Reproducibility of a Self-administered Food Frequency Questionnaire Used in the 5-year Follow-up Survey of the JPHC Study Cohort I to Assess Food and Nutrient Intake

We examined the reproducibility of a self-administered semiquantitative food frequency questionnaire (FFQ) used in the 5-year follow-up survey for the Japan Public Health Center-based prospective Study on cancer and cardiovascular diseases (JPHC Study) to estimate nutrient and food intake by using repeated FFQs at a 1-year interval in 101 men and 108 women. Between energy and each of 32 nutrients, the correlation coefficients in crude values varied from 0.41 for vitamin B12 to 0.83 for alcohol (median=0.59) in men and 0.52 for alpha-carotene to 0.77 for iron (median=0.67) in women. In 21 food groups, it varied from 0.42 for seasonings and spices to 0.80 for pickled vegetables (median=0.61) in men and 0.45 for seasonings and spices and 0.74 for pulses, milks, and pickled vegetables (median=0.63) in women. The correlation coefficients for the energy-adjusted values (medians were 0.49 and 0.50 for nutrients and 0.50 and 0.49 for food groups in men and women, respectively) were somewhat lower than for the crude values. The difference in mean intakes between the two FFQs was less than 10% in most of the nutrient and food groups. The results suggest that the reproducibility of the FFQ used for the JPHC study was moderate to high in most of the nutrient and food groups.

Nutrient and food intakes have been assessed with a selfadministered food frequency questionnaire (FFQ) in several nutritional epidemiologic studies for chronic diseases.1.2 In most observational epidemiologic studies, single measurements of dietary habits have been used as representative of long-term dietary habits of individuals.To use a questionnaire for this purpose, the reproducibility as well as the validity should be examined to assure the consistency of the dietary habits in the target population.We therefore examined reproducibility of a FFQ used in the 5-year follow-up survey of the Japan Public Health Centerbased prospective Study on cancer and cardiovascular diseases (JPHC Study) using two data sets obtained one year apart.

METHODS
The study design and subject characteristics have been reported elsewhere.3Subjects included in the analysis were 101 men and 108 women who completed FFQ twice at a one-year interval in the Ninohe, Yokote, Saku and Ishikawa Public Health Center (PHC) areas in the JPHC Study.In the Ninohe, Yokote and Saku PHC areas, the first FFQ (FFQ1) was conducted in February, 1995 as a part of the 5-year follow-up survey for the entire JPHC study cohort I, and the second FFQ (FFQ2) was conducted in February, 1996.In the Ishikawa PHC area, the FFQ1 was conducted in February, 1995, and the FFQ2 in February, 1996.In the previous report on the design of this study 3 the FFQ immediately after the completion of dietary records (DR) was described as the FFQ for validity (FFQ V) and the other as the FFQ for reproducibility (FFQ R) in order to distinguish the validity of the FFQ in conjunction with DR.For the purpose of this report, however, we aimed to examine the reproducibility of two FFQs at a oneyear interval in chronological order.Therefore, for the Ninohe, Yokote and Saku PHC areas, the FFQ_V was regarded as FFQ1, and the FFQ R as FFQ2.For Ishikawa PHC, the FFQ R was regarded as FFQ1 and the FFQ V as FFQ2.
The methods for computing nutrient and food intakes from the FFQ have been described in this Supplement4 The mean crude intakes of energy, 32 nutrients and 21 food groups were calculated from each FFQ (the first and the second FFQs).Vegetables were divided into three groups, i.e., green and yellow, non-green Reproducibillty of Food Frequency Questionnaire and yellow (namely, other), and pickled vegetables.Beverages were divided into two groups, alcoholic and non-alcoholic.Intake of sugars and sweeteners was not computed in this FFQ so the food group was not included in the analysis .Mean energy-density values were calculated as intake of nutrient and food groups per 1000 kcal.The percent difference between the first and second FFQ was computed by dividing the difference in mean intake between the two FFQs by the FFQ1 mean: (FFQ2 mean-FFQ1 mean)/FFQ1 mean.
Intakes of nutrients and foods were adjusted for total energy intake using a residual model 5. Spearman rank correlation coefficients between the two FFQs were computed for crude and energy-adjusted intakes.The Spearman rank correlation coefficient was used for the correlation analysis because the distribution was skewed in most values.Because our purpose was to quantify measurement error rather than test a hypothesis, p values were not presented for correlation coefficients.All the analyses were performed separately for men and women.The computation was performed using the data with the 4 areas combined.

RESULTS
Table 1 shows the mean and standard deviation (SD) of nutrient intakes in crude values and the differences between the two FFQs.The percent difference was less than 5% in most nutrients except for n-6 polyunsaturated fatty acid, sodium, carotenes (alpha-and beta-), vitamin B12, daidzein and genistein in men, and saturated fatty acid, n-6 polyunsaturated fatty acid, alcohol, alpha-carotene, vitamin B12 and genistein in women.
Table 2 shows the mean and SD of nutrient intakes in energydensity values and the differences between the two FFQs.The percent difference was less than 5% in most nutrients except alcohol, carotenes (total, alpha-and beta-) in both sexes, and vitamin B12, daidzein and genistein in men.
Table 3 shows the mean and SD for food intake by food group in crude values and the differences between the two FFQs.The variation in the differences was greater than in nutrient intakes, i.e., the difference was more than 5% in 11 food groups within the 21 food groups examined both in men and women.The variation in the differences in energy-density values was similar to that in crude values (Table 4).
Table 5 shows Spearman rank correlation coefficients for nutrients between the two FFQs.The correlation coefficients for crude nutrient intakes varied from 0.41 in vitamin B12 to 0.83 in alcohol in men, and from 0.52 in alpha-carotene to 0.77 in iron in women.The median correlation coefficients for crude nutrient intakes were 0.59 and 0.67 in men and women, respectively.After nutrient intakes were adjusted for energy intake, the correlation coefficients varied from 0.30 in vitamin B12 to 0.82 in alcohol in men, and from 0.32 in protein to 0.68 in alcohol in women.The median correlation coefficients for energy-adjusted nutrient intakes were 0.49 and 0.50 in men and women, respectively.
Table 6 shows Spearman rank correlation coefficients for food groups between the two FFQs.The correlation coefficients for crude food intake by food group varied from 0.42 in seasonings and spices to 0.80 in pickled vegetables in men, and from 0.45 in seasonings and spices to 0.74 in pulses, milks, and pickled vegetables in women.The median correlation coefficients for crude food intakes were 0.61 and 0.63 in men and women, respectively.After food intakes by food groups were adjusted for energy intake, the correlation coefficients varied from 0.38 in total beverages to 0.71 in alcoholic beverages in men, and from 0.30 in nuts and seeds to 0.74 in milks in women.The median correlation coefficients for energy-adjusted food intakes were 0.50 and 0.49 in men and women, respectively.

DISCUSSION
In the present study, we examined the reproducibility of the FFQ which was repeatedly administered at a one-year interval for estimating dietary intake of nutrients and foods.We compared two repeated measurements by Spearman rank correlation coefficients, which ranged from 0.41 to 0.83 among various nutrients for crude nutrient intake in men and women.The correlation coefficients for most of the nutrients were higher in women than in men, but lower after energy adjustment both in men and women.The correlation was relatively high in nutrients, such as n-3 polyunsaturated fatty acid, vitamin C, daidzein and genistein, whose food sources were limited to a few food items.The correlation coefficients for crude food intake ranged from 0.42 to 0.80 among various food groups in men and women.Correlation coefficients for men and women were similar, and lower after energy adjustment; they were relatively high for confectionaries, pickled vegetables, and alcoholic beverages which tended to be consumed according to individual preferences.
Reproducibility of our FFQ was comparable to the results from the validation study of similar questionnaires previously developed in Japan.Correlation coefficients between repeated measurements in our study for selected nutrients were compared to that in two other similar questionnaires in Table 7 .6-8Our results indicated the same declining tendency as the study by Imaeda et al.8 after nutrient intakes were adjusted for energy .The correlation coefficients also decreased for food groups after the energy adjustment in our study.The same tendency after energy adjustment was observed in the validity of our questionnaire in estimating nutrients and food groups.9. 10 Correlation coefficients were slightly higher in women than in men for most of the nutrient and food groups, in contrast to where the validity was concerned .9.10 Correlation coefficients between repeated measurements for estimating dietary intake are usually on the order of 0 .5-0.7.1 This 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 epidemiologic studies.1 Except for vitamin B12, cereals , algae, and seasoning and spices, the reproducibility of our FFQ for estimating intake of nutrients and food groups was r>0.5, which was suffi-Table 6. Spearman rank correlation coefficients between two FFQs administered at one-year interval for estimated nutrient intakes 1 Food intakes were adjusted for energy intake by residual method .

Table 7 .
Comparison Residual method of Food Frequency Questionnaire ciently reasonable for epidemiologic use.With the reasonable validity of our FFQ,9.10wecan be sure that our FFQ at one point in time can determine the usual intakes of individuals over the period of one year.In conclusion, the reproducibility of the FFQ used in the JPHC Study cohort I in estimating most of the nutrient and food group intakes was comparable to that of the other questionnaires developed in Japan. *