Background: In 1997, we launched a large population-based cohort study, the Shanghai Women Health Study (SWHS), to investigate diet in relation to cancer risk among 74 943 Chinese women. Simultaneously, a dietary calibration study was conducted among 200 SWHS participants with biweekly 24-h dietary recall (24HDR) over a 1-y period in order to evaluate the validity and reliability of the SWHS food frequency questionnaire (FFQ).
Objective: The objectives of the current study were to evaluate the nature and magnitude of variances for intake of 26 nutrients among SWHS participants, and to estimate the number of 24HDR needed for estimate intake of the 26 nutrients examined in the study population.
Design: In all, 1-y biweekly 24HDR collected from 200 healthy, free-living women aged between 40 and 70 y, who participated in the SWHS dietary calibration study, was analyzed by mixed effects model and ratios of within-individual and between-individual dietary intake variances (w2/b2) were estimated.
Results: In agreement with reports from studies conducted in the US, we found that within-individual variances were larger than between-individual variances in dietary intake of most nutrients. The sum of all other variation (eg, weekday and weekend, seasonal, interviewer) accounted for less than 5% of total variation. Ratios of within- to between-individual variances (for log transformed data) ranged from 1.05 (carbohydrate) to 2.79 (fat) for macronutrient intake, 1.74 (niacin) to 8.48 (δ-tocopherol) for vitamin intake, and 1.35 (phosphorus) to 5.59 (sodium) for mineral intake.
Conclusions: The results of this study suggest that within- and between-individual differences in nutrient intake are the major sources of variation in this population of adult Chinese women. Cultural practices as well as seasonal supply and consumption patterns of vegetable intake are likely the major contributors to the variation. Implications of these results are discussed.
Sponsorship: This study is supported by the NIH grant R01CA70867.
The relationship between dietary intake and health outcomes has been a focus for scientific research and public concern for years (Stamler et al, 1996; Hunter, 1999; Willett, 2001). However, diet is extremely difficult to measure in free-living populations (Freudenheim & Marshall, 1988; Hebert & Miller, 1988; Subar et al, 2001). Accurate and precise assessment of dietary intake is a major challenge in the study of diet and health risk (Kohlmeier, 1995). Both repeated 24-h dietary recall interviews (24HDR) and dietary records have often been used as the best proximate estimate for the ‘true’ intake of food or nutrients within a defined period of time (Hebert et al, 1995; Buzzard et al, 1996; Willett, 1998). However, as the self-report method with least overall variability (Posner et al, 1992; Buzzard et al, 1996; Hebert et al, 1998) and the ability to be used in illiterate or semiliterate populations (Willett, 1998; Hebert et al, 2000), the 24HDR is often used as the comparison method in validation studies. These estimates derived from the 24HDR are then used as proxies for the underlying ‘true’ intake of foods or nutrients. They may then be used to estimate food or nutrient effects on specific disease outcomes, or they may be used to evaluate the validity of a specific food frequency questionnaire (FFQ) (Posner et al, 1992; Hebert et al, 1998; Stram et al, 2000).
Dietary intake in free-living persons differs from day to day and seasonally. Such variations may occur randomly or nonrandomly, and often are in response to environmental and cultural factors. Estimates of natural variation in dietary intake are needed to understand the nature of food and nutrient exposures in a human population. Also, such information is important for identifying the numbers of days of 24HDR required to reliably characterize food and nutrient intakes of individuals in a population. Such information directly influences the study design and statistical power available to detect differences between treatment and control groups in experimental studies, or its ability to validate the FFQ. Once the underlying pattern of dietary intake is estimated with confidence, the information can be used to identify specific sources of bias in structured questionnaires such as the FFQ.
The vast majority of published studies focused on evaluating the within- and between-individual variation of dietary intake have been conducted in North American populations (Beaton et al, 1979, 1983; Hunt et al, 1983; Sempos et al, 1985; Ziegler et al, 1987; Liu, 1989; Tarasuk & Beaton, 1991; Hebert et al, 1998). Only few studies outside of this region have been conducted to determine the sources of variance using 24HDR (Hebert et al, 2000) or dietary records/weighted dietary records (Ogama et al, 1999; Oh & Hong, 1999; Tokudome et al, 2002). In many of those studies variance components have been estimated only for macronutrients. There are no published estimates of within- and between-individual variances in consumption of macronutrients and micronutrients among Chinese people.
In 1997, we launched a large population-based cohort study, the Shanghai Women Health Study (SWHS), to investigate dietary and cancer risk among Chinese women. In-person interviews were conducted for 74 943 women. Simultaneously, a dietary calibration study was conducted among 200 SWHS participants in order to evaluate the validity and reliability of the SWHS FFQ. In the calibration study, biweekly 24HDR was implemented over a 1-y period.
The major objectives of this paper are to describe the major sources of dietary variation in the study population and to compare the ratios of within-individual to between-individual variances (w2/b2) for 26 nutrients obtained from the 24HDR with findings from other populations. The number of 24HDR required for future dietary calibration study among the study population was estimated and implications for epidemiologic studies are discussed.
Subjects and methods
Subjects and data collection
The SWHS is a population-based cohort study of 74 943 women aged between 40 and 70 y. The study participants were recruited from seven communities in urban Shanghai during 1997–2000. A dietary calibration study was conducted in 1998 involving 200 SWHS participants, who were randomly selected from the SWHS roster from 15 neighborhoods in three study communities. In all, 25 primary and 50 alternate contacts were identified for possible recruitment by each interviewer. Among participants in this study, approximately 30% were primary contacts. Study participants were contacted twice a month by 10 interviewers during a 12-month period to provide estimates of their food intake over the past 24 h. Thus, each woman has a total of 24 days of 24HDR. The days of 24HDR were chosen to assure a balanced representation of weekdays (18 days) and weekend days (6 days) for each participant. All recalls were obtained by an unannounced in-person interview in the evening after dinner. Subjects were asked the name and amount of foods that they consumed at each meal during the past 24 h (Shu et al, 2004).
The major objective of this statistical analysis was to estimate the nature and magnitude of within- and between-individual variance of nutrient intakes derived from 24HDR in the 1-y assessment period. Sources of dietary variance include: (1) between-individual (ie, between the 200 subjects); (2) season (ie, variation of dietary intake across the four seasons); (3) day of the week (ie, weekday and weekend day); (4) sequence (ie, order of the interview; 1–24); (6) interviewer; (7) age of the respondent; and (8) within-individual (ie, day-to-day variation unaccounted by other sources).
Daily nutrient intake data were derived from reported food intake using a Chinese food composition table (Wang & Shen, 1991). The nutrient data were first checked for skewness and kurtosis. In general, intakes of most macronutrients were nearly normally distributed. By contrast, intakes of most micronutrients were not normally distributed, necessitating a log transformation (see Table 2). Therefore, the variances ratios (VRs) reported in this paper are based on both untransformed and natural log transformed data.
The data were analyzed using a random effects model:
Where yijklm is the intake nutrient for the ith participant of the jth weekday or weekend during the kth season at the lth sequence and visited by mth interviewer; μ is the mean of a nutrient intake; subjecti is the random variable for variation among subjects; x1,x2, etc represent the separate random effects of day, month, sequence, and interviewer; ɛijklm is an error term, including the within-person variance. Estimates of within-individual variance (w2) and between-individual variance (b2) were calculated by setting mean squares (MS) equal to their expected values. Also the residual approach of energy adjustment is used for calculating VRs. To do this, the dependent variable yijklm is replaced by the residual, which is derived from the regression of the specific nutrient value on total energy intake (nutrient=a0+a1 energy+ɛ), which represents the difference in nutrient intake not being attributed to differences in total energy intake.
As a few study participants had fewer, and some had more, than 24 days of intake, the analysis was based on an unbalanced model. Variances were estimated using MIXED procedure using the Statistical Analysis System (SAS). The ratios of within-individual to between-individual variance components was estimated by
The correlation coefficient (ρ0) between a person's ‘true’ unobserved usual intake and the average of k (k=24 in our study) diet recalls can be estimated by the following formula:
Thus, for a given correlation coefficient for a person's ‘true’ unobserved usual intake with the average diet recalls (eg ρ0=0.7) and a known VR for a nutrient, the required repeat dietary recalls (k) for a dietary calibration study could be estimated.
In all, 98% of participants completed 24 days of 24HDRs. Four subjects had fewer than 24 24HDRs; that is, one each had 9, 17, 20, and 21 days of 24HDRs. A total of 32 persons had more than 24 records. These yield a total of 4 892 dietary records for the current analysis.
Table 1 shows the demographic characteristics of the study participants. The average age for the study participants is 55.3 y (s.d.=9.0) and 12.5% of them had completed college. About 80% had family income below 30 000 yuan per year (at the then-current exchange rate this was equivalent to ≈$3600/y), and approximately half worked as manual laborers.
Table 2 presents the summary statistics of nutrient intake in the study population. The macronutrient intake had a smaller variation than consumption of micronutrients. The macronutrient intake levels were close to normally distributed. Distributions of many, but not all, of the micronutrients were badly skewed, kurtotic or both. The variances of weekday and weekend (day), seasonal, interviewers; and sequential variance account for less than 5% of total variance (Table 3), indicating that these factors are not the main sources of variation in dietary intake in the study population. Ratios of within-person (residual) to between-person (subject) variance are generally greater than 1, with the exception of carbohydrate intake. For the vast majority of vitamin intakes, the within- to between-subject variance ratios for untransformed data (VR7 in Table 3) are above 2, with the highest ratio being noted for retinol (VR=22.37). The ratios for minerals range between 1.46 (phosphorus) and 11.56 (sodium).
In general, variance ratios for the log-transformed data are smaller than the untransformed data and such changes are most evident for micronutrients (VR8 in Table 3 and VR9 in Table 4). When the nutrients are adjusted for energy using the residual method, nearly all of VRs increase in magnitude (compare Table 4 to Table 3). For some of the nutrients, such as zinc, sodium, manganese, iron, thiamin, and niacin, the VRs are ⩾3 times higher than those without energy adjustment. The change in VR is smaller when log-transformed data were analyzed. The VRs for fat, riboflavin, total-vitamin A, δ-tocopherol, and calcium are decreased. Adjusting energy using the density method produced results similar to those derived using residuals (data not shown).
Table 5 shows the within- and between-person variance ratios (VR4 for nontransformed data; VR5 for log-transformed data) stratified by the age of study participants. The VRs in the younger age group (40–59 y old) generally are slightly larger than those of the older age group (60–70 y old), with the exception of fat, copper, and sodium. The differences appeared to be more evident for micronutrient intake. This is mainly attributed to a smaller number of food entries in 24HDR (529 vs 628 foods) among older (60–70 y) than younger (40–59) women. Body mass index, income, and education appeared to have a minimum influence on the variability of nutrient intake in this population (data not shown).
Formula (3) illustrates the effects of the variance ratio, the number of days of 24-h dietary recall on the attenuation of the correlation coefficient. In general, the smaller the VR and more the days of recalls, the less the attenuation of the correlation coefficient. For example, if VR is ⩽5, 10–15 days of 24HDR would result in a correlation greater than 0.7, while 20 days of 24HDR would result in a correlation of 0.8. According to the VRs presented in Table 3, 12 days of 24HDR would be adequate for a validation study to evaluate intakes of most nutrients in this Chinese population.
It is difficult to measure long-term or ‘habitual’ diet in free-living populations, a necessary step in most epidemiologic studies of diet and health. As dietary intakes of individuals tend to vary considerably from day-to-day, methods like the 24HDR are extremely vulnerable to these within-individual sources of variation unless an adequate number of replicate measurements is collected (Liu et al, 1978; Sempos et al, 1985; Liu, 1989). However, with a sufficient number of replicates, they have the lowest error of any assessment method (Hebert et al, 1998). While seldom used for primary dietary exposure information in analytic epidemiologic studies, short-term methods such as the 24HDR are used to assess fluctuations in dietary intake (ie, to ascertain the magnitude of within-individual error) and to provide comparative data for studies of the relative validity of some other method, such as the FFQ. Other reasons for collecting data using a method such as the 24HDR include comparing the intakes of groups and ranking individuals according to nutrient intake (which also is the primary concern of structured questionnaires such as the FFQ).
In studies of diet and health where the concern is focused primarily on ranking (eg, for estimating the relative risk using quantile categorizations) or establishing usual intake of individuals (eg, for establishing the true nature of the dose response), it is essential that the dietary exposures of individuals be estimated as accurately as possible. This requires some knowledge of the sources of variation in a study population including the relative contributions of within- and between-individual variability. Despite the potential utility of this information, very little work has been done on identifying contributions to total variability in dietary intake in human populations. This is the first such study conducted in China.
Design of this study is predicated on the fact that estimates from a random subsample may be used under the assumption that the within-individual variance tends to be similar across subjects in the same underlying population (William et al, 1983). The estimation of between- and within-individual variation is useful in determining the study sample size and the optimum number of replicate measurements required to study a diet–disease relationship or for conducting a dietary instrument validation study. Most North American studies only involve 6 days (Beaton et al, 1983) or 7 days (Mcgee et al, 1982) of diet recalls or records. In our study, we implemented 24 days of 24HDR. This is because the Chinese diet is still influenced by seasonal availability of certain foods, particularly vegetables. Therefore, it was anticipated that more replicates would be needed to capture the resulting increased interpersonal dietary variation.
Our findings suggest that among the SWHS participants the major sources of variation in dietary intake were attributable to between- and within-individual variation. Taken together, other sources contributed less than 5% to total variation, usually around 1–2%. For total energy and carbohydrate intake, the within-individual variance was approximately equal to the between-individual variance, a ratio similar to that which we observed in India (Hebert et al, 2000), but lower in general, than those observed in Western populations, Korea, and Japan (Beaton et al, 1979, 1983; Hunt et al, 1983; Sempos et al, 1985; Hebert et al, 1998; Ogama et al, 1999; Oh & Hong, 1999; Tokudome et al, 2002) (see Table 6). Beaton et al (1979) have suggested that within-person variation in nutrient intake is largely culturally determined. It is clear that both Chinese and Indian populations have less day-to-day variation in macronutrient intake, but day-to-day variation of macronutrient intake in Korea and Japan is a little higher than that in both China and India and is much close to that in Western Populations. This may be due to cultural practices regarding regulating total food intake. Also, the nature of the food supply in these four countries may play an important role in the variance ratio differences. Age and seasonal food appear to be the major attributors to the dietary intake variability in the study population. We did not find that income, education, and body mass index have a major impact on the variation.
For the micronutrients, particularly Vitamin E, Vitamin A, and sodium, the within-individual variation was larger than between-individual variation, and was larger than what we had observed in populations that we studied in India (Hebert et al, 2000 and see Table 6), but was nearly consistent with that reported in Korea, Japan, and Western population. It should be noted that the difference may be attributable, in part, to the fact that the Indian study was conducted in rural populations, and therefore they may have less food or nutrient heterogeneity than this urban Chinese population. Whatever the underlying reason, this difference has important implications for epidemiologic study design. For the micronutrients, repeated 24HDR is required to derive a reliable estimation of ‘true’ usual dietary intake. For study, consisting of 200 participants, we found that 10–15 days of dietary recall are sufficient for most micronutrients that have a VR smaller than or equal to 5. For the macronutrients, where the VR is ≈1.0, only a few days of 24HDR will be needed. As the day-to-day and seasonal variations tended to be small, timing is not of major concern. This will guarantee the correlation coefficients between ‘true’ usual daily dietary intake and estimated usual intake being 0.7 or more (Hartman et al, 1990; Mcavay & Rodin, 1998). If nutrients that have a very large VR are the major exposure of interest (eg vitamin A, carotene, δ-tocopherol, and sodium), more replicates will be needed to derive a reliable and valid estimation of ‘true’ intake.
SWHS participants had a higher VR for fiber intake than did the Nurses Health Study participants (Posner et al, 1992). We note that there is a tendency for VRs of most nutrients to decrease with the age of study participants, and this may explain the lower VRs observed in the study by Hunt et al (1983). Virtually without exception, the VRs for older (⩾60 y) women in our study were lower than those for younger women (⩽59 y). Clearly, this would facilitate an evaluation of dietary intake among the elderly.
In conclusion, the present investigation suggests that day-to-day fluctuation was the major source of variance in daily nutrient intake for this middle-aged and elderly female population in Shanghai China. In all, 10–15 days of 24HDR is needed if this methodology is to be used as the ‘gold’ standard for validating a dietary intake.
Beaton GH, Milner J, Corey P & McGuire V et al. (1979): Source of variance in 24-hour dietary recall data: implications for nutrition study design and interpretation. Am. J. Clin. Nutr. 32, 2546–2549.
Beaton GH, Milner J, McGuire V, Feather TE & Little JA (1983): Source of variance in 24-hour dietary recall data: implications for nutrition study design and interpretation. Carbohydrate sources, vitamins, and minerals. Am. J. Clin. Nutr. 37, 986–995.
Buzzard IM, Faucett CL, Jeffery RW, McBane L, McGovern P, Baxter JS, Shapiro AC, Blackburn GL, Chlebowski RT, Elashoff RM & Wynder EL (1996): Monitoring dietary change in a low-fat diet intervention study: advantages of using 24-hour dietary recalls vs food records. J. Am. Diet. Assoc. 96, 574–579.
Freudenheim JL & Marshall JR (1988): The problem of profound mismeasurement and the power of epidemiological studies of diet and cancer. Nutr. Cancer 11, 243–250.
Hartman AM, Brown CC, Palmgren J, Pietinen P, Verkasalo M, Myer D & Virtamo J (1990): Variability in nutrient and food intakes among older middle-aged men. Am. J. Epidemiol. 132, 999–1012.
Hebert JR, Clemow L, Pbert L, Ockene IS & Ockene JK (1995): Social desirability bias in dietary self-report may compromise the validity of dietary intake measures. Int. J. Epidemiol. 24, 389–398.
Hebert JR, Gupta PC, Mehta H, Ebbeling CB, Bhonsle RR & Varghese F (2000): Sources of variability in dietary intake in two distinct regions of rural India: implications for nutrition study design and interpretation. Eur. J. Clin. Nutr. 54, 479–486.
Hebert JR, Hurley TG, Chiraboga DE & Barone J (1998): A comparison of selected nutrient intakes derived from three diet assessment methods used in a low-fat maintenance trial. Public Health Nutr. 1, 207–214.
Hebert JR & Miller DR (1988): Methodologic considerations for investigating the diet–cancer link. Am. J. Clin. Nutr. 47, 1068–1077.
Hunt WC, Leonard AG, Garry PJ & Goodwin JS (1983): Components of variance in dietary data for an elderly population. Nutr. Res. 3, 433–444.
Hunter DJ (1999): Role of dietary fat in the causation of breast cancer: counterpoint. Cancer Epidemiol. Biomarkers Prev. 8, 9–13.
Kohlmeier L (1995): Future of dietary exposure assessment. Am. J. Clin. Nutr. 61 (Suppl), 702S–709S.
Liu K (1989): Consideration of and compensation of intra-individual variability in nutrient intakes. In Epidemiology, nutrition and health, eds Kohlmeier L & Helsing E London: Smith-Gordon; Niigate, Japan: Nishimura.
Liu K, Stamler J, Dyer A, McKeever J & Mckeever P (1978): Statistical methods to assess and minimize the role of intra-individual variability in obscuring the relationship between dietary lipids and serum cholesterol. J. Chron. Dis. 31, 399–418.
Mcavay G & Rodin J (1998): Interindividual and intraindividual variation in repeated measures of 24-hour dietary recall in the elderly. Appetite 11, 97–110.
Mcgee D, Rhoads G, Hankin J, Yano K & Tillotson J (1982): Within-person variability of nutrient intake in a group of Hawaiian men of Japanese ancestry. Am J. Clin. Nutr. 36, 657–663.
Ogawa K, Tsubono Y, Nishino Y, Watanabe Y, Ohkubo T, Watanabe T, Nakatsuka H, Takahashi N, Kawamura M, Tsuji I & Hisamichi S (1999): Inter- and Intra-individual variation of food and nutrient consumption in a rural Japanese population. Eur. J. Clin. Nutr. 53, 781–785.
Oh S-Y & Hong MH (1999): Within- and Between-person variation of nutrient intakes of older people in Korea. Eur. J. Clin. Nutr. 53, 625–629.
Posner BM, Martin-Munley SS, Smigelski C, Cupples LA, Cobb JL, Schaefer E, Miller DR & D'Agostino RB (1992): Comparison of techniques for estimating nutrient intake: The Framingham Study. Epidemiology 3, 171–177.
Sempos CT, Johnson NE, Smith EL & Gilligan C (1985): Effects of intraindividual and interindividual variation in repeated dietary records. Am. J. Epidemiol. 121, 120–130.
Shu XO, Yang G, Jin F, Liu DK, Gao YT & Zheng W (2004): Validity and reproducibility of the food frequency questionnaire used in the Shanghai Women's Health Study. Eur. J. Clin. Nutr. 58, 17–23.
Stamler J, Caggiula A, Grandits GA, Kjelsberg M & Cutler JA (1996): Relationship to blood pressure of combinations of dietary macronutrients. Findings of the multiple risk factor intervention trial (MRFIT). Circulation 94, 2417–2423.
Stram DO, Hankin JH, Wilkens LR, Pike MC, Monroe KR, Park S, Henderson BE, Nomura AM, Earle ME, Nagamine FS & Kolonel LN (2000): Calibration of the dietary questionnaire for a multiethnic cohort in Hawaii and Los Angeles. Am. J. Epidemiol. 151, 358–370.
Subar AF, Thompson FE, Kipnis V & Subar et al. (2001): respond to ‘A further look at dietary questionnaire validation’ and ‘Another perspective on food frequency questionnaires’. Am. J. Epidemiol. 154, 1105–1106.
Tarasuk V & Beaton GH. (1991): The nature and individuality of within-subject variation in energy intake. Am. J. Clin. Nutr. 54, 464–470.
Tokudome Y, Imaeda N, Nagaya T & Ikeda M et al. (2002): Daily, weekly, seasonal, within- and between-individual variation in nutrient intake according to four season consecutive 7 day weighed diet records in Japanese female dietitians. J. Epidemiol. 12, 85–92.
Wang GY & Shen ZP (eds) (1991): Chinese Food Composition Table. Beijing: People's Health Publishing House.
Willett W (1998): Nutritional Epidemiology 2nd edn. New York: Oxford University Press.
Willett WC (2001): Diet and cancer: One view at the start of the millennium. Cancer Epidemiol. Biomarkers Prev. 10, 3–8.
William HC, Andrea LG, Philip GJ & James GS (1983): Components of variance in dietary data for an elderly population. Nutr. Res. 3, 433–444.
Ziegler RG, Wilcox HB, Mason TJ, Bill JS & Virgo PW (1987): Seasonal variation in intake of carotenoids and vegetables and fruits among white men in New Jersey. Am. J. Clin. Nutr. 45, 107–114.
About this article
Cite this article
Cai, H., Shu, X., Hebert, J. et al. Variation in nutrient intakes among women in Shanghai, China. Eur J Clin Nutr 58, 1604–1611 (2004) doi:10.1038/sj.ejcn.1602013
- 24-hour dietary recall
- component of variance
- within-individual and between-individual variation
- dietary calibration study
Journal of Nutritional Science and Vitaminology (2019)
Long-Term Effectiveness of a Lifestyle Intervention: A Pragmatic Community Trial to Prevent Metabolic Syndrome
American Journal of Preventive Medicine (2019)
Nutrition Research and Practice (2018)
Protective Effect of Dietary Calcium Intake on Esophageal Cancer Risk: A Meta-Analysis of Observational Studies
Seasonality of sodium and potassium consumption in Switzerland. Data from three cross-sectional, population-based studies
Nutrition, Metabolism and Cardiovascular Diseases (2017)