Validation of simplified diet history questionnaire.

Simplified methods to estimate long term nutrient intakes would be needed for not only nutritional epidemiologic studies but also other ones. Based on data of diet history questionnaires (DHQ) which ask frequency and portion size for 169 items from 2,371 participants randomly selected from our cohort study in a rural city of Japan, we simplified the DHQ by eliminating some items using stepwise regression method. To examine the validity of the simplified DHQ (SDHQ), we obtained the SDHQs from 31 volunteers who had finished to complete one-day records once a month during the previous year, we calculated Pearson's correlation coefficient and calorie-adjusted correlation coefficient between the SDHQ and twelve one-day records for each nutrient intake value. Thirty one items were selected for our SDHQ. The mean values of most nutrient intakes from the SDHQ were more than those from twelve one-day records. The correlation coefficients between the SDHQ and twelve one-day records were more than 0.30 except for fat and monounsaturated fat. Calorie adjustment increased in the correlation coefficients for some nutrients. Our results suggest that the SDHQ is a validated and easy-to-use method for assessing long term 17 selected nutrient intakes.

Epidemiologic studies have indicated relationships between nutrient or food intakes and chronic diseases. Not only nutritional epidemiologic studies but also other ones should include some validated nutrient or food intake assessment methods in their researches. Hankin et al 1). developed the diet History Questionnaire (DHQ). It has been modified and used to estimate average food intakes of the residents in a rural city of Japan). However, the DHQ asked average frequency and quantity in the previous year for 169 foods and dishes and it took about one hour to complete it. Such a long questionnaire might not be suitable in epidemiologic studies where major interests are not in nutrition. We developed a shorter version of the DHQ, i.e., the simplified Diet History Questionnaire (SDHQ) to estimate 17 selected nutrient intakes. This study aims to know the validity of the SDHQ by comparing with mean estimated nutrient intake values from twelve one-day records.

Development of the SDHQ
In September 1992, we distributed the DHQs to all residents aged 35 years old and over in a rural city of Gifu prefecture, Japan, at the beginning of our cohort study. We received the DHQs from 34,018 among 36,990 residents (response rate: 92%). We randomly selected 2,371 subjects from 23,708 respondents who met our conditions and aged under 64 years old, and utilized the DHQ data from these subjects (They were 1,137 males including 58 subjects who had missing values in body mass index (BMI) (with mean age = 49.6 years; SD = 8.4, mean BMI = 22.8 kg/m2; SD = 2.8) and 1,234 females including 35 subjects who had missing values in BMI (with mean age = 49.3 years; SD = 8.3, mean BMI = 22.1 kg/m2; SD = 2.8) to develop the SDHQ 2).
We modified the original DHQ 1) by adding Japanese traditional foods (raw fish, tofu etc.) which resulted in 169 foods and dishes in our DHQ 2). The frequency was asked using 8 response categories except for alcohol beverages, coffee and tea, i.e., "never or hardly ever", "once a month", "2 to 3 times a month", "once a week", "2 to 3 times a week", "4 to 6 times a week", "once a day", "2 or more times a day". For alcohol beverages, coffee and tea, the frequency was asked using 9 categories, further classifying the category of "2 or more times a day" into "2 to 3 times a day" and "4 or more times a day" . The quantity was asked using 3 response categories except for alcohol beverages, i.e., "a half cup or less" , "one cup", "2 cups or more". The quantity of each alcohol beverage was asked using 4 categories, i.e., "one can or less", "2 cans", "3 cans" , "4 cans or more" . For some dishes and foods which were difficult to describe the quantity, the portion sizes were shown using 3 photo pictures. We did not ask quantity about coffee and tea.
The DHQs were distributed to the subjects by the volunteers . The subjects were asked to mark their likely frequency and quantity after reading the instruction page. The DHQs were collected by the same volunteers and sent to our department . We scanned the DHQ data by an automatic optical card reader . We calculated a food score for each item by multiplying the frequency per month by the quantity. Based on Japan Food Composition Table, 4th edition), we calculated 26 nutrient intakes per month by multiplying each food score by standard weight of the food item and nutrients per gram in the food item. Vitamin supplements were not included into the calculation. Each estimated nutrient intake per day was assessed by the summing up amount of nutrient of 169 items per month and dividing it by 30. We described the detail in previous report Z.
To develop our SDHQ, we selected 17 major nutrients ( Table 2). We employed a stepwise regression method to select food items from the 169 DHQ items 4). The stepwise regression method was conducted for each of the 17 nutrient intake as a dependent variable, which was estimated by the DHQ, using the 169 food scores as independent variables. We obtained a set of food items for each nutrients where the cumulative coefficient of determination (R2) was 0.6 or greater . We combined 17 sets of food items and excluded duplications, then we got 31 items as our SDHQ. Missing data of frequency and quantity in the SDHQ was treated as in the DHQ2) a Validation of the SDHQ In July 1994, we recruited 55 volunteers who were residents in Gifu prefecture and aged 35 years old and over including 39 participants in our cohort and were asked to complete one-day records once a month during one year (July, 1994 -June , 1995). The volunteers were noticed randomly selected day once a month just before the day and were asked to record each food item and the weight which they ate on the day . When they were not sure about the food weight, they were asked to record the volume of each food stuff (e.g. half of carrot, etc .) and later a dietitian estimated the weight from the standard value. Forty volunteers among them completed twelve one-day records. In February 1996, we distributed the SDHQs to the volunteers and asked them to complete the questionnaire concerning their food intakes during the one year. We excluded 8 persons who aged under 35 or over 64 years old and one male subject for whom estimated vitamin A intake from his SDHQ was extremely higher (i.e. four times) than for other subjects. We examined the data from 31 volunteers (13 males (mean age = 44.2 years old; S.D. = 9.3, mean BMI = 22.7 kg/m2; S.D. = 2.1) and 18 females (mean age = 47.6 years old; S.D. = 10.8, mean BMI = 21.3 kg/m2; S.D. = 2.6)) to assess the validation of the SDHQ.
The nutrient intakes per day based on the twelve one-day records were calculated using the same food consumption table as for the DHQ. Mean values from twelve one-day records were used for the following validation assessment. Log transformation was employed for intake values except for total calorie, fat and carbohydrate. We compared means of each intake value between the twelve one-day records and the SDHQ by paired t-test. Pearson's correlation coefficients between estimated intake values from one-day records and the SDHQ were calculated. Calorie-adjusted correlation coefficients were also obtained by calculating correlation coefficients between respective residuals from a regression model with total calorie intake as the independent variable and each absolute nutrient intake as the dependent variable 5). The analyses were performed by SAS Ver 5.18 6).

RESULTS
The SDHQ included 31 food items ( Table 1). Coefficients of determination (R2) were 0.8 and greater for all nutrients (for partial regression coefficients and partial R2 for each nutrient, Appendix).
The mean of estimated nutrient intakes from the SDHQ were significantly higher than those from twelve one-day records (Table 2), except for retinol and vitamin D. Similar trends were observed in both sexes (data not shown).
Pearson's correlation coefficients between nutrient intakes from twelve one-day records and the SDHQ were greater than 0.30, except for fat and monounsaturated fat ( Table 2). Calorie adjustment increased the correlation coefficients, except for fat, carbohydrate, retinol, vitamin A, salt, polyunsaturated fat and monounsaturated fat. Correlation coefficients between nutrient intakes from the SDHQ and twelve one-day records were greater than 0.30 in each sex except for fat, retinol, vitamin A, saturated fat and monounsaturated fat. After calorie adjustment, such high correlations were observed in each sex except for fat, retinol, vitamin A, salt, polyunsaturated fat and monounsaturated fat (data not shown). Appendix. Partial regression coefficients and partial coefficients of determination (R2) of each item and adjusted R2 for 17 nutrients (1) Appendix. Partial regression coefficients and partial coefficients of determination (R2) of each item and adjusted R2 for 17 nutrients (2) Appendix. Partial regression coefficients and partial coefficients of determination (R2) of each item and adjusted R2 for 17 nutrients (3) Appendix. Partial regression coefficients and partial coefficients of determination (R2) of each item and adjusted R2for 17 nutrients (4)

DISCUSSION
The correlation coefficients between the SDHQ and twelve one-day records were greater than 0.3 except for fat and monounsaturated fat. In previous reports 5, 14)lay, their results indicated almost same correlation coefficients (r, 0.17 -0.76) between their short food frequency questionnaires and their food records as our results, and higher correlation coefficients for fat (r, 0.27 -0.67) than our results. In general, our results suggest that the SDHQ is a valid method for estimating nutritional intakes. The SDHQ has some limitations. It estimates only 17 nutrients intakes and is not able to estimate long term food or food group intakes. However, it might reduce the burden on respondents. The SDHQ might be useful as an additional questionnaire to a long questionnaire of epidemiologic studies in which nutrient intake is assessed as a covariate.
Many previous studies a 4,5). 7-12) reported reduction of food items to develop their food frequency questionnaire. Most of them employed a stepwise linear regression method to reduce the number of items. The stepwise regression method would be useful to select food items which contribute to the betweenperson variance. However, it might select high correlating items which does not contain the target nutrient). We should be careful about this fact, but this procedure might be adequate to develop the SDHQ for classifying subjects according to nutrient intakes. We selected items those cumulative RI exceed 0.6 for each nutrient. This cut off value of RI was lower than those of previous reports 5.8)11). However, RI value of the final form of SDHQ which excluded duplication was 0.8 or greater for each nutrient, which seems comparable with these reports.
It has been suggested adjustment for total calorie intake improve the accuracy of specific nutrient measurements 13). We also observed increases in correlation coefficients for some nutrients as previous studies 5, 14). However, the adjustment decreased in the correlation coefficients of fat related nutrients (fat, retinol, polyunsaturated fat and monounsaturated fat) and carbohydrate. Unadjusted correlation coefficients of these fat related nutrients also were lower than other nutrients. One reason might be due to measurement error by the subjects who reported higher calorie intakes in the SDHQ and / or one-day records. Heitmann et al. observed underreporting their calorie intakes especially their fat calorie intakes in obese peoples 16) The subjects who reported higher calorie intakes in the SDHQ also might underreported their food intakes containing much fat related nutrients or carbohydrate like obese peoples. We observed overreporting of their nutrients intakes in the SDHQ compared with one-day records. The subjects may recall more than usual intakes, because they concentrate on only 31 items in the SDHQ.
We validated the SDHQ using twelve one-day records which were conducted once a month during a year. We had not validated the one-day records method. The 24-hour recall method has been often employed in major nutritional investigations. Because our one-day records less depend on recall, our method might be less affected by bias and errors due to recall. We considered seasonal variations of food intakes also to estimate average intakes during the previous year. Owaki et al. mentioned the importance of seasonal variations of some nutrient intakes for Japanese population 15). Twelve one-day records which we employed would cancel out seasonal variations of food intakes. Previous reports used 4 seven-day records recorded in four seasons 5) and 2 seven-day records recorded in almost same season 14) as a standard. Willett mentioned that canceling out day-to-day variation of food intake should be needed to estimate true long-term intake values especially in industrialized countries 13). Such long-term consecutive records of nutritional intakes might be a better standard to validate the questionnaire method. A further validation study is needed using dietary records for a longer term as a standard.