown in Table 2. There was a clear linear association between whether a food was consumed in the diary and increasing frequency of consumption in the FFQ22, with the exception of dry fish and rice. Moreover, the consistency between the dietary measures was generally greatest at a high frequency of consumption; 80-90% of subjects who stated an almost daily intake of eggs, fish, fruit, green-yellow vegetables and salted foods, tofu, miso soup, coffee and rice recorded consuming these foods during a given 24-hour period.
The mean intake (g/day) of each food item as estimated from the 24-hour diary according to each food-frequency category of the FFQ22 is shown in Table 3. With the exception of dry fish, the mean intake of each food item increased with increasing foodfrequency categories. For example, the mean intake of fruit as estimated from the 24-hour diary increased from 58g/day (which is equivalent to a quarter of an apple or half a banana) in those who recorded they never consumed fruit on the FFQ22, to 189g/day in those who reported eating fruit almost daily. The association between the FFQ22 and the 24-hour diary was strongest for coffee, with a Spearman correlation coefficient of 0.51. Correlations for the other food items ranged between 0.11 for chicken and 0.34 for rice. There was no correlation between the FFQ22 and the 24-hour diary for dry or broiled fish. Table 4 presents the correlation coefficients according to sex and age groups. In general, no difference in correlation coefficient was observed between men and women or between the young and the old age groups. However, a slight difference between men and women for beef and pork, fish, seaweed, fruit, salted foods, and confectionery was observed. Also, a difference between the younger and the older groups was noted for eggs, green-yellow vegetables, salted foods, and confectionery.
The daily mean intake of carotene and vitamin C as estimated from the 24-hour diary according to the frequency of consumption of green-yellow vegetables and fruit as stated on the FFQ22 is shown in tion between the mean carotene and vitamin C intake in relation to green-yellow vegetables and fruit consumption suggests that the FFQ22 can be used as a valid tool in which to measure dietary intake in this population. Dry fish was the only food item on the FFQ22 that was not associated with the 24-hour diary, most likely a reflection of its low consumption in this population, and should therefore be used with caution. Although some participants who stated that they never consumed food items (such as fish, green-yellow vegetables and rice), actually recorded the food in the 24-hour diary; they represented only about 1% for fish and vegetables, and 0.1% for rice (see Tablet).
Furthermore, the "never" consumers had a low intake as estimated from the 24-hour diary. For example, the mean intake of pork products in the "never" category representing one fifth of a bacon slice or sausage; 22g of egg correspond to less than a half of a hen egg; and 58g of fruit is equivalent to a quarter of an apple, or half a banana. With regard to those with a missing frequency of a given food in the FFQ22, their proportion of consumption (Table2) and their mean intake (Table3) were similar to the values of those who answered "never" or "once or less per week or per day". However, the missing values for green tea and rice were comparable to the "2-4 times per day" category, but these subjects with missing data represented only 3% of the study population. It is therefore likely that the participants did not report the intake frequency of a food because they never or scarcely ate this food.
For future research based on the present FFQ22, the "never" and possibly "missing" category would be better grouped with the "once or less per week" , or "once or less per day" category.
The administration of a FFQ is particularly relevant in large prospective studies since it may be self-administered, it is easy to use, and is relatively inexpensive to process." However, FFQs are based on a limited number of food items and rely heavily on the respondents' memory and their interpretation of the questions. A more direct method, such as a 24-hour dietary record, is much more accurate since the intake of foods is recorded in real time and without restrictions, and allows a direct estimation of the portion size and nutrient intake." However, such a method increases the burden on study participants and requires substantially more resources for data management. As there is no gold standard in measuring habitual dietary intakes,12 direct record methods are often used on a small number of subjects to obtain referent data for validating a less accurate method such as a FFQ. [13][14][15] The present study was performed on a large number of subjects and, although we observed modest correlation coefficients for most foods, they are of a similar magnitude to those found in Firstly, the diaries were recorded only for a 24-hour period and may therefore not capture day-to-day variation in food intake,'' leading to possible underestimation of the association." A recent validation study of a FFQ among a Japanese population, based on four 4-day diary records showed higher correlation coefficients than our study. 19 In addition, the 24-hour diaries were completed between three and four years after the FFQ22, thus increasing the possibility that an individual's dietary habit may have changed. However, on a population level, an adult's dietary pattern is thought to remain fairly stable over time,10 and national food consumption and food availability patterns in Japan were relatively constant during the period between the two surveys (1980 and 1984). 23 In conclusion, all food items as measured by the FFQ were linearly associated with food intake as measured by the 24-hour diary, with the exception of dry fish. These findings suggest that the FFQ22 can be used to measure habitual dietary intake in the Life Span Study and supplement the previous reports based on this questionnaire.5-7