Validity of a Self-Administered Food Frequency Questionnaire for Middle-Aged Urban Cancer Screenees: Comparison With 4-Day Weighed Dietary Records

Background The validity of estimates of dietary intake calculated using a food frequency questionnaire (FFQ) depends on the specific population. The 138-item FFQ used in the 5-year follow-up survey for the Japan Public Health Center-based Prospective Study was initially developed for and validated in rural residents. However, the validity of estimates based on this FFQ for urban residents, whose diet and lifestyle differ from those of rural residents, has not been clarified. We examined the validity of ranking individuals according to level of dietary consumption, as estimated by this FFQ, among an urban population in Japan. Methods Among 896 candidates randomly selected from examinees of cancer screening provided by the National Cancer Center, Japan, 144 participated in the study. In 2007–2008, at an average 2.7 years after cancer screening, participants were asked to respond to the questionnaire and to provide 4-day weighed diet records (4d-DRs) for use as the reference intake. Spearman correlation coefficients (CCs) between the FFQ and 4d-DR estimates were calculated, after correction for intraindividual variation of 4d-DRs. Results The median (range) deattenuated CC for men and women was 0.57 (0.23 to 0.89) and 0.47 (0.08 to 0.94), respectively, across 45 nutrients and 0.51 (0.10 to 0.98) and 0.51 (−0.36 to 0.88) for 43 food groups. Conclusions Although the FFQ was developed for a rural population, it provided reasonably valid measures of consumption for many nutrients and food groups in middle-aged screenees living in urban areas in Japan.


INTRODUCTION
Accuracy in measuring individual dietary intake is an important issue in the analysis and evaluation of results from epidemiologic studies of the association between diet and disease. Food frequency questionnaires (FFQs) provide a view of usual food or nutrient intake over time and have been developed and validated in target populations of epidemiologic studies. 1 Because the foods listed in an FFQ are selected according to their percentage contribution to the total consumption of nutrients among representatives of the target population for whom the FFQ is to be used, they might not necessarily reflect the foods eaten by a different population. Further, accuracy in remembering foods consumed appears to differ by education level and the degree of interest in diet. 1 The validity of FFQ estimates of dietary intake therefore appears to depend on the specific population.
The FFQ used for the Japan Public Health Center-based Prospective Study 5-year follow-up survey was developed for use with residents of rural cohort areas. 2 Of these resident, 27% worked in management, clerical, sales, or services, and 21% were employed in the agriculture, forestry, and fisheries sector. 3 Further, the FFQ was validated among subsamples of these rural residents. [4][5][6] It is therefore unclear whether this FFQ is accurate in estimating dietary intake among Japanese with an urbanized lifestyle. In addition, to our knowledge no such validation study has been restricted to an examination of subjects living in urban and adjacent areas. 7 To confirm the suitability of this FFQ for use in epidemiologic studies of cancer screenees at the National Cancer Center, such as the participants in the Colorectal Adenoma Study in Tokyo, we evaluated the validity and reproducibility of ranking individuals by levels of dietary consumption-as estimated by this FFQ after minor modification-as a means of assessing dietary intake among middle-aged urban cancer screenees. 8

Study setting and participants
The study participants were selected from adults who underwent cancer screening at the Research Center for Cancer Prevention and Screening, National Cancer Center, "Japan from January 2004 through July 2006. Eligibility criteria were age between 40 and 69 years, residence in metropolitan Tokyo, and no previous or present diagnosis of cancer, cardiovascular disease, or diabetes mellitus. Eligible subjects were stratified by sex and age (40-49, 50-59, and 60-69 years) and randomly numbered for recruiting priority.
Among the 896 invited candidates, 187 (response rate: 20.9%) agreed to participate in the study. After excluding those who could not attend the study orientation, 144 participated in the study. As an incentive to participate, participants received a report of their results regarding the consumption of energy and nutrients based on 4-day dietary records, a small gift (an instrument for measuring the salt concentration of soup), and a free invitation to attend a class on healthful cooking. The study was approved by the Institutional Review Board of the National Cancer Center, Tokyo, Japan. All participants provided their written informed consent for participation, at the study orientation.

Data collection
The reference intake was 4-day weighed diet records (4d-DRs), which were obtained over 4 consecutive days during the period from May 2007 through April 2008. Before the start of data collection, all participants were invited to attend the study orientation, where the 4d-DR procedure was explained by trained dietitians. The self-administered FFQ was first administered during 2004-2006 at the time of cancer screening (FFQ0) and then during 2007-2008 at the orientation session (FFQ1).

Dietary assessment
The 4d-DR included 3 weekdays and 1 weekend day and was used as the reference method. Food portions were measured by each participant during meal preparation using supplied digital scales and measuring spoons and cups. For foods purchased or consumed outside the home, the participants were instructed to record the approximate quantity of all foods in the meal and/or the names of the product and company. Daily weighed records were faxed to the study office at the Research Center for Cancer Prevention and Screening, National Cancer Center on the morning after completion of that day's record. Trained dietitians checked the record with the examinee by telephone and coded the foods and weights. Stores and restaurants were asked about the recipes of certain meals eaten outside the home.
The FFQ consisted of 138 food and beverage items and 9 frequency categories, which ranged from almost never to 7 or more times per day (or to 9 glasses per day, for beverages), and asked about the usual consumption of listed foods during the previous year. The food list, which was initially developed for the Japan Public Health Center-based Prospective Study, 2 was modified for a middle-aged urban population as follows: 11 foods mainly consumed in specific areas (Okinawa and Nagano) or at specific times were excluded (luncheon meats, vivipara, qing-geng-cai [bok choy], leaf mustard, bitter gourd, chard, loofah, mugwort, yushi-tofu [soft, boiled tofu], calcium beverages, and beta carotene beverages), and 11 foods consumed throughout the year in urban areas were added (beef, stir-fried; chicken, stir-fried; chicken, stew; low-fat milk; Japanese amberjack; Welsh onion; eggplant; edible burdock; konnyaku foods [devil's tongue]; and jam, strawberry or marmalade). Portion size was specified for each food item, using 3 standard sizes: medium (the standard amount), small (50% smaller), and large (50% larger).
Intakes of energy, 45 nutrients, and 43 food groups were calculated using the Standardized Tables of Food Composition, Fifth revised edition 9,10 and a specially developed food composition table for isoflavones and lycopene in Japanese foods. 11,12 We collapsed the individual food items into 18 predefined food groups according, to the Food Composition Tables, and 25 stream-specific subgroups. The grouping scheme for subgroups, eg, cruciferous vegetables and red meat, was based on the similarity of nutrient profiles or culinary usage among the foods and was somewhat similar to that used in other studies.

Statistical analysis
The mean intake of each nutrient and food group estimated using the FFQ1 was compared to that estimated using the 4d-DR among the 143 participants who completed both. Percentage differences were calculated for each nutrient and food group by dividing the difference in intake on the FFQ1 from that on the 4d-DR by those using the 4d-DR. To determine the validity of the FFQ, Spearman rank correlation coefficients (CCs) between intake estimates of the FFQ1 and 4d-DR were calculated for crude and energy-adjusted values. Regression coefficients between nutrient intakes according to the FFQ1 and 4d-DR were calculated for energy-adjusted values to examine the degree of attenuation in a diet-disease association in a hypothetical study using the FFQ. 1 A residual model was used for energy adjustment. 1 We corrected the observed CCs for the attenuating effect of random intraindividual error from the usual intake of each energy and nutrient and each food group. The deattenuated value was corrected using the ratios of the within-to between-individual variances based on the 4-day DRs according to the following formula: where the observed en-CC x is the correlation in energyadjusted value for nutrient x, λ x is the ratio of within-to between-individual variance, and n is the number of dietary records (4 days). 1 To measure the validity of categorization, we computed the number of participants classified into the same, adjacent, and extreme categories by joint classification according to both quintiles using the FFQ1 and the 4d-DR. For reproducibility, CCs between the FFQ1 and FFQ0 were calculated for crude and energy-adjusted values for the 144 participants who completed both FFQs. We confirmed the cumulative percentage among the top 20 foods for energy, because food variety was important in confirming the extent to which the list of FFQ items could be covered. Percentages of the sum of energy by individual foods eaten to total energy during the 4 days were also calculated. All analyses were performed using SAS Version 9.1 (SAS Institute Inc., Cary, NC).

Participants in the validation study
Age distribution (40s, 50s, 60s) at recruitment (2004)(2005)(2006) was n = 11, 29, and 29, respectively, for men and n = 16, 30, and 29 for women. Mean body mass index (standard deviation) for men and women was 23.5 (2.5) and 21.5 (2.5), respectively. Overall, 51% of the participants were employed in management, clerical, sales, or services, and 2% worked in agriculture, forestry, or fisheries. Table 1 shows daily intakes of energy and 45 nutrients, as assessed by 4d-DR and FFQ1, percentage differences between FFQ1 and the 4d-DR, and their correlations among men and women. Although estimated intake levels for energy were very similar between the 2 methods (difference: −6% for men, 2% for women), the percentage difference in nutrient intake between the 4d-DR and FFQ1 varied from −35% and −20% for beta-carotene to +99% and +198% for cryptoxanthin in men and women, respectively. The CCs of the crude values varied from 0.12 for retinol equivalents to 0.71 for daidzein in men and from 0.10 for polyunsaturated fatty acid to 0.57 for vitamin K in women. The median across the 45 nutrients was 0.43 for both men and women. After energy adjustment and deattenuation, the median CC improved to 0.57 in men and 0.47 in women. The regression coefficient for nutrient intake varied from 0.16 for retinol equivalents to 0.61 for copper in men and from 0.05 for cryptoxanthin to 0.63 for pantothenic acid in women (data not shown). Table 2 shows daily intakes of 43 food groups assessed by the 4d-DR and FFQ1, the percentage difference between

Mean intakes and FFQ validity
where λ x is the ratio of within-to between-individual variance for nutrient x, and n is number of dietary records; observed CCs were based on energy-adjusted values other than energy intake. FFQ1 and 4d-DR, and their correlations among men and women. The percent difference in intakes between the 4d-DR and FFQ1 varied from −83% and −86% for seasonings and spices in men and women, respectively, to +111% for other cereals in men and +153% for citrus fruit in women. The CCs of the crude values varied from 0.04 and −0.28 for seasonings to 0.81 and 0.82 for coffee in men and women, respectively. The medians across 43 food groups for men and women were 0.45 and 0.35, respectively. After energy adjustment and deattenuation, the median CC slightly improved to 0.51 (varying from 0.10 for seasonings to 0.98 for noodles) in men and 0.51 (varying from −0.36 for seasonings to 0.88 for coffee) in women.

Joint classification by quintile
We conducted further analysis to compare FFQ1 with the 4d-DR based on joint classification by quintile. Most nutrients and food groups were classified into the opposite extreme categories by 5% or less of men or women, with a corresponding median value for men and women of 1% and 3%, respectively, for nutrients, and of 3% and 3%, respectively, for food groups (Supplemental Tables 1  and 2). In contrast, retinol for men and women showed a relatively high percentage of extreme categories by joint classification (6% and 12%, respectively) and a relatively low CC (0.32 and 0.11, respectively) and regression coefficient (0.18 and 0.15, respectively). Further, cryptoxanthin for women showed a relatively low percentage of the same and adjacent categories (53%) and a relatively low CC (0.07) and regression coefficient (0.05).

Reproducibility
We also examined the reproducibility of dietary intake estimated by 2 identical FFQs (  Tables 3 and 4).
Percentage contributions of the top 20 foods to total energy Finally, we conducted an additional analysis of the cumulative percentage contributions of the top 20 foods for energy, based on the 4d-DRs, to assess the foods listed in the FFQ. The cumulative percentage of the top 20 foods for energy was 44.0% and 41.0% for men and women, respectively (Supplemental Table 5).
where λ x is the ratio of within-to between-individual variance for nutrient x, and n is number of dietary records; observed CCs were based on energy-adjusted values other than energy intake.
-: not applicable for calculation.

DISCUSSION
We examined the validity of ranking middle-aged urbandwelling cancer screenees in Japan by level of dietary intake using an FFQ, with 4-day DR data as the reference method.
The FFQ was initially developed and validated in rural populations. As compared with reference intakes, differences in mean absolute consumption based on the FFQ varied and tended to be underestimated. However, using the FFQ, dietary assessment of many nutrients and food groups showed moderate validity and reproducibility in ranking urban residents, whose diet and lifestyle might differ from those of rural residents.
In comparison with 4d-DRs corrected for intraindividual variance, for most nutrients, the validity of the FFQ was similar to or better than that observed in a comparison with 28-day weighed diet records among the rural residents for which the FFQ was developed. 6 In that initial validation study, median CCs for energy and 45 nutrients were 0.43 and 0.39 for men and women, respectively, and 0.38 and 0.32 for 19 main food groups. Evaluation of diet might be complicated by the apparently wider variety of foods eaten by urban as compared with rural residents in Japan (percent energy from cereal areas among the former was less than that among the latter 13 ), as has been seen in China 14 and Morocco, 15 although we saw no large difference in the validity of intakes, as estimated by the FFQ, between urban and rural populations in the present study.  36  77  3  23  60  5  Other fruit  36  77  1  30  69  1  Fungi  33  71  3  24  62  3  Algae  33  58  6  22  64  3  Fish and shellfish  28  71  1  23  61  3   Meats  38  78  6  28  66  7  Processed meat  28  67  1  32  72  3  Red meat  29  71  6  32  62  1  Poultry  25  61  3  28  54  5  Eggs  36  74  1  26  61  3  Milk and dairy products  41  78  3  35  78  0  High-fat milk  41  67  4  42  87  3  Low-fat milk  36  78  3  42  82  3  Fats and oils  30  61  4  39  72  0  Butter  32  64  6  34  70  5  Margarine and oils  29  61  4  34  66  1  Confectionaries  20  67  1  28  62  3  Japanese confectionery  22  64  4  14  55  5  Western confectionery  32  65  0  24  60  5  Alcoholic beverages  46  91  0  42  72  0  Nonalcoholic beverages  26  64  1  22  65  3  Green tea  48  80  0  27  65  0  Coffee  45  93  0  50  91  0  Other beverage  29  68  0  26  65  4  Seasonings and spices  16  49  4  16  42  12   MEDIAN  30  68  3  28  65  3 Wakai 7 reviewed 21 validation studies of FFQs developed in Japan and reported a median CC for energy intake of 0.46 (range 0.20 to 0.87) and a median CC among the 21 studies ranging from 0.22 (n-6 PUFA) to 0.58 (calcium) for energy and 24 nutrients. As compared with the median CCs among the 21 studies for energy and 24 nutrients and 17 food groups, the CCs for the many nutrients and food groups evaluated in the present study were not substantially different or higher. 7 Attenuation caused by measurement error may be unavoidable in studies that use FFQs to investigate diet-disease associations. For example, based on a true relative risk of 2.0, if the regression coefficient of intakes according to an FFQ and DR varies from 0.6 to 0.2, the corresponding relative risk is further attenuated from 1.52 to 1.15. 1 A similar attenuation might be unavoidable in any examination that uses the present FFQ to assess diet-disease associations. Further investigation will be needed to examine the effects of measurement error on diet-disease associations in an actual dataset.
The CC for energy intake among women in this study (deattenuated CC: r = 0.34) was lower than the median of 21 previous studies. Further, the CCs of intakes based on the FFQ appeared to be lower in women than in men for most of the energy and nutrients examined (median deattenuated CC: 0.57 and 0.47 for men and women, respectively). This lower correlation in women than men has been previously observed in Japanese and Western populations. 7,16 Sex differences in validity might be partly due to disparities in the ease of response to the structured questionnaire that result from differences between men and women in their interest in dietary habits. 4 Moreover, we also found that the cumulative percentage among the top 20 foods for energy was lower for women than for men and that it was also lower than among subjects during the development of the initial FFQ (men: 63.9%, women: 56.3%). 17 These results suggest that the lower validity for energy intake among women is partly attributable to a lower contribution to energy by individual foods in women than in men, as was seen among subjects during the development of the initial FFQ. Our study has several potential limitations. First, the response rate was not necessarily high, although the participants were randomly chosen and recruited from among cancer screenees. Selection bias, eg, a higher proportion of health-conscious subjects than in the actual population, was likely present, and thus the possibility of overestimating the validity of the FFQ cannot be ruled out. This response rate is nevertheless reasonable considering the burden posed by studies such as this. Second, reference intakes were based on 4-day values, versus the 28-day values used for the initial validation study of the FFQ. 4-6 A simple comparison of CCs might have been difficult, even though the present CCs were corrected for intraindividual variance.
Moreover, although the dietary records were completed on consecutive days (ie, in the same season), the FFQ inquired about the previous year. In addition, responses to the FFQ might have depended on the season, 18 and FFQ1 was conducted in the season during which the dietary record was done. Thus, the possibility that validity might have been overestimated cannot be ruled out, especially for seasonal foods such as fruit and vegetables. Third, in the examination of reproducibility, we were unable to consider the "true" change in diet. Although we would have liked to examine the effects of random variation in response to the FFQ, the effects of such variation and the "true" change of diet could not be readily separated, and both might have attenuated the reproducibility of the FFQ. 1 Therefore, the reproducibility of this FFQ (in random variation in response) might have been underestimated. In general, the advantages of FFQ-based dietary assessment are that the burden on participants is not heavy, an interviewer is unnecessary, costs are relatively low, 19 and the long-term diet can be ranked. In the present study, too, the median percentages of extreme categories based on joint classification by quintile between FFQ and DR for nutrients and food groups were 1% and 3%, indicating that this FFQ is suitable for the ranking of individuals with regard to intakes of many nutrients and food groups in large-scale studies of urban populations. However, some nutrient and food group intakes estimated by this FFQ showed relatively low CCs and regression coefficients; thus, any application of this FFQ to the examination of diet-disease associations, such as investigations of retinol and cryptoxanthin, must carefully address the problem of classification.

Supplementary
In conclusion, these results indicate that the present FFQ, which was initially developed for rural populations, provides reasonably valid measures in ranking middleaged cancer screenees in urban areas in Japan according to level of consumption of many nutrients and food groups.