Validity and Reproducibility of a Self-administered Food Frequency Questionnaire Used in the Baseline Survey of the JPHC Study Cohort I

We examined the validity and reproducibility of a 44-item food frequency questionnaire used in a baseline survey of the Japan Public Health Center-Based Prospective Study Cohort I conducted in February 1990. Subjects were 94 men and 107 women selected on a voluntary basis among respondents to the baseline survey. Four or five years after the baseline survey, they provided four 7-day diet records during a 1-year period, and then responded to the same questionnaire a second time. The median (range) for energy-adjusted correlation coefficients between 30 nutrients measured by the questionnaire and the diet records was 0.36 (0.06-0.81) for men and 0.37 (0.11-0.52) for women. The median correlation (range) for 17 food groups was 0.30 (0.08-0.75) for men and 0.28 (0.08-0.46) for women. The median correlation (range) for energy-adjusted correlation coefficients between the two questionnaires was 0.24 (0.04-0.69) for men and 0.50 (0.27-0.60) for women for the nutrients, and 0.34 (0.15-0.63) for men and 0.48 (0.18-0.55) for women for the food groups, respectively. The results indicate that this brief food frequency questionnaire provides reasonably valid and reproducible measures of consumption for many nutrients and food groups, and is useful for examining the association between diet and health in the Japanese population.

Center Districts (Ninohe, Yokote, Saku, Katsushika-kita, and Ishikawa) were asked to complete a self-administered questionnaire including a 44-item food frequency section, which is referred to as the baseline food frequency questionnaire 1 (BFFQ1)9. Completed questionnaires were collected from 50,245 subjects, with a response rate of 82%.

Subjects
The subjects of this validation study were a subsample of the participants in the prospective study in four districts, omitting one area in Tokyo (Katsushika-kita). We selected 122 men and 125 women on a voluntary basis. Table 1 summarizes the sequence of data collection. In three districts of mainland Japan (Ninohe, Yokote and Saku), we obtained 7-day diet records four times in different seasons in 1994, five years after the baseline survey, and administered the same questionnaire (BFFQ2) again in February, 1995. In one district of Okinawa Island, we collected 7-day diet records in the winter and summer of 1995, and the second questionnaire was administered in February 1996. We collected the diet records only twice in this sub-tropical area because the seasonal variations in diet were small. We also collected fasting blood and 24-hour stored urine samples twice, in February and August of 1994 (three mainland districts) or in 1995 (Okinawa district). Twenty men and 12 women who failed to provide their complete diet records were excluded, along with 8 men and 6 women who either did not respond or provide complete information for the BFFQ1 or BFFQ2. Subsequently , the remaining 94 men and 107 women were included in this analysis .

Food Frequency Questionnaire
The food frequency questionnaire asked about the usual consumption of 44 food items during the previous month . Standard portions/units were neither specified nor asked. Four frequency categories were employed for the majority of foods (almost never / 1-2 days per week / 3-4 days per week / almost everyday) . For rice and bean paste soup, inquiries were made as to the number of bowls consumed per day. Questions on the frequency of alcohol consumption covered six frequency categories (almost never / 1-3 days per month / 1-2 days per week / 3-4 days per week / 5-6 days per week / almost everyday), and queried those who consumed 1-2 days per week or more about the usual combinations and amounts of five beverages (sake, shochu, beer , whiskey, and other) consumed in a day, e.g., one bottle of beer and one bottle of sake. For 9 non-alcoholic beverages, six frequency categories were used (almost never / 1-2 days per week / 3-4 days per week / 1-2 cups per day / 3-4 cups per day / 5 or more cups per day).
Based on the dietary record data, we developed a food composition table that corresponded to the food items listed in the questionnaire, and determined standard portions/units for each food category using the observed median values. The food composition table was developed and standard portions/units were assigned separately for men and women. Daily nutrient intakes were calculated by multiplying the consumption frequency of each food by the nutrient content of the standard portions/units and summed these values for all foods. Calculation of total energy and 30 nutrients was done with the Standard Food Composition Table11 and other sources. Consumption of 17 food groups was also determined.

Dietary Records
We collected weighed dietary records over seven consecutive days by a method used in the National Nutrition Survey12 with some modifications. Research dietitians instructed the subjects to record all foods and beverages prepared and consumed in a specially designed booklet. Participants were asked to provide detailed descriptions of each food , including the method of preparation and recipes whenever possible . The dietitians checked the records at the subjects1 homes during the survey and reviewed them in a standardized way after completion . The mean daily consumption of nutrients and food groups was calculated .

Statistical Analysis
To assess the correspondence between DRs and BFFQ2 , we compared the mean daily intakes and computed Spearman correlation coefficients for each nutrient and food group . For selected nutrients, we calculated Spearman correlation coefficients between serum and urinary levels (mean of the two measurements) and intakes measured by BFFQ2 . To examine the 5-year reproducibility of the questionnaire, we compared the mean daily nutrient intakes and computed Spearman correlation coefficients between BFFQ1 and BFFQ2. The correlation coefficients were energy-adjusted by the residual method .1 Table 2 shows the validity of BFFQ2 for nutrient intakes . The mean daily intakes calculated from BFFQ2 tended to be lower than those computed from DR . For men, energy-adjusted correlation coefficients between BFFQ2 and DR ranged from 0 .06 (polyunsaturated fatty acid) through 0.81 (alcohol) , with a median of 0.36. For women , it ranged from 0.11 (niacin) through 0.52 (alpha-carotene), with a median of 0.37. The correlations for fatty acids were generally higher when their respective consumption was expressed in terms of the percentage of total fatty acid intake . Table 3 shows the validity of BFFQ2 for consumption of food groups. The mean daily intakes calculated from BFFQ2 tended to be lower than those computed from DR . For men, energy-adjusted correlation coefficients between BFFQ2 and DR ranged from 0.08 (seasonings and spices) through 0 .75 (alcoholic beverages), with a median of 0 .30. For women, it ranged from 0.08 (seasonings and spices) through 0 .46 (milk and dairy products) , with a median of 0.28. Table 4 presents the validity of BFFQ2 for food intakes (g/day)     Table 7. Food group intakes (g/day) and Spearman rank correlation between food group intakes assessed with BFFO11 and BFFO22 1 BFFQ1, Baseline food frequency questionnaire. in 1990 2 BFFQ2, Baseline food frequency questionnaire in 1995 (for Ishikawa, in 1996) 3 Energy was adjusted by residual method. 4 Pickled plum (umeboshi) was included in pickled vegetables, and not in total vegetables but rather in fruits. For n=94, r>0.21 = p<0.05, r>0.27 = p<0.01, r>0.34 = p<0.001. For n=107, r>0.19 = p<0.05, r>0.25 = p<0.01, r>0.32 = p<0.001. and food frequency (times/day). The validity for food intakes (g/day) and food frequency (times/day) was comparable, although the correlation coefficients were different with the kind of food. Table 5 indicates the validity of BFFQ2 using biomarkers of selected nutrients as the reference standard. Among men, dietary intake measured by the questionnaire showed a moderate correlation with serum levels of n-3 polyunsaturated fatty acid, eicosapentaenoic acid, docosahexanoic acid as a percentage of total fatty acid, serum levels of alpha-carotene, beta-carotene, daidzein, genistein, and urinary levels of sodium. Among women, correlations between the questionnaire estimates and biomarkers were generally low. Table 6 shows the 5-year reproducibility of the questionnaire for nutrient intakes. The mean daily intakes were generally comparable between BFFQ1 and BFFQ2. For men, energy-adjusted correlation coefficients between BFFQ1 and BFFQ2 ranged from 0.04 (lycopene) through 0.69 (alcohol), with a median of 0.24. For women, it ranged from 0.27 (lycopene) through 0.60 (total fat), with a median of 0.50. Table 7 shows the 5-year reproducibility of the questionnaire for consumption of a food group. The mean daily intakes were generally comparable between BFFQ1 and BFFQ2. For men, energy-adjusted correlation coefficients between BFFQ1 and BFFQ2 ranged from 0.15 (algae) through 0.63 (alcoholic beverages), with s median of 0.34. For women, it ranged from 0.18 (algae) through 0.55 (non-alcoholic beverages), with a median of 0.48.

DISCUSSION
We found that the mean daily intakes were lower in the questionnaire than in the diet records for most nutrients and food groups. This is probably due to the relatively small number of food items listed in the questionnaire (44 items). Thus, the questionnaire would underestimate the absolute levels of consumption of nutrients and food groups.
Correlations between the questionnaire and the dietary records were lower than those reported in Western studies,1 but comparable with those found in Japanese studies. [2][3][4][5][6][7][8] The results indicate that the questionnaire is reasonably valid in ranking subjects and classifying them into quintiles according to the consumption of many nutrients and food groups.
Reproducibility of our questionnaire tended to be lower than reported in previous Western1 and Japanese2-8 studies. We administered the two questionnaires at a 5-year interval, which is substantially longer than the intervals in other studies (typically one year). The longer interval of questionnaire administration would result in the observed lower correlations. The reproducibility of the questionnaire in the present study was generally higher in women than in men, suggesting that patterns of nutrient and food consumption are more stable over time in women than in men among the Japanese population.
In conclusion, this brief, 44-item food frequency questionnaire used in the baseline survey of the Japan Public Health Center-Based Prospective Study Cohort I provides reasonably useful measures of intake for many nutrients and food groups. The questionnaire would be useful in examining the association between diet and health in the Japanese population.