Within- and Between-Individual Variation in Energy and Nutrient Intake in Japanese Adults: Effect of Age and Sex Differences on Group Size and Number of Records Required for Adequate Dietary Assessment

Background Information on within- and between-individual variation in energy and nutrient intake is critical for precisely estimating usual dietary intake; however, data from Japanese populations are limited. Methods We used dietary records to examine within- and between-individual variation by age and sex in the intake of energy and 31 selected nutrients among Japanese adults. We also calculated the group size required to estimate mean intake for a group and number of days required both to rank individuals within a group and to assess an individual’s usual intake, all with appropriate arbitrary precision. A group of Japanese women (younger: 30–49 years, n = 58; older: 50–69 years, n = 63) and men (younger: 30–49 years, n = 54; older: 50–76 years, n = 67) completed dietary records for 4 nonconsecutive days in each season (16 days in total). Results Coefficients of within-individual variation and between-individual variation were generally larger in the younger group than in the older group and in men as compared with women. The group size required to estimate a group’s mean intake, and number of days required to assess an individual’s usual intake, were generally larger for the younger group and for men. In general, a longer period was required to rank women and older adults. Conclusions In a group of Japanese adults, coefficients of within-individual variation and between-individual variation, which were used to estimate the group size and number of records required for adequate dietary assessment, differed by age, sex, and nutrient.


INTRODUCTION
Fluctuations in daily dietary intake values, which frequently hamper analysis of nutritional data, result from within-and between-individual variation. [1][2][3] Within-individual variation is subject to several factors such as true day-to-day variation, variation by day of the week and season, and residual variation, including measurement error. Between-individual variation is strongly influenced by factors such as age and sex. [1][2][3][4][5][6][7][8] These variations should be considered whenever dietary intake is assessed in individuals and groups. 3,9 Properly designed nutritional research that includes dietary assessment should thus consider the number of subjects required in 1 group (group size) and the number of days required to implement the assessment efficiently. 3,10 These variables can be estimated using within-and between-individual variation of nutrient intake. [1][2][3]7 Dietary assessment is usually conducted for 1 of 3 purposes: (1) to compare the mean intake of different groups, (2) to rank individuals within a group, or (3) to assess an individual's usual intake. Thus, knowledge of within-and between-individual variation is required in order to determine group size in studies comparing mean intake between groups, 7 and the ratio of within-to betweenindividual variation is required in order to determine the number of days required for dietary assessment in studies that assess diet-disease associations using rankings of subjects within a group (eg, in estimating relative risk using quartile categorizations). 1,5,11 Moreover, within-individual variation influences the number of days required to assess the usual intake of individuals (eg, to establish the true nature of doseresponse). 1,3,9 The magnitude of within-and between-individual variation in nutrient intake is largely determined by cultural and ecologic factors. 2,3,12 The group size and number of days required for precise estimation of usual nutrient intake has been studied, but results have differed, 7,13 and these variables might differ by age, sex, and country, due to different dietary habits. 13 However, investigation of these issues has been limited in Japan. 4,14,15 Here, we examined within-and between-individual variation in dietary intake by age and sex among Japanese adults. We assessed energy and 31 selected nutrients derived from dietary records (DRs) that were maintained for 4 nonconsecutive days in each season (16 days in total). We also estimated the group size required to estimate a group's mean intake and the number of days required to rank individuals within a group and to assess an individual's usual intake with adequate precision.

METHODS Subjects
The study was conducted in 4 areas in Japan that differed in geographic conditions and dietary habits, namely Osaka (Osaka City: 11 743 persons/km 2 ; urban), Nagano (Matsumoto City: 786 persons/km 2 ; rural inland), Tottori (Kurayoshi City: 285 persons/km 2 ; rural coastal), and Okinawa (Ginowan City: 4446 persons/km 2 ; urban island), 16 between November 2002 and September 2003. [17][18][19][20] We recruited apparently healthy women aged 30 to 69 years who were willing to participate with a cohabiting husband. The subjects were volunteers and were asked by local staff to participate in the study. Subject recruitment was continued until a sufficient number of participants was obtained. In each of the 4 areas, each 10-year age band (30-39, 40-49, 50-59, and 60-69 years) included 8 women; the age of the husband was not considered. Thus, a total of 128 women and 128 men were invited. Dietitians were excluded from the study. None of the subjects had recently received dietary counseling from a doctor or dietitian or had a history of educational hospitalization for diabetes or nutritional education from a dietitian. Before the study, group orientations were held to explain the study purpose and design. Written informed consent was obtained from each subject. The study did not undergo ethical approval because it was conducted before ethical guidelines for epidemiologic research were enforced in Japan. However, use of data from this study was approved by the Ethics Committee at the University of Tokyo Faculty of Medicine (No. 3421). A total of 121 women aged 30 to 69 years and 121 men aged 30 to 76 years completed 16-day DRs and were included in the present analysis.  [17][18][19][20] The 4 recording days consisted of 3 randomly selected weekdays and 1 weekend day. During the orientation session, local staff (registered dietitians) gave subjects both written and verbal instructions on how to keep the dietary record, using a completed recording sheet as an example. Each couple was given blank recording sheets and a digital scale (Tanita KD-173, ±2 g precision for 0-250 g and ±4 g precision for 251-1000 g). Subjects were also instructed on how to weigh each food item and drink and were asked to record and weigh all foods and drinks consumed on each recording day. When weighing was difficult (eg, when eating out), we instructed them to record the size and quantity of foods they ate as precisely as possible, using household measures. For each recording day, the subjects were asked to fax the completed forms to the local staff. The staff reviewed the submitted forms and, if necessary, asked the subject to augment and/or modify records by telephone or fax. The responses were faxed or, in some cases, handed directly to the staff.
All collected records were checked by trained registered dietitians in each local center and then again in the data center. The coding of records and conversion of measurements into grams were performed by trained registered dietitians in the survey center in accordance with uniform procedures. A total of 1398 food and beverage items appeared in the dietary records. Intake of energy and 31 selected nutrients was assessed based on the estimated intake of all items and the Standard Tables of Food Composition in Japan. 21 Anthropometric measurements, physical activity level, and reporting adequacy of reported energy intake Body height and weight were measured to the nearest 0.1 cm and 0.1 kg, respectively, with subjects wearing light clothing and no shoes. Body mass index (BMI) was calculated as body weight (kg) divided by the square of body height (m). Basal metabolic rate (BMR) was calculated for each subject from age, measured body height, and weight with the use of the equations of Ganpule et al. 22 Physical activity level (PAL) was obtained from a questionnaire that queried subjects on their occupation and leisure-time activity. PAL was classified into 1 of 4 categories, and the categorical classification of PAL was then converted to 1.5 for sedentary or light, 1.75 for active or moderate, and 2.0 for vigorous and heavy PAL (Ministry of Health, Labour and Welfare of Japan, 2009). 23 Estimated energy requirement (EER) was calculated as the product of PAL and BMR. We used the ratio of reported energy intake (EI) to EER (EI/EER) as an indicator of the adequacy of energy intake reporting and defined a ratio of 1.0 as adequate reporting for the group.

Statistical analysis
All statistical analyses were performed separately for women and men in 2 age groups (younger: 30-49 years for both women and men; older: 50-69 years for women and 50-76 years for men) using SAS statistical software, version 9.2 (SAS Institute Inc., Cary, NC, USA). Means, coefficients of within-individual variation (CV w ) and between-individual variation (CV b ), variance ratio, required group size, and required number of days were compared between age groups and sexes.
Means, SD, CV w , and CV b for intakes were calculated. Variances of intake were estimated into 2 sources by 1-way ANOVA: (1) between-individual variance (σ b 2 ) and (2) within-individual variance (σ w 2 ) (ie, day-to-day variation unaccounted for by other sources). Estimates of σ w 2 and σ b 2 were calculated by setting mean squares equal to their expected values.
We used untransformed data to analyze within-and between-individual variation in energy and all nutrients because a previous study showed that the estimated relative contribution of sources of variance was not considerably affected by logarithmic transformation 2 and because other previous studies showed that a logarithm and Box-Cox transformation did not improve the assumption of homoscedasticity across covariates in the models, that estimates based upon transformed nutrient data were difficult to interpret meaningfully, and that back-transformation would introduce bias to variance estimates. 24,25 The group size of DR (G) required to estimate mean intakes with 95% CIs within the specified percentage deviation (D 0 ) of group mean from group usual ("true") mean intake was calculated using the following formula 2 : G = 1. ]. The number of days of DR (N R ) required to ensure a specified level of correlation coefficient (r) between observed and unobserved usual ("true") mean intakes in individuals was calculated using the following formula 1,7 : N R = [r 2 /(1 − r 2 )] × VR, where VR is the variance ratio as determined by σ w 2 /σ b 2 . For this analysis, r is thus a measure of confidence of ranking or classification of individuals into fractions (eg, fourths).
The number of days of DR (N I ) required to estimate mean intakes with 95% CIs within the specified percentage deviation (D 1 ) of individual mean from usual ("true") mean intake based on CV w was calculated using the following formula 1-3 : N I = (1.96 × CV w /D 1 ) 2 . Table 1 shows the physical characteristics of men and women in the 2 age groups. The mean value of EI/EER was around 1.0 in all groups; the smallest value, 0.94, was for younger men, and largest value, 1.08, was for older women. Table 2 shows means, SD, CV w , CV b , and VR of daily intake of energy and 31 selected nutrients. Mean intake was larger in the older than in the younger group in both sexes for most nutrients and larger in men than in women for energy and all nutrients. CV w was larger than CV b for energy and most nutrients irrespective of age or sex. CV w was larger in the younger than in the older group for both women (for energy and 26 nutrients; ±1%-65% differences) and men (for energy and 28 nutrients; ±2%-25% differences). The findings for CV b were similar among both women (for energy and 26 nutrients; ±5%-12% differences) and men (for energy and 29 nutrients; ±8%-11% differences). Additionally, CV w was larger in men than in women for both the younger (for energy and 21 nutrients; ±8%-4% differences) and older groups (for energy and 22 nutrients; ±7%-51% differences). Similar findings were obtained in CV b for both the younger (for energy and 29 nutrients; ±1%-8% differences) and older groups (for energy and 18 nutrients; ±4%-8% differences). VR was greater than 1 for all except water (in younger women and men and older men) and carbohydrate (in younger men). In contrast to the results for CV w and CV b , VR was larger in the older than in the younger group for both women (for energy and 21 nutrients) and men (for energy and 26 nutrients) and larger in women than in men for both the younger (for energy and 27 nutrients) and older groups (for energy and 16 nutrients).  Table 3 shows the group size required to estimate mean intake of energy and nutrients with 95% CIs within a specified (ie, 2.5%, 5%, 10%, and 20%) deviation of a group's mean from the group's usual ("true") mean intake by DR. The group size required to determine the mean intake of the group was larger in the younger than in the older group for both women (for energy and 29 nutrients) and men (for energy and 30 nutrients) and was larger in men than in women for both the younger (for energy and 26 nutrients) and the older groups (for energy and 22 nutrients). Table 4 presents the number of days required to ensure specified (ie, 0.75, 0.80, 0.85, 0.90, and 0.95) correlation coefficients between observed and usual ("true") mean intake of energy and nutrients by DR. The number of days required to rank individuals within a group by intake was larger in the older than in the younger group for both women (for energy and 20 nutrients) and men (for energy and 25 nutrients) and was larger in women than in men for both the younger (for energy and 29 nutrients) and older groups (for energy and 16 nutrients). Table 5 shows the number of days required to assess mean intake of energy and nutrients with 95% CIs within a specified (ie, 5%, 10%, 20%, and 30%) deviation of an individual's mean from usual ("true") mean intake by DR. The number of days needed to assess the usual intake of individuals was larger in the younger than in the older group for both women (for energy and 26 nutrients) and men (for energy and 28 nutrients) and was larger in men than in women for both the younger (for energy and 20 nutrients) and older groups (for energy and 21 nutrients).

DISCUSSION
In this study of Japanese women and men, we found that coefficients of within-individual variation and betweenindividual variation were generally larger in the younger Table 2. Mean daily energy and nutrient intake, coefficients of variation, and within-to between-individual variance ratios according to sex and age group Women (n = 121) Men (n = 121) Younger a (n = 58) Older a (n = 63) Younger a (n = 54) Older a (n = 67)  ). e Sum of β-carotene, α-carotene/2, and cryptoxanthin/2. f Sum of retinol, β-carotene/12, α-carotene/24, and cryptoxanthin/24. g Sum of eicosapentaenoic acid, docosapentaenoic acid, and docosahexaenoic acid. group than in the older group, whereas variance ratio was larger in the older group than in the younger group. Similarly, both CV w and CV b were generally larger in men than in women, whereas VR was larger in women than in men. To our knowledge, this study is the first to examine within-and between-individual variation in dietary intake with respect to both age and sex in a Japanese population and in Asian men.
The results of this study are comparable with those of previous studies in Japan, 4,14,15 namely, CV w was larger than CV b , and CV w , CV b , and VR were relatively small for energy, carbohydrate, protein, and water, intermediate for minerals, dietary fiber, and fat, and large for fatty acids, cholesterol, and vitamins. Ogawa et al used four 3-day DRs to investigate women (aged 47-76 years) and men (aged 45-77 years) living in a rural area. 15 Their results for CV w and CV b were similar to our estimates. Egami et al used four 4-day DRs to assess women and men (aged >40 years) living in a coastal area. 14 CV w was generally larger than in our results, whereas CV b was smaller. Tokudome et al used four 7-day DRs to investigate female dietitians (aged 32-66 years). 4 CV w and CV b were generally smaller than in our study, possibly due to differences in eating patterns between their and our groups, which arose from the greater nutritional knowledge of their subjects.
Our findings are also consistent with those of several studies that examined CV w and CV b by age or sex. 7,15,26 In a study of UK adults (a comparison among 4 groups categorized by sex and age, with younger groups aged 18-57 years using 7-day DRs vs older groups aged 60-80 years using three 7day DRs), CV w and CV b were larger in the younger than in the older group for both sexes. 7 In studies of Japanese adults living in a rural area (mentioned above), 15 UK adults (mentioned above), 7 and US elderly adults (aged >60 years using 3-day DRs), 26 CV w and CV b were larger in men than in women. In a study of Chinese women (aged 40-59 years vs 60-70 years using 24-h dietary recall), while CV w was Table 3. Group size required to estimate mean intake of energy and nutrients with 95% CIs within the specified % deviation (D 0 ) of a group's mean from the group's usual ("true") mean intake by dietary record according to sex and age group a  Energy  442  111  28  7  302  76  19  5  497  124  31  8  353  88  22  6  Protein  569  142  36  9  448  112  28  7  639  160  40  10  476  119  30  7  Fat  980  245  61  15  884  221  55  14  1186  297  74  19  976  244  61  15  Carbohydrate  517  129  32  8  352  88  22  5  556  139  35  9  400  100  25  6  Dietary fiber  1081  270  68  17  937  234  59  15  1145  286  72  18  872  218  54  14  Water  520  130  32  8  473  118  30  7  732  183  46  11  448  112  28  7  Sodium  889  222  56  14  881  220  55  14  1032  258  64  16  846  212  53 7191  1798  449  112  5235  1309  327  82  6585  1646  412  103  6254  1563  391  98  Folate  2001  500  125  31  1219  305  76  19  2147  537  134  34  1741  435  109  27  Vitamin C  2261  565  141  35  1483  371  93  23  2564  641  160  40  1980  495  124  31  SFA  1344  336  84  21  1243  311  78  19  1789  447  112  28  1251  313  78  20  MUFA  1284  321  80  20  1222  305  76  19  1471  368  92  23  1378  344  86  22  PUFA  1155  289  72  18  1127  282  70  18  1245  311  78  19  1162  291  73  18  n-6 PUFA  1244  311  78  19  1290  323  81  20  1362  341  85  21  1326  332  83  21  n-3 PUFA  2170  543  136  34  2224  556  139  35  2301  575  144  36  2332  583  146  36  Marine origin n-3 PUFA e  9315  2329  582  146  7134  1784  446  111  10 124 2531  633  158  6624  1656  414  103  Cholesterol  2000  500  125  31  1862  465  116  29  1803  451  113  28  1715  429  107  consistent with our study, CV b was larger in the older than in the younger group. 5 Additionally, in studies of Japanese adults living in a coastal area (mentioned above), 14 Korean elderly (mean [SD] age 70.4 [5.8] years using 5-or 6-day 24-h dietary recall), 12 and Canadian adults (aged 25-44 years using 24-h dietary recall), 2,27 CV w and CV b were larger in women than in men. These inconsistent results in some previous studies may be due to differences in study design: these studies 2,5,12,27 used 24-h dietary recall, whereas we used DR. Cultural factors also likely played a role. 2,3 The present results have implications for the design and interpretation of dietary assessment. First, among older adults and women, nutrient intake may be more homogeneous from day-to-day and among subjects than for younger adults and men, because smaller CV w and CV b were observed in older groups and in women than in their respective counterparts. Thus, as compared with men and younger adults, women and older adults may require a smaller group size and fewer days to assess the group's and individual's usual nutrient intake. Second, subjects can be more precisely ranked in groups of younger adults and/or men, because a smaller VR was observed in these groups. A smaller VR means that σ w 2 is relatively small compared with σ b 2 and that the difference in intake between individuals can be more easily distinguished. Therefore, if dietary assessment is conducted in individuals or groups by the same methods (number of days and group size) regardless of age or sex, the level of precision of the assessment will differ among the individuals or groups. If an analysis includes estimates of intake with a low level of precision, even in only 1 group, this may decrease the power of the statistical analysis and lead to misinterpretation of the association between dietary factors and an outcome. 2,7,9,12 Third, regardless of age or sex, a large CV w means that many DR days would be required to characterize an individual's usual intake-for example, 4 to 481 days would be needed to achieve within 20% deviation for younger women. Therefore,  9  12  17  29  62  11  15  22  36  78  6  9  13  20  44  7  9  14  22  49  n-3 PUFA  10  14  20  33  72  12  16  23  38  83  8  12  17  28  60  10  13  19  32  69  Marine origin n-3 PUFA e 21  29  42  70  151  18  25  37  60  131  17  24  35  58  126  13  18  26  42  92  Cholesterol  8  11  16  25  55  8  12  17  28  61  6  8  12  19  42  6  8  11  18  40 Abbreviations: SFA = saturated fatty acids; MUFA = monounsaturated fatty acids; PUFA = polyunsaturated fatty acids. a Number of days of dietary record = [r 2 /(1 − r 2 )] × VR, where r = unobservable correlation coefficient between observed and usual ("true") mean intakes of individuals and VR = within-individual/between-individual variance ratio (σ w use of an alternative method (eg, a semi-quantitative food frequency questionnaire) that can estimate usual intakes over a longer period than DR or dietary recall may be necessary to accord with the study objective, study design, demographic characteristics of the population, and available resources. 3,6,12,15 Several limitations of this study warrant mention. First, the generalizability of our results is hampered by the fact that the present subjects were not randomly sampled from the general Japanese population but were instead volunteers and possibly health-conscious. As we lacked information on the subjects' characteristics, including education and occupation, we could not determine how such characteristics influenced our findings. Mennen et al 13 assumed that the dietary recall of subjects who completed a protocol is more precise (smaller CV w ) than that of subjects who dropped out. Hebert et al 11 suggested that CV b is smaller in a population with higher socioeconomic status (SES). Thus, because of precise recording, CV w might have been smaller in our volunteers than in the general population consuming a similar diet. CV b might have been smaller because of limited variation in some variables (eg, health-consciousness). If so, the group size required to estimate a group's mean intake in the general population would be larger than the estimates observed here (Table 3). Additionally, the number of days required to precisely estimate an individual's usual intake in the general population would be larger than the estimates observed here (Table 5). Conversely, as we did not know whether VR was lower or higher in our volunteers than in the general population, the number of days required to rank individuals based on their intakes within the general population is unclear, that is, we cannot conclude that the required number of days is larger or smaller than the estimates observed here (Table 4).
Second, the subjects were married men and women living together, who likely frequently have the same meals. This implies that the CV w and CV b of men in this study might be underestimated as compared with the general male population because the daily menu is probably usually decided by Table 5. Number of days required to assess mean intake of energy and nutrients with 95% CIs within the specified % deviation (D 1 ) of an individual's mean from usual ("true") mean intake by dietary record according to sex and age group a Women (n = 121) Men (n = 121) Younger b (n = 58) Older b (n = 63) Younger b (n = 54) Older b (n = 67) 10%  20%  30%  5%  10%  20%  30%  5%  10%  20%  30%  5%  10%  20%  30%   Energy  65  16  4  2  52  13  3  1  69  17  4  2  53  13  3  1  Protein  100  25  6  3  85  21  5  2  99  25  6  3  87  22  5  2  Fat  188  47  12  5  187  47  12  5  211  53  13  6  198  49  12  5  Carbohydrate  65  16  4  2  53  13  3  1  67  17  4  2  61  15  4  2  Dietary fiber  176  44  11  5  161  40  10  4  178  45  11  5  143  36  9  4  Water  65  16  4  2  44  11  3  1  84  21  5  2  53  13  3  1  Sodium  174  44  11  5  181  45  11  5  195  49  12  5  178  45  women, who in our study had a smaller CV w and CV b . Third, although we compared within-and between-individual variation between sexes and age groups (younger vs older), several unanticipated confounding factors, such as SES, might be present in our analysis. If the distribution of SES differs between sexes or age groups, and SES has an effect on dietary habits, it should be adjusted for in the analysis. However, we designed the study so as to consider important confounding factors that may affect the comparisons. For example, sex itself is an important confounding factor in a comparison between age groups, and age is the same in a comparison between sexes. To address this problem, we recruited the same number of subjects for each sex and age category. Living area, season, and timing of data collection (weekday or weekend day) are other possible confounding factors, and they were equalized between sexes and age groups. [1][2][3][4]28,29 Finally, DR is susceptible to measurement error due to erroneous recording and potential changes in eating behavior. 3 However, the adequacy of reported energy intake was likely adequate at the group level, given that the mean value of EI/EER was around 1.0.
In conclusion, the present study of Japanese adults showed that CV w and CV b were larger in a younger group than in an older group and larger in men than in women for energy and most nutrients. Precise estimation of usual nutrient intakes requires consideration of differences not only in CV w and CV b by age and sex, but also in group size and number of days estimated using CV w and CV b . The present findings may have important implications for the design and interpretation of dietary assessment in Japanese adults.

ONLINE ONLY MATERIAL
Abstract in Japanese.

ACKNOWLEDGMENTS
This work was supported by grants from the Japanese Ministry of Health, Labour and Welfare. All the authors contributed to the preparation of the manuscript and approved the final version submitted for publication. A.F. performed statistical analyses and wrote the manuscript. K.A. and K.M. assisted in writing and editing the manuscript. S.S. contributed to the concept and design of the study, study protocol, and data collection, and assisted in writing and editing the manuscript. H.O., N.H., A.N., H.T., A.M., M.F., and C.D. were involved in the study design, data collection, and data management.
Conflicts of interest: None declared.