Reproducibility of a Semi-quantitative Food Frequency Questionnaire in Japanese Female Dietitians

Objective: To examine reproducibility of assessed intake of foods and nutrients according to a semi-quantitative food frequency questionnaire (SQFFQ) in Japanese female dietitians. Subjects and Methods: An SQFFQ was self-administered to 106 (21 male and 85 female) Japanese dietitians in Aichi prefecture in autumn 1996 and the same questionnaire was repeated in autumn 1997. Reproducibility was evaluated in terms of consumption of 15 foods and energy and 30 macro- and micro-nutrients based on the SQFFQ from 84 Japanese female dietitians. Results: For intake of foods, Pearson’s correlation coefficients (CCs) with log-transformation and energy adjustment (minimum - median - maximum) ranged from 0.35 (beverages) - 0.61 - 0.71 (dairy products). ANOVA intraclass correlation coefficients (ICCs) with log-transformation and energy adjustment ranged from 0.49 (beverages) - 0.74 - 0.82 (dairy products). Spearman’s rank CCs with energy adjustment ranged from 0.43 (confectionery) - 0.57 - 0.76 (dairy products). Weighted kappa statistics with energy adjustment ranged from 0.34 (confectionery) - 0.49 - 0.71 (dairy products). For consumption of nutrients, Pearson’s CCs with log-transformation and energy adjustment ranged from 0.23 (zinc) - 0.55 - 0.74 (insoluble dietary fiber). ANOVA ICCs with log-transformation and energy adjustment ranged from 0.37 (zinc) - 0.70 - 0.84 (insoluble dietary fiber). Spearman’s rank CCs with energy adjustment ranged from 0.25 (zinc) - 0.56 - 0.74 (magnesium). Weighted kappa statistics with energy adjustment ranged from 0.25 (zinc) - 0.50 - 0.68 (insoluble dietary fiber). Conclusions: Substantially high reproducibility of consumption of foods and nutrients was attained from an SQFFQ self-administered to Japanese female dietitians.


INTRODUCTION
In developed countries, the major causes of death are chronic/lifestyle-related diseases including cancer, cerebrovascular disease and heart disease 1,2).In Japan two-thirds of all deaths consist of these causes, which are associated with accumula-tion of gene aberrations caused by interaction of genetic factors and daily lifestyle.The pathogenesis proceeds in a multi-factor, multi-hit and multi-stage fashion with a long latency period.
Genetic factors are of interest in terms of etiology but lifestyle is more important in that we can modify it with a view to disease prevention and health promotion.
Lifestyle factors include food intake , alcohol drinking, smoking, physical exercise , and stress.Among these, smoking is the most potent single factor contributing to chronic diseases.Food consumption also appears to play a crucial role 3) , but in this case both deleterious and beneficial aspects are evident.Observations may remain inconsistent , however, because information on diet is not necessarily valid or reproducible due to fluctuations in dietary intake itself , and in procedures to obtain information on diet and study subjects .
We have developed a data-based SQFFQ to secure longterm dietary intake 4) and conducted a relative validation/calibration of consumption of nutrients as well as foods based on the SQFFQ versus findings with four season consecutive 7 day weighed diet records (abbreviated WDRs hereafter) 5) .Here, we studied reproducibility/repeatability of intake of foods and nutrients by comparing results of two SQFFQs self-administered at a one-year interval to Japanese female dietitians .

Subjects and SQFFQ
We earlier designed an evidence-based SQFFQ according to multiple regression and contribution analyses, and validated the SQFFQ versus 28 day WDRs as described elsewhere 4-6) .
Briefly, in autumn 1996, we recruited 106 (21 male and 85 female) middle-aged Japanese dietitians living in Aichi prefecture, Central Japan, and first mail-administered the SQFFQ (SQFFQ96) and surveyed consecutive 7 day WDRs approximately one week later, and then at about 3 month intervals in winter, spring and summer 1997 (Figure 1).In autumn 1997 , the same questionnaire (SQFFQ97) was applied again.
All female dietitians completed both SQFFQ96 and SQFFQ97: however, one subject was excluded because her response on energy lay beyond 4 and 5 standard deviations from the group means with the SQFFQ96 and SQFFQ97, respectively.Since the number of male dietitians was rather small , they were not included in the present study.

Foods and Nutrients Selected
We chose fifteen foods/food groups and beverages including rice, bread, noodles and potatoes, confectionery, oil, soybean and soybean products, fish and other seafoods, meat, eggs, dairy products, green-yellow vegetables, other vegetables, seaweed, fruit, beverages, and alcohol.
In addition, energy and thirty macro-and micro-nutrients were selected, including protein, fat, carbohydrate, total dietary fiber (TDF) (soluble DF and insoluble DF), minerals (potassium, calcium, magnesium, phosphorous, iron, zinc and copper) and vitamins (carotenes and vitamins A, D, E and C).

Intake of Foods and Nutrients
For regular foods and recipes, except for type and frequency of breakfast and consumption of beverages, food frequency was categorized into eight: that is, never or seldom, 1-3 times/month, 1-2 times/week, 3-4 times/week, 5-6 times/week, once/day, twice/day, and more than twice/day.Intake of beverages was assessed in an open-ended manner.Serving sizes were half, one, 1.5 and two times the standard portion size, and an open-ended bracket was also prepared.
We ascertained average daily consumption of foods and nutrients by multiplying the food intake (in grams) or serving size and the nutrient content per 100 grams of food as listed in the Standard Tables of Food Composition, Version 4, the Follow-up of Standard Tables of Food Composition 7,8) , Composition beverages, and energy and 30 macro-and micro-nutrients according to the SQFFQ96 and SQFFQ97.The differences of means were examined by paired t-test.Secondly, we calculated Pearson's correlation coefficients (abbreviated CCs hereafter), log-transformed CCs, energyadjusted and log-transformed CCs with 95% confidence intervals 11-16) ANOVA intraclass CCs (ICCs hereafter), log-transformed ICCs, and energy-adjusted and log-transformed ICCs with 95% confidence intervals, and Spearman's rank CCs and energy-adjusted Spearman's rank CCs with 95% confidence intervals for intake of selected foods and nutrients with the two SQFFQs.
Thirdly, after categorizing daily consumption of the selected foods and nutrients measured in the two SQFFQs into three groups, we computed percentages of exact agreement and complete disagreement, and kappa and weighted kappa statistics with 95% confidence intervals 17,18)

Intake of Foods
Table 1 shows comparisons between daily intake of foods according to the SQFFQ96 and SQFFQ97.The values with SQFFQ96 were equal to or larger than those of SQFFQ97, except in the alcohol case.There were no statistical differences in daily consumption of foods between the two questionnaires, except for confectionery.

Intake of Nutrients
Table 4 lists comparisons between daily intake of energy and macro-and micro-nutrients based on SQFFQ96 and SQFFQ97.The values of SQFFQ96 were almost equal to or larger than those of SQFFQ97.There were no statistical differences in daily consumption of nutrients between the two, except for vegetable fat, PUFAs, n-6 PUFAs, linoleic acid, n-3 PUFAs, linolenic acid, n-6 PUFAs/n-3 PUFAs, vitamin E and magnesium.

DISCUSSION
When dealing with epidemiologic data, validity and reproducibility are the two most important elements.Relative validity of an SQFFQ versus 28 day WDRs was found to be higher than the values conducted in Japan and essentially equal to those obtained in other areas of the world 5).Many articles of reproducibility of intake of foods and nutrients according to FFQs/SQFFQs have been reported using various types of questionnaire, study subjects and time frames 14, 16) Here we referred reports as cited below assessing reproducibility between two questionnaires administered in one-year interval and compared their values with the present study.
With respect to consumption of foods, the values for reproducibility in this study were generally similar to or greater than those reported earlier [19][20][21][22][23] with some exceptions 24-29).Although we did not initially inform our study subjects about the intention to administer SQFFQ97, we cannot preclude the possibility that there are education/leaming effects on SQFFQ97 due to four season consecutive 7 day WDRs.Further, figures for reproducibility were presumably improved by the fact that the participants were professional dietitians, while the wide diver-Table 5. Pearson's CCs, ANOVA ICCs and Spearman's rank CCs between daily intake of selected nutrients based on SQFFQ96 and SQFFQ97.
sity of Japanese foods may have exerted an opposing influence.Thus, Japanese FFQs/SQFFQs for foods 30.31) seem to have performed as a whole rather less well than those in other regions of the world, cited previously.Similarly, with reference to nutrients, indices of reproducibility in the present study were generally similar to or greater than those reported around the world 19,23,29,32-40) with exceptions 25,41).By the same token, as mentioned, Japanese FFQs/SQFFQs for nutrients 42) did not seem to have performed in general well compared with those in other regions .
In this context the interval between the first and second questionnaires is important.One week or a month may be too short because the memory would be strong.In order to evade influence of memory and change of physical conditions, a threemonth interval within the same season may be appropriate for assessment of repeatability.A six-month interval would introduce effects of seasonal variation in consumption of foods and nutrients [43][44][45][46][47][48].In this study we set one-year interval to delete Table 6.Agreement, disagreement and kappa statistics according to tertile classification of daily intake of selected nutrients based on SQFFQ96 and SQFFQ97.
bias owing to memory and seasonal variance; however, we cannot exclude true change due to disability/disease, which would be compatible with the trend for the longer the administration interval, the lower the value for reproducibility 14,16,49) We prepared an item inquiring whether they intentionally/ unduly modified their dietary habit and lifestyle along with any changes in medical and family history within the year, but none of the study subjects answered in the affirmative.There are several approaches for assessing reproducibility, including Pearson's CCs, ANOVA ICCs, Spearman's CCs, percentages of agreement/disagreement and kappa statistics.Further, energy adjustment does not appear to be powerful for amending the figures of CCs.
In conclusion, the reproducibility of assessed intake of foods and nutrients according to the SQFFQ from Japanese female dietitians obtained in the present study was in line with the satisfactory relative validity with an SQFFQ versus 28 day WDRs gained earlier 5).Having observed seasonal variation in consumption of foods and nutrients in four-season consecutive 7 day WDRs, we have been executing the JADE (Japanese Dietitians' Epidemiologic) Study, making proposals to all 47 prefectural dietetic associations in Japan and securing agreement for participation from a total of 26 of these.We have now mail-administered the validated/modified SQFFQ with a request for written informed consent to members of the associations and the alumnae of several women's colleges for nutrition in autumn 1999, 2000 and 2001.The hope is that highfidelity data can be secured to facilitate assessment of the association between lifestyle and health/disease.
Table of Processed Foodstuff 9) and Table of Trace Element Contents in Japanese Foodstuffs 10) or per 100 grams of model recipe.
Figure 1.Schedule for validation/reproducibility studies of an SQFFQ.

Table 1 .
Comparison of daily intake of selected foods according to SQFFQ96 and SQFFQ97.

Table 2 .
Pearson's CCs.ANOVA ICCs and Spearman's rank CCs between daily intake of selected foods based on SOFFQ96 and SQFFQ97.

Table 3 .
Agreement, disagreement and kappa statistics according to tertile classification of daily intake of selected foods based on SQFFQ96 and SQFFQ97.

Table 4 .
Comparison of daily intake of selected nutrients according to SQFFQ96 and SQFFQ97 .