Self-Administered Diet History Questionnaire Developed for Health Education: A Relative Validation of The Test-Version by Comparison with 3-Day Diet Record in Women

A self-administered diet history questionnaire has been developed for the use in health education in Japan. The relative validity of the test-version was examined using 3-day diet record (DR) as a reference method. Subjects were mildly hypercholesterolemic 47 women aged 38-69 years living in Hikone, Japan. The questionnaire provided close estimation of nutrients compared to the DR (mean of the 3-days), 1-25 % differences between the two methods were observed for total energy and 17 nutrients examined. The differences were in general smaller for macronutrients, 1-3 %, than for micronutrients, 1-25 %. Pearson correlation coefficients between the questionnaire and the DR (mean of the 3-days) ranged from 0.16 for niacin to 0 .60 for saturated fatty acid (mean = 0.41). The energy-adjustment using regression analysis and the correction of attenuation due to error from a limited number of days of DR slightly improved the results. The correlations ranged from 0.19 for niacin to 0.75 for saturated fatty acid (mean = 0.48). Average 37 and 6 % subjects were classified into same quartile and opposite quartile respectively between the two methods. This questionnaire may be useful to assess individual nutrient intake level at least for the subjects examined. J Epidemiol, 1998 ; 8 : 203-215 .


Study design
Between May and June, 1995, 189 mildly hypercholesterolemic subjects, serum cholesterol = 220-259 mg/dL (5.69-6.70 mmol/L) for women aged more than or equal to 55 years and 200-239 mg/dL (5.17-6.18mmol/L) for women aged less than 55 years and men, were screened from the annual health check-up service performed in Hikone, Japan, and enrolled to a health education program organized by the city.The subjects were considered to need a health education before pharmaceutical treatment.
A self-administered diet history questionnaire has been mailed with an inviting letter to the program in the beginning of September, 1995.Ninety subjects agreed to participate in the program.Public health nurses visited their home after making appointment by telephone, and checked and collected the questionnaire one to two weeks later.When participants could not answer the questions correctly, the nurses asked the questions again with brief explanations and obtained the answers.Seventy subjects completed the DHQ.At the moment, the subjects were not informed that they would be asked to keep DR later.About two weeks later, when the program has started in the late of September, the participants took a 30 min lecture using a written pamphlet on how to keep DR from one dietitian.Fifty-five subjects kept the DR for three days.Among them four were excluded because of the incompleteness of the DR.In the remained fifty-one subjects, 47 and four were women and men respectively.Only 47 women were included in the analysis.Mean (standard deviation) age was 55.9 (9.9) years old, and age-range was 38 -69 years old.No subject was a current smoker.

Three-day diet record
The subjects kept DR, either consecutively or non-consecutively, for three weekdays.The three days were chosen by subjects not to include a day with substantially abnormal eating.The subjects were recommended to weigh all foods eaten.But some did not have a scale at home.Foods eaten out of their home were not weighed.They were advised to report volume of foods eaten when the weight was not available and they did so.One dietitian estimated weight of foods eaten described with volume and other explanations by subjects.In this population, most of foods were eaten at home and they were cooked by themselves or their family members.A computer program with the Japanese food composition tables, 4th edition 13,14), and additional tables for fatty acid compositions 15) was used to analyze the diet records 16)

Self-administered diet history questionnaire
A self-administered diet history questionnaire has been developed for the use in health education.It consists of seven sections; 1) eleven questions on dietary behaviors such as `how often and how much do you usually use "Shoyu (soy-sauce)" at table?' 2) semi-quantitative frequency questions on selected 110 food items, 3) questions on frequency and quantity on core-foods on rice, 4) major cooking methods for vegetables, fish, and meats, 5) frequency and quantity of alcohol beverage ("go"/drink), and major types of alcoholic beverages, 6) types, frequency, and quantity of supplement, 7) open-ended questions to describe food names and the quantities eaten regularly (at least once a week).The questions were designed to ask diet in a previous month.For questions about rice, sizes of bowls usually used were asked.For foods listed in the semi-quantitative frequency questions, commonly consumed portion sizes were explained in words such as `half for apple and `one big leaf for cabbage.The 110 food items and the portion sizes listed in the semi-quantitative frequency question section were chosen as foods consumed commonly in Japan.They were selected mainly from a food list used in National Nutrition Survey of Japan').Foods with similar compositions were categorized into one food item.Some foods with different compositions listed as one food item in the food list of National Nutrition Survey of Japan were divided into more than one food items.For example, milk was divided into two types of milk, i.e., regular fat milk and low fat milk.Local foods or menus were not considered for the.food selection.This was done for that the results of this validation study can indirectly indicate validities when the DHQ was used to other populations in Japan.The food-names listed in the questionnaire and the food-codes used for nutrient calculation are shown in Appendix.The frequency was answered using a combination of the two questions; 1) per day, week or month, and 2) once, twice, three times, or more than or equal to four times.Relative quantity was answered compared to the normal portion size indicated in a question; very small (50 % or less), small (about 70-80 %), medium (about same as the portion size), large (20-30 % larger), or very large (50 % larger or more).In the section of the open-ended questions, one dietitian estimated the weight of foods reported when the sizes of the foods or of the packages were described.When a missing or questionable answer was detected after the collection of the DHQ, the public health nurses checked and filled the answer by a telephone interview.
Nutrient intake was calculated using an ad-hoc computer program developed to analyze the questionnaire .As a data-:)ase of food composition tables, the Japanese food composi-:ion tables, 4th edition 13.14), and the other 15) were used .sodium, sugar, and cooking oil added during cooking were ,stimated from the questions in the dietary behavior questions , he major cooking methods of vegetables, fish , and meats, and weight of foods obtained in the semi-quantitative food frequen-;y questions.Salt from Miso-soup and the use of table salt were added to the salt intake.Fat preference asked in the dietary behavior questions were considered to choose foodcodes for meats.For some foods without compositions for fatty acids and/or cholesterol, similar foods with the compositions were substituted (Appendix).In this case, the composition of substituted fatty acids was adjusted for total fat.Additional information was used for fatty acid compositions of some foods 19.For foods, most of them are beverages, fruits or vegetables, with total fat less than 1.0 % of total energy, fatty acid compositions remained missing when without the compositions.For cooked foods which are not listed in the food composition tables of Japanese foods 13, 14, the nutrient compositions were calculated using the recipes prepared by one dietitian.Foods listed in the open-ended questions were checked by dietitians.They were erased, removed to the correct questions, or remained in this section.After considering the contribution to nutrient intake levels, all foods remained in the section were not included in the analysis.

Nutrients included in the study
We calculated intake levels of total energy, four macro-and 11 micro-nutrients from the both surveys, i.e., the DHQ and the DR.Fat was furthermore divided into three subgroups, i.e., saturated fatty acid (SFA), monounsaturated fatty acid, and polyunsaturated fatty acid.Eighty-nine and 72 % of the subjects reported no alcohol intake in the 3-day DR and in the DHQ respectively.Alcohol was therefore excluded from the analysis.Nutrients from supplements were not included in the analyses in this study.

Statistical methods
Firstly, we compared the subject means and standard deviations between the DR and the DHQ.For the DR, the mean of the 3-days was used as a representative value of each nutrient for each subject.Both crude values and ones expressed as energy density, except for total energy, were calculated.To examine the difference between the means for each nutrient between the DR and the DHQ, the mean differences of nutrients between them with the 95 % confidence intervals by a paired t-test were calculated.The relative differences with the DR as a reference were also calculated.
Secondly, Pearson product-moment correlation coefficients were used to compare the two dietary assessment methods.We calculated also Spearman correlation coefficients.They were not apparently different each other, therefore only the Pearson correlation coefficients are presented in this report.To examine the correlation between nutrient intakes after removing covariation due to total energy intake, we calculated energy-adjusted nutrient intakes, which indicate the nutrient composition of the diet by regressing nutrient intake on total energy 14.The distributions of most nutrients were skewed towards high values.Nutrients were transformed (loge) to improve normality for the correlation analysis.The presence of day-to-day intra-individ-ual variation tends to attenuate the correlation between the DR and the DHQ 19).In order to correct this factor, we used the within-and between-person components of variation in the DR intake (treating each day as random unit of observation).We thus obtained 'de-attenuated' correlation coefficient (see footnote of Table 3) to estimate the correlation between nutrient intake calculated from a single question and multiple days of diet recording.
The distributions of nutrients derived from the DR and from the DHQ were divided into quartiles to examine how well individuals were classified into quartiles by means of the DHQ compared with the DR.Classification error and overall agreement.were examined here.Furthermore, we calculated the mean intake of nutrients from the DR for groups of subjects who were defined by quartile of intake of the same nutrient on the DHQ.This validation can be used to fit models for true nutrient intake (from the DR) as a function of measured intake (from the DHQ) 20).The means between the lowest quartile and the corresponding quartiles were compared with one-way ANOVA (Dunnett's method).A significance level was put at p<0.05.In the analysis for Pearson product-moment correlation coefficients, most of all coefficients were statistically significant.Moreover, since the purpose of the analysis was on the quantification of misclassification rather than hypothesis testing, we have not presented p-values in this analysis.

Average nutrient intakes
Table 2 shows means and standard deviations of energy and 17 nutrient intake levels.Slightly higher means were obtained from the DHQ than from the DR in most nutrients examined.The relative differences (%) of the mean intake levels obtained from the DR and from the DHQ were 1 -3 % for total energy and macronutrients and 1 -25 % for the micronutrients.For calcium, phosphorus, cholesterol, and iron, the means obtained from the DHQ were statistically lower than those from the DR (p<0.05 -0.001 by paired t-test).In thiamin and riboflavin, the means obtained from the DHQ were statistically higher than  those from the DR (p<0.05 -0.01 by paired t-test).The relative differences (%) between the means of the two assessment methods were not substantially different between the analysis with crude values and with energy density.

Validity
Table 3 shows Pearson correlation coefficients for unadjusted and energy-adjusted daily nutrient intakes between the average intake measured by means obtained from the DR and corresponding values obtained from the DHQ.The comparisons between unadjusted nutrient intake ranged from r = 0.16 for niacin to r = 0.60 for SFA.After the energy-adjustment for nutrient intake, the mean of correlation coefficients improved from 0.41 to 0.46.For some nutrients, sodium in particular, the degree of the correlation decreased by the energy adjustment.Table 3 also contains the de-attenuated correlation coefficients taking within-and between-person variations in the DR into consideration.The correlation coefficients ranged from r = 0.19 for niacin to r = 0.75 for SFA (mean = 0.48).
Table 4 summarizes the cross classifications for nutrient intake using quartiles by the two assessment methods.Taking total energy as an example, 58 and 67 % of the subjects classified into the lowest quartile by the DR was classified into the lowest quartile and the lowest two quartiles by the DHQ respectively.Among the subjects classified into the highest quartile by the DR, 42 and 75 % was classified into the highest Table 3. Pearson correlation coefficients for comparison of self-administered diet history questionnaire scores with the means of 3-day diet records both unadjusted and adjusted for total energy (data based on information provided by 47 women aged 38-69 in Hikone.Japan.1995).a Nutrient values were transformed (loge) to improve normality .n Intakes adjusted for total energy intake using regression analysis .
The de-attenuated correlation coefficient is calculated using the ratio of the within-to between-person variance measured from the 3-day diet records.The formula for this corrected correlation is calculated as re = to [1+(S2w/S2b)/n]1/2 where to is the observed correlation, S2w is the within-person variation, and S2b is the between-person variation and n is the number of replicate measurements.For this calculation n = 3 representing each day of the diet record.d De-attenuation only Table 4. Cross-classification (%) of total energy and 17 nutrient intakes between the mean of 3-day diet records and the selfadministered diet history questionnaire for energy-adjusted intake except for total energy (data based on information provided by 47 women aged 38-69 in Hikone.Japan, 1995)a.
a All nutrient , except for total energy, intakes were adjusted for total energy intake using regression analysis.
quartile and the highest two quartiles by the DHQ respectively.Thirty-four % of the subjects were classified into the same quartile by the DR and by the DHQ.Two percent of them were classified into the opposite quartile (the lowest versus the highest).More than or equal to 40 % of the subjects were classified into same quartile for seven nutrients out of total energy and 17 nutrients examined.Table 5 shows the mean intake of total energy and 17 nutrients from the DR for groups of subjects who were defined by quartile of intake of the same nutrient on the DHQ.In total energy and 6 nutrients, mean intakes increased monotonically across quartiles.When the means of the highest and of the lowest quartiles were compared, in total energy and 13 nutrients a statistical difference was observed (p<0.05-0.001 by Dunnett's method).

Development of a dietary assessment questionnaire for
Oriental populations has scarcely been reported21).We examined a relative validity of a self-administered diet history questionnaire and its associated nutrient analysis program developed for Japanese population using a 3-day DR as reference.The group means suggested a reasonable ability of the questionnaire to estimate population means of nutrient intakes.The observed mean total energy intakes were 807 kJ/day for the DHR and 861 kJ/day for the DR lower than those observed in women of a similar age range, 50 -59 years , in the National Nutrition Survey, 1995 22) .The subjects in this study who were health conscious might have underreported the foods eaten in both survey methods.The correlations observed suggested a moderate ability of the questionnaire to rank individuals along the distribution of intake.The overall median correlations observed in this study (median = 0.42 for crude values and 0.48 for energy adjusted and de-attenuated ones) were similar to or slightly lower than those reported in previous studies .The mean or median correlation coefficients of nutrients in previous studies, which used DR as a reference method, ranged Table 5. Mean intake of total energy and 17 nutrients from 3-day diet records within quartiles of intake determined by the selfadministered diet history questionnaire (data based on information provided by 47 women aged 38-69 in Hikone, Japan, 1995)a.a All nutrient , except for total energy, intakes were adjusted for total energy intake using regression analysis.
To our knowledge, some self-administered dietary assessment methods have been developed in Japan 8-12).Date et al. developed semi-quantitative menu-based frequency questionnaire and reported a reasonable validity 8).They used voluntary dietetic students for the study subjects.Kono et al. developed a semi-quantitative food frequency questionnaire and failed to obtain enough validity 9).Takatsuka et al. validated a simplified diet history questionnaire and reported that the mean correlation coefficients for 17 nutrients were 0.41 for crude values and 0.37 after energy-adjustment 10).Some other studies reported a reasonable validity for macronutrients and limited micronutrients 11,12).On the other hand, Japanese showed higher degree of Pearson correlation coefficients of nutrient intakes between diet history method and diet records among the five major ethnic groups in Hawaii 34).The questionnaire used in the study in Hawaii asked more number of foods and more dietary behaviors in detail, and the structure was more complicated than most questionnaires previously developed in Japan.The structure of the questionnaire of the study is more complicated and has more volume of questions compared to the question-naires of previous studies in Japan [8][9][10][11][12].Although the results were not satisfactory enough, they were thought to be comparable to those of the self-administered diet assessment methods developed in Western countries18, [23][24][25][26][27][28][29][30][31][32][33].The questionnaire produced close group mean nutrient intake estimates compared to the 3-day DR.The results were in agreement with previously developed similar questionnaires in the United States 4,21).
When the correlations between the DR and the DHQ (Table 3) were compared between the nutrients, fat, especially SFA, showed a relatively high degree of correlation.It was also relatively higher than those reported in previous studies in Western populations 18,23-33).It may partly be because that fat-, especially SFA-, rich foods were limited in Japanese diets compared to Western diets.In contrast, the degree of the correlation on sodium was low.It may partly be due to the quality of the program for estimating seasoning use.An estimation of salt added during cooking and at table is difficult by DR too 34).We examined validity of the DHQ on sodium and potassium with single 24-hour urinary excretions using 223 university students35).Although a reasonable validity was observed on potassium, the result on sodium was inconclusive because the duration of Appendix. (Continued) Appendix.(Continued) 8 AL : alcoholic beverage section , C : calculated from the combination of questions, F : food frequency section, M: main staple section."Label" indicates the composition was obtained from food label ."Recipe" indicates the composition was calculated from the recipe ."Ref" indicates the composition was obtained from the reference (15) .
urine collection was too short.The results on niacin, iron, vitamin A, and cholesterol were less satisfactory compared to macronutrients.Longer days of diet recording are necessary to assess individual habitual intakes for these nutrients 36.37),0. Therefore, the relatively lower degree of the correlations between the DR and the DHQ on these nutrients may be due to a shortcoming of the study design and not to a quality of the DHQ.Three-day DR is not long enough to estimate individual habitual diet also for other nutrients 36.37.38).
Seasonal variation in dietary habits exists in rural Japanese population 39).The comparison with repeated records across one year should be performed 26,29).We, however, asked to answer dietary habit of the previous month in the DHQ.A possible seasonal variation of diet is presumably not a major problem of this study-design.
The 110 foods listed in the closed-ended semi-quantitative frequency question section of the DHQ were mainly selected from a food list used in the National Nutrition Survey of Japan 17).The foods listed there were not substantially different from the reported major foods actually eaten in rural Japanese populations 40).But some local foods, for example a kind of preserved fish, which were not listed in the Japanese food composition table 13.14), was replaced by a similar food listed in it for calculation of nutrients.It might be a bias.A data-based approach is usually used for a selection of food-items for a development of a dietary assessment questionnaire 41) with an exception 18).Because the food items and their portion sizes were not determined from data obtained from the population examined, the observed validity of the DHQ may be somewhat lower than a questionnaire developed by a data-based approach for food-item selection and their portion-size determination.Self-administered dietary assessment questionnaires do not necessarily use picture-or model-aids to estimate portion size except some 26,33,42).Some studies on portion size estimates in food frequency questionnaire concluded that the value of the aid was limited43,44,45).The questionnaire validated in the study did not use any aid for estimation of portion size.
This questionnaire has been developed for the use in health education.Both to estimate individual intakes of nutrients of interest and to rank individual in the population for nutrients are necessary.The results indicated that the questionnaire and its associated nutrient analysis program may be useful to assess individual habitual nutrient intake in health education at least in the population examined.However, the small sample size, the short days of diet recording, the incompleteness of foodweighing, and the low response rate of the dietary recording have limited the validity of the study.Validity of the DHQ to other populations such as men and/or healthy subjects has also remained to resolve.Further validation studies overcoming these problems should therefore be performed.Improvement is still required especially for some nutrients which are important for the prevention of cardiovascular disease and cancer.
aRange =38-69 b 1 mg/dL = 0 .02586mmol/LTable 2. Mean (standard deviation in parenthesis) daily intake of total energy and 17 nutrients (excluding supplements) measured by three-day diet records and a self-administered diet history questionnaire (data based on information provided by 47 women