QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Full Text (PDF) Purchase Article View Shopping Cart Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Bhargava, A. Search for Related Content PubMed PubMed Citation Articles by Bhargava, A.

---

Nutritional Epidemiology

Fiber Intakes and Anthropometric Measures are Predictors of Circulating Hormone, Triglyceride, and Cholesterol Concentrations in the Women's Health Trial¹

Department of Economics, University of Houston, Houston, TX 77204-5019

^* To whom correspondence should be addressed. E-mail: bhargava{at}uh.edu.

	ABSTRACT

TOP ABSTRACT Introduction Materials and Methods Results Discussion LITERATURE CITED

 
The unhealthy eating patterns and obesity among women in the U.S. are indicated by changes in biomarkers, such as insulin, lipoproteins, and estradiol, that are risk factors for breast cancer and cardiovascular diseases. This article models the inter-relations among diet, serum insulin, estradiol, and sex hormone binding globulin (SHBG) concentrations, plasma LDL and HDL cholesterol, and net triglyceride concentrations, using the data at baseline and 12 mo on 379 and 615 postm enopausal women in the Control and Intervention groups, respectively, of the Women's Health Trial: Feasibility Study in Minority Populations. Subjects in the Intervention group received detailed advice over a period of 1 y for reducing fat intakes and increasing the consumption of whole grains and fruits and vegetables. The main findings were that there were significant differences between the Control and Intervention groups in the changes from baseline to 12 mo in LDL and HDL cholesterol and SHBG concentrations. Second, using a comprehensive random effects modeling framework, the ratio of fiber to energy intake was significantly associated (P < 0.05) with lower insulin and triglyceride levels, and with a higher HDL cholesterol concentration in the Intervention group. Third, the subjects' waist-to-hip ratio and BMI were significantly associated with insulin, SHBG, LDL and HDL cholesterol, and triglyceride concentrations. Fourth, insulin levels were significantly negatively associated with SHBG and HDL cholesterol, and positively associated with LDL cholesterol, triglyceride, and estradiol concentrations. Overall, weight loss, especially around the waist, and increased fiber intakes are likely to be beneficial for lipid, cholesterol, and hormone profiles of U.S. women.

	Introduction

TOP ABSTRACT Introduction Materials and Methods Results Discussion LITERATURE CITED

Further, dietary intakes and sedentary lifestyles are important factors in explaining anthropometric measures, especially of women from low-education backgrounds (10); the evolution of lipid, lipoprotein, and hormone profiles depends on interactions between dietary, lifestyle, and biological factors. Moreover, central obesity reflected in the ratio of waist-to-hip circumference has been associated with higher estradiol concentrations (11); BMI and the waist-to-hip ratio are likely to have differential effects on biomarkers. Also, diets low in fat and high in fiber can affect subjects' triglyceride, cholesterol, and SHBG concentrations within a short time frame, although biomarkers such as estradiol may be less responsive.

In the previous literature on diet and biomarkers, rice intakes by Chinese women in a cross-sectional analysis were positively associated with SHBG concentrations (12). The fiber intakes by men in the U.S. were positively associated with SHBG, controlling for the insulin levels (13). Moreover, changes in fiber intakes over a 12 mo period were predictors of change in bioavailable estradiol among women with a history of breast cancer (14). However, previous analyses did not control for confounding factors such as subjects' anthropometric measures and insulin levels that reflect the history of dietary intakes. Moreover, the unobserved subject-specific characteristics partly reflect genetic factors and can complicate the interpretation of the estimated model parameters. Thus, it is desirable to analyze longitudinal data on dietary intakes and biomarkers on a large number of subjects with variation in diet composition.

The data on dietary intakes, anthropometric measures, lipids, lipoproteins, and hormone concentrations of postmenopausal women in the Control and Intervention groups of the Women's Health Trial: Feasibility Study in Minority Populations (WHTFSMP) at baseline, 6, and 12 mo provide an opportunity for a comprehensive analysis (15). Because the fat intakes were significantly reduced and consumption of whole grains, fruits, and vegetables were increased in the Intervention group (16), one can model the effects of dietary intakes on insulin, LDL and HDL cholesterol, triglyceride, estradiol, and SHBG concentrations, while controlling for subjects' ethnicity, lifestyles, and anthropometric measures. Furthermore, the inter-relations between the biomarkers are important for model specification. For example, subjects' insulin levels are likely to affect LDL and HDL cholesterol, triglyceride, and SHBG concentrations and should be included as explanatory variables in empirical models for these biomarkers. Similarly, because a high proportion of estradiol is bound by SHBG, one might expect subjects with higher SHBG concentrations to have lower estradiol concentrations. Moreover, insulin and SHBG concentrations are often positively correlated, and some researchers have argued that low SHBG concentrations are predictors of the "metabolic syndrome" (17,18). It is evident that an analysis of the WHTFSMP data using a comprehensive modeling framework that incorporate the various biological and dietary aspects can provide useful insights for improving women's health.

	Materials and Methods

TOP ABSTRACT Introduction Materials and Methods Results Discussion LITERATURE CITED

 
    Subjects. The WHTFSMP was a multicenter, randomized trial in 1991–95, sponsored by the National Cancer Institute, involving 2,208 women in Atlanta, Birmingham, and Miami, with the goals of reducing energy intakes from fat to 20% and increasing the consumption of fruits, vegetables, and grain products in the Intervention group (16). The participants (28% Black, 16% Hispanic, and 54% White) were postmenopausal women in the age group 50–79 y; 40% and 60% of the subjects were randomly assigned, respectively, to the Control and Intervention groups. The informed consent protocol was approved prior to the study (16), and secondary analyses were approved by the Human Subjects Committee of the University of Houston. The subjects were observed at baseline, 6, and 12 mo although, because of budget constraints, blood specimens collected at 6 mo were analyzed for only a randomly selected 40% of the subjects. Most subjects were observed for a period of 1 y, with subjects being recruited on a rolling basis. Thus, complete longitudinal data at baseline and 12 mo on dietary intakes, anthropometric variables and biomarkers were available for 379 women in the Control group and 615 women in the Intervention group.

    Diet, anthropometry, and socioeconomic and biological variables. The women in the Intervention group, led by a nutritionist, met weekly in groups of 8–15 for the first 6 wk, biweekly for the next 6 wk, and monthly thereafter for 9 mo and received detailed advice on reducing fat intakes. The subjects in the Control group were given pamphlets containing minimal information on healthy eating (19). The dietary intakes at baseline, 6, and 12 mo were measured in the 2 groups by a Food Frequency Questionnaire developed for this study (16,20).

The subjects' age, marital status, and reproductive history were recorded in the questionnaire. Education levels were coded into 4 categories that increased with the years of education. Subjects' height and weight were measured at baseline and at 12 mo. Height was measured with a stadiometer by rounding off to the nearest half-inch; weight was measured to the nearest pound, using a calibrated balanced beam scale. Waist circumference was measured at the smallest horizontal circumference between the ribs and the iliac crest, using a fiberglass tape. Hip circumference was measured at the maximum extension of the buttocks. The weight measurements were converted to kilograms and those for height, and waist, and hip circumferences were converted to meters. The subjects' patterns of "mild" and "strenuous" physical exercise for at least 30 min were investigated on a scale of 1–5 (1 = never and 5 = everyday). However, these variables were not significant predictors in the models and are not reported in this paper.

Blood specimens were drawn from the subjects in the Control and Intervention groups after a 12 h fast at baseline and 12 mo. With the subject seated, 20 mL of blood was drawn into one 10 mL plain tube and one 10 mL tube containing K3EDTA. The lipid analyses were performed using standardized protocols (CDC/ NHLBI Cholesterol Reference Method Laboratory Network). Plasma total cholesterol was determined by an enzymic assay in an automated analyzer. HDL cholesterol was determined by the same technique after precipitation of lipoproteins containing apolipoprotein-B with dextran sulfate (21). Net triglycerides were analyzed by a glycerol phosphate oxidase assay with glycerol used as the blank. LDL cholesterol was estimated using the Friedewald equation (22).

Serum estradiol concentrations were assayed with radioimmunoassay reagents (Diagnostic Products), using antibody-coated tube technology. The SHBG concentrations were assayed with immunoradiometric reagents (Diagnostic Systems Laboratories). Plasma glucose was measured with an automated analyzer, using a hexokinase method that was standardized to the National Institute of Standards and Technology. Serum insulin concentrations were determined by radioimmunoassay, using coated tubes. Lastly, the drugs affecting blood pressure, lipids, lipoproteins, hormones, and blood glucose were recorded in the dataset.

    The Model. We postulated the random effects model for insulin levels of n subjects at 2 time points (baseline and 12 mo) (i = 1,2,...,n; t = 1,2):

(1)

Here, ln represents natural logarithms and a₀, ..., a₈ are unknown coefficients; a₀ is the coefficient of the overall constant term. The model in Eq. 1 included indicator variables for ethnicity, insulin lowering medications, and for drugs affecting hormones. The subjects' age, BMI, waist-to-hip ratio, and the ratio of fiber to energy intake were transformed into natural logarithms (23), partly to reduce heteroscedasticity i.e., different variances of the errors for the subjects. Whereas fiber and energy intakes were introduced as separate variables in an alternative version of the model to account for the overall energy intake (24), fiber intake in Eq. 1 was expressed as the ratio to energy intake after the application of a statistical test (10). Coefficients of the variables in logarithms were the "elasticities" (percentage change in the dependent variable resulting from a 1% change in the independent variables). The variables Black and Hispanic were indicator (0–1) variables for the subjects' ethnicity. The models for estradiol, SHBG, triglycerides, and LDL and HDL cholesterol were similar to the model in Eq. 1 but also included subjects' insulin levels as explanatory variables; indicator variables for lipoprotein-lowering medications and for drugs affecting hormones were included in these models. The ratio of saturated fat to energy intakes was included in the models for triglycerides and cholesterol (25).

The u_{i t} were random error terms that can be decomposed in a random effects fashion as

(2)

where _i were subject-specific random variables that were assumed to be normally distributed with 0 mean and a constant variance, and v_it were normally distributed random variables with 0 mean and constant variance (26). The error terms capture the unobserved between-subject differences in the biomarkers levels that may be in part due to genetic differences.

    Statistical methods. For assessing the changes between the baseline and 12 mo periods in the Control and Intervention groups, paired t tests were used to test the null hypotheses that there were no differences among the means of dietary intakes, anthropometric measures, and biomarkers. The software package SPSS (SPSS for Windows version 10) was used to compute descriptive and t-statistics. Differences between the changes in the means in the Control and Intervention groups were tested using independent t tests and were considered significant at P < 0.05.

Because only 2 time observations were available on the subjects, the estimation of the model in Eq. 1 assumed that the number of women (n) was large but the number of time periods was fixed. The errors affecting the model were assumed independent across women, but correlated over time with a positive definite variance-covariance matrix. These assumptions were more general than the simple random effects model in Eq. 2. For example, the variances of the unobserved subject-specific random effects (_i) and the general errors (v_{i t}) could differ in the 2 time periods, and the v_{i t}'s could be serially correlated. These assumptions were better suited to the present application, where there were potentially large changes between the 2 time periods in the dependent and independent variables, due to the dietary intervention. The model parameters were estimated using a stepwise procedure developed for longitudinal data that used an estimated serial covariance matrix to produce efficient estimates (with the smallest possible variance) of the model parameters at the final stage (27). Constancy of model parameters across the Control and Intervention groups was tested using chi-square tests that were equivalent to likelihood ratio tests.

	Results

TOP ABSTRACT Introduction Materials and Methods Results Discussion LITERATURE CITED

 
    Descriptive statistics and t tests. Characteristics of the subjects in the 2 groups were similar at baseline (Table 1). There were significant changes in the Control group between baseline and 12 mo in body weight, intakes of saturated fat, fiber, and energy, and in LDL cholesterol and SHBG concentrations. In the Intervention group, changes in all variables except the waist-to-hip ratio and estradiol concentrations were significant. The differences in the changes between baseline and 12 mo between the Control and Intervention groups were significant for subjects' weight, hip and waist circumferences, intakes of saturated fat, fiber, and energy, and LDL and HDL cholesterol and SHBG concentrations. Thus, there was sufficient variation over time, especially in the Intervention group, in the dietary intakes, cholesterol, and SHBG concentrations to facilitate modeling of the pathways through which these changes were achieved.

The final set of columns in Table 2 present the results for the SHBG concentrations. The coefficients of the indicator variable for if the subjects were on medications affecting hormones were positive and significant in both the Control and Intervention groups. Moreover, the coefficients of the variable for subjects' age were positive and significant in both groups. The coefficients of BMI and waist-to-hip ratios were negative and significant for both groups, showing the importance of obesity and central obesity, respectively, for SHBG concentrations. The coefficients of the ratio of fiber to energy intakes were estimated with positive signs but were not significant. However, the insulin concentrations were significantly negatively associated with the SHBG concentrations in both groups.

    Empirical results from the models for LDL and HDL cholesterol and triglycerides for the 2 groups. The results from estimating the models for LDL and HDL cholesterol and triglyceride concentrations for the Control and Intervention groups are in Table 3. First, the subjects taking cholesterol lowering medications had significantly lower LDL concentrations in both groups. However, the coefficients of the indicator variable for

The results for HDL cholesterol showed that Black women in both groups had higher HDL concentrations. The medications for lowering LDL cholesterol had positive and significant effects on HDL cholesterol levels. The subjects' BMI was negatively and significantly associated with HDL cholesterol in the Control group, and the waist-to-hip ratio was significantly negatively associated with HDL cholesterol in both groups. Whereas the ratio of saturated fat to energy intakes was not significantly associated with HDL cholesterol, the ratio of fiber to energy intakes was positively associated in the Intervention group. In both the groups, subjects with higher insulin concentrations had significantly lower HDL cholesterol concentrations.

Finally, the results for net triglyceride concentrations showed that Black women had lower concentrations than White women in both groups. The coefficients of indicator variables for lipoprotein-lowering medications were estimated with positive signs that were significant for both groups. In the Control group, BMI was positively and significantly associated with triglyceride concentrations. The waist-to-hip ratio was estimated with large positive coefficients that were significant for both groups. The ratio of saturated fat to energy intakes was estimated with an unexpected negative result in the Intervention group. However, the ratio of fiber to energy intakes was negatively and significantly associated with subjects' triglyceride concentrations in the Intervention group. Lastly, the subjects' insulin concentrations were positively and significantly associated with triglyceride concentrations in both groups.

	Discussion

TOP ABSTRACT Introduction Materials and Methods Results Discussion LITERATURE CITED

 
This paper presented a comprehensive analysis of the effects of a dietary intervention in the WHTFSMP on the subjects' insulin, estradiol, SHBG, LDL and HDL cholesterol, and triglyceride concentrations, controlling for confounding factors such as body weight, waist-to-hip ratio, and background variables. Whereas there were significant differences between the Control and Intervention groups in the changes from baseline to 12 mo in LDL and HDL cholesterol and SHBG concentrations, the analysis of the data provided several useful insights. First, the ratio of fiber to energy intakes was negatively associated with insulin concentrations in the Intervention group, presumably due to greater variation in diet composition following the nutrition educational program. Moreover, the subjects' BMI and waist-to-hip ratio were positively and significantly associated with insulin concentrations in both the Control and Intervention groups.

Second, insulin concentrations were significantly negatively associated with SHBG and HDL cholesterol and positively associated with triglyceride concentrations in the Control and Intervention groups. Moreover, insulin concentrations were positively associated with estradiol and LDL cholesterol in the Intervention group. Because the chances of insulin resistance are higher among obese subjects, and the onset of insulin resistance is occurring at earlier ages in the U.S. due to obesity, lower body weight achieved through diet and exercise is likely to reduce the prevalence of insulin resistance (29). This, in turn, would be beneficial for the lipid, lipoprotein, and hormonal profiles of postmenopausal women.

Third, considering the effects of general obesity versus central obesity, the coefficients of BMI and waist-to-hip ratios were similar in the models for insulin and SHBG concentrations. However, in the models for LDL and HDL cholesterol, the magnitudes of coefficients of waist-to-hip ratios were generally larger than the corresponding coefficients of BMI. Moreover, in the model for triglyceride concentrations, the coefficients of waist-to-hip ratio were significant for Control and Intervention groups and the magnitude of the coefficients was at least 4 times higher than those of BMI. Excess fat around the waist among the subjects in WHTFSMP had deleterious consequences for the triglyceride and LDL cholesterol concentrations.

Fourth, the theoretical inter-relations between estradiol and SHBG have been emphasized in the literature (6,8). The bivariate correlations between estradiol and SHBG at baseline and 12 mo were 0.44 and 0.41, respectively, and were statistically significant. However, including SHBG in the empirical model for estradiol led to a small positive coefficient of SHBG that was not significant in the Control or the Intervention groups. Similarly, the coefficients of estradiol concentrations in the model for SHBG were not significant in the 2 groups. It is likely that the measurement of different components of estradiol, such as the proportions that are bound by SHBG and the bioavailable estradiol, would have been useful for investigating these inter-relations.

Fifth, the autocorrelations and within-subject variations (30) were estimated using the data on insulin, estradiol, SHBG, LDL and HDL cholesterol, and triglyceride concentrations at 3 time points (baseline, 6, and 12 mo) (results not shown). In contrast with other biomarkers, estradiol concentrations exhibited 3 times as large within-subject variation, and the between-subject variances were not statistically significant in the Control or the Intervention groups. Thus, it was not surprising that the model for estradiol (Table 2) identified medications affecting hormones as the most important explanatory variable.

Finally, researchers have found that low levels of SHBG predict higher chances of type 2 diabetes (18). The bivariate correlation between insulin and SHBG concentrations at baseline was –0.31, which was statistically significant. Although insulin concentrations were significant predictors of SHBG in both groups (Table 2), including SHBG as an explanatory variable in the model for insulin led to significant coefficients of SHBG, although with smaller magnitudes. The estimation techniques took into account possible bidirectionality in the relationships (10). Further, we estimated 2 sets of models where insulin levels at baseline were used to predict SHBG concentrations at 12 mo, and where baseline SHBG concentrations predicted insulin levels at 12 mo. However, the coefficients of previous insulin or SHBG concentrations were invariably significant for both groups, thereby throwing little additional light on the direction in the empirical relation between insulin and SHBG concentrations. Thus, it would be useful to analyze longitudinal data on anthropometric measures and insulin and SHBG concentrations spanning a longer time frame for understanding the evolution of insulin and SHBG concentrations. Overall, the results from our comprehensive analysis of the WHTFSMP data demonstrated the importance of reducing central obesity in particular and increasing the intakes of dietary fiber for improving the lipid, lipoprotein, and hormonal profiles of postmenopausal women.

	ACKNOWLEDGMENTS

 
While retaining the responsibility for the views, the author would like to thank Dr. Rudolf Kaaks for helpful comments.

	FOOTNOTES

 
¹ This research was initiated by a grant from the National Cancer Institute (R03 CA 97738).

Manuscript received 19 April 2006. Initial review completed 22 May 2006. Revision accepted 1 June 2006.

	LITERATURE CITED

TOP ABSTRACT Introduction Materials and Methods Results Discussion LITERATURE CITED

 

1. Stamler J, Daviglus ML, Garside DB, Dyer AR, Greenland P, Neaton JD. Relationship of baseline serum cholesterol levels in 3 large cohorts of younger men to long-term coronary, cardiovascular, and all-cause mortality and to longevity. JAMA. 2000;284:311–8.[Abstract/Free Full Text]

2. Nestle M, Jacobson MF. Halting the obesity epidemic: A public health policy approach. Public Health Rep. 2000;115:12–24.[Medline]

3. Turgeon J, McDonnell D, Martin K, Wise P. Hormone therapy: Physiological complexity belies therapeutic simplicity. Science. 2000;304:1269–73.

4. Cushman M, Kuller LH, Prentice R, Rodabough RJ, Psaty BM, Stafford RS, Sidney S, Rosendaal FR, Women's Health Initiative Investigators. Estrogen plus progestin and risk of venous thrombosis. JAMA. 2004;292:1573–80.[Abstract/Free Full Text]

5. Simonen P, Gylling H, Howard A, Miettinen T. Introducing a new component of the metabolic syndrome: low cholesterol absorption. Am J Clin Nutr. 2000;72:82–8.[Abstract/Free Full Text]

6. Kaaks R. Nutrition, hormones, and breast cancer: Is insulin the missing link? Cancer Causes Control. 1996;7:605–25.[Medline]

7. Kaaks R, Bellati C, Venturelli E, Rinaldi S, Secreto G, Biessy C, Pala V, Sieri S, Berrino F. Effects of dietary intervention on IGF-I and IGF-binding proteins, and related alterations in sex steroid metabolisms: The diet and adrogens (DIANA) randomized trial. Eur J Clin Nutr. 2003;57:1079–88.[Medline]

8. Tymchuk C, Tessler S, Barnard R. Changes in sex hormone-binding globulin, insulin and serum lipids in postmenopausal women on a low-fat, high fiber diet combined with exercise. Nutr Cancer. 2000;38:158–62.[Medline]

9. Plymate S, Matej L, Jones R, Friedl K. Inhibition of sex hormone binding globulin production in the human hepatoma (Hep G2) cell line by insulin and prolactin. J Clin Endocrinol Metab. 1988;67:460–4.[Abstract/Free Full Text]

10. Bhargava A, Guthrie J. Unhealthy eating habits, physical exercise and macronutrient intakes are predictors of anthropometric indicators in the Women's Health Trial: Feasibility Studies in Minority Populations. Br J Nutr. 2002;88:719–28.[Medline]

11. Barnett J, Woods M, Rosner B, McCormack C, Longcope C, Houser R, Gorbach S. Sex hormone levels in premenopausal African-American women with upper and lower body fat phenotypes. Nutr Cancer. 2001;41:47–56.[Medline]

12. Gates J, Parpia B, Campbell T, Junshi C. Associations of dietary factors and selected plasma variables with sex hormone-binding globulin in rural Chinese women. Am J Clin Nutr. 1996;63:22–31.[Abstract/Free Full Text]

13. Longcope C, Feldman H, McKinlay J, Araujo A. Diet and sex hormone-binding globulin. J Clin Endocrinol Metab. 2000;85:293–296.[Abstract/Free Full Text]

14. Rock C, Flatt S, Thomson C, Stefanick M, Newman V, Jones L, Natarajan L, Ritenbaugh C, Hollenbach K, et al. Effects of a high-fiber, low-fat diet intervention on serum concentrations of reproductive steroid hormones in women with a history of breast cancer. J Clin Oncol. 2004;22:2379–87.[Abstract/Free Full Text]

15. Hall W, Ziding F, George V, Lewis C, Oberman A, Huber M, Fouad M, Cutler J. for the Women's Health Trial: Feasibility Study in Minority Populations. Low-fat diet: Effect on anthropometrics, blood pressure, glucose, and insulin in older women. Ethn Dis. 2003;13:337–43.[Medline]

16. Bowen D, Clifford CC, Coates R, Evans M, Feng Z, Fouad M, George V, Gerace T, Grizzle JE, et al. The women's health trial feasibility study in minority populations: Design and baseline description. Ann Epidemiol. 1996;6:507–19.[Medline]

17. Oh J-Y, Barrett-Connor E, Wedick N, Wingard D. Endogenous sex hormones and the development of type 2 diabetes in older men and women: The Rancho Bernardo Study. Diabetes Care. 2002;25:55–60.[Abstract/Free Full Text]

18. Golden S, Ding J, Szklo M, Schmidt M, Duncan B, Dobs A. Glucose and insulin components of the metabolic syndrome are associated with hyperandrogenism in postmenopausal women. Am J Epidemiol. 2004;160:540–48.[Abstract/Free Full Text]

19. USDA/DHSSNutrition and your health: Dietary guidelines for Americans. Washington: Department of Health and Human Services, 1990.

20. Kristal AR, Feng Z, Coates RJ, Oberman A, George V. Associations of race/ethnicity, education, and dietary intervention with the validity and reliability of a food frequency questionnaire. Am J Epidemiol. 1997;146:856–69.[Abstract/Free Full Text]

21. Warnick G, Benderson J, Albers J. Dextran sulfate-Mg²⁺ precipitation procedure for quantitation of high-density-lipoprotein cholesterol. Clin Chem. 1982;28:1379–88.[Free Full Text]

22. Friedewald W, Levy R, Frederickson D. Estimation of the concentration of low-density lipoprotein cholesterol in plasma, without use of preparative ultracentrifuge. Clin Chem. 1972;18:499–502.[Abstract]

23. Nelson M, Black AE, Morris JA, Cole TJ. Between-and-within subject variation in nutrient intake from infancy to old age: Estimating the number of days to rank dietary intakes with desired precision. Am J Clin Nutr. 1989;50:155–67.[Abstract/Free Full Text]

24. Willett W. Nutritional Epidemiology, 2^nd edition. Oxford: Oxford University Press; 1998.

25. Keys A. Serum cholesterol response to dietary cholesterol. Am J Clin Nutr. 1984;40:351–9.[Abstract/Free Full Text]

26. Laird N, Ware J. Random effects models for longitudinal data. Biometrics. 1982;38:963–74.[Medline]

27. Bhargava A. Identification and panel data models with endogenous regressors. Rev Econ Stud. 1991;58:129–40.[Medline]

28. Liu K, Stamler J, Dyer A, McKeever J, McKeever P. Statistical methods to assess and minimize the role of intra-individual variability in obscuring the relationship between dietary lipids and serum cholesterol. J Chronic Dis. 1978;31:399–418.[Medline]

29. Lofgren I, Herron K, West K, Zern T, Brownbill R, Ilich J, Koo S, Fernandez M. Weight loss favorably modifies anthropometrics and reverses the metabolic syndrome in premenopausal women. J Am Coll Nutr. 2005;24:486–93.[Abstract/Free Full Text]

30. Bhargava A, Forthofer R, Mc Pherson S, Nichaman M. Estimating the variations and autocorrelations in dietary intakes on weekdays and weekends. Stat Med. 1994;13:113–26.[Medline]

This Article Abstract Full Text (PDF) Purchase Article View Shopping Cart Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Bhargava, A. Search for Related Content PubMed PubMed Citation Articles by Bhargava, A.