Jump to ContentJump to Main Navigation
Obesity Epidemiology$

Frank Hu

Print publication date: 2008

Print ISBN-13: 9780195312911

Published to Oxford Scholarship Online: September 2009

DOI: 10.1093/acprof:oso/9780195312911.001.0001

Show Summary Details
Page of

PRINTED FROM OXFORD SCHOLARSHIP ONLINE (oxford.universitypressscholarship.com). (c) Copyright Oxford University Press, 2020. All Rights Reserved. An individual user may print out a PDF of a single chapter of a monograph in OSO for personal use.  Subscriber: null; date: 30 October 2020

Dietary Assessment Methods

Dietary Assessment Methods

Chapter:
(p.84) 6 Dietary Assessment Methods
Source:
Obesity Epidemiology
Author(s):

Frank B. Hu

Publisher:
Oxford University Press
DOI:10.1093/acprof:oso/9780195312911.003.0006

Abstract and Keywords

This chapter begins by discussing the strengths and limitations of various dietary assessment methods—24-hour recall, food records, diet history, food-frequency questionnaires (FFQs), and biomarkers—and their validity and applications in obesity epidemiologic research. It then considers the assessment of, and adjustment for, total energy intake in epidemiologic research. It discusses methods to correct for random and systematic measurement errors in nutritional epidemiologic studies. Finally, the chapter reviews statistical approaches for assessing the impact of overall diet through dietary patterning analyses.

Keywords:   obesity research, obesity epidemiological studies, dietary composition, food-frequency questionnaires

For most people, significant weight gain results from a small but persistent positive energy balance over a long period of time, an balance that is too small to be detected by most instruments used in epidemiologic studies. This limitation notwithstanding, reasonably accurate methods are available to assess dietary composition and food patterns in free-living populations. Epidemiologic studies have demonstrated that many dietary factors have significant relationships with obesity and weight gain, either independently of, or mediated through, total energy intake (see Chapters 11 and 12). However, none of the dietary assessment approaches are perfect, and there are important methodological issues related to appropriate choices for specific research settings in obesity research.

In this chapter, we first discuss strengths and limitations of various dietary assessment methods—24-hour recall, food records, diet history, food-frequency questionnaires (FFQs), and biomarkers—and their validity and applications in obesity epidemiologic research. Next, we cover the assessment of, and adjustment for, total energy intake in epidemiologic research. We then discuss methods to correct for random and systematic measurement errors in nutritional epidemiologic studies. Finally, we review statistical approaches for assessing the impact of overall diet through dietary patterning analyses. The methods discussed in this chapter are not unique to obesity epidemiologic studies, but are relevant to virtually all areas of epidemiology involving dietary exposures.1

Dietary Assessment Methods in Obesity Epidemiologic Studies

Several methods are available to assess individual intakes of foods, nutrients, and total energy; these include single 24-hour recalls, FFQs, diet histories, food or diet records, and biomarkers. Each approach has its strengths and limitations. In the following section, we briefly discuss the application of these methods to assess dietary intake in epidemiologic research. For more detailed information, please refer to texts in nutritional epidemiology1 and dietary assessments.2

(p.85) 24-Hour Recalls

The 24-hour dietary recall involves the collection of detailed information on all foods and beverages consumed by a subject in the previous day or past 24 hours. This method is most widely used by national nutritional surveys (e.g., the National Health and Nutrition Examination Survey (NHANES), USDA Nationwide Food Consumption Surveys, and the Continuing Surveys of Food Intakes by Individuals) to estimate average intakes of populations. The recall method, especially the unannounced recall (see below), is frequently used in dietary intervention trials to monitor adherence. A single 24-hour dietary recall can be useful for estimating mean intakes of a population in national surveys, but it alone cannot be used to estimate usual intakes of individuals, or provide correct distributions for the population because of large within-person variations in dietary intakes.3

The 24-hour recall is usually conducted by a trained or certified interviewer. The interview is often face to face, but can also be done by telephone. In the face-to-face interview, visual aids, such as food models or shapes, can be used to obtain more accurate information on quantities of foods. To help with portion size estimation in telephone-administered interviews, two-dimensional food portion visual aids or photographs are sometimes mailed to the respondents’ homes beforehand.4 To avoid changes in participants’ eating habits, 24-hour recalls are best administered unannounced, that is, not scheduled on a specific day. The surprise aspect of the unannounced telephone interview is especially important for monitoring compliance in dietary intervention studies.5

Traditional paper-and-pencil or computerized systems can be used to collect data from 24-hour recalls. The Minnesota Nutrient Data System (NDS) has been specifically designed for conducting real-time interactive interviews6 and a computer program, EPIC-SOFT, has been developed to standardize interview-based 24-hour recalls in the European Prospective Investigation into Cancer and Nutrition (EPIC) study.7

The multiple-pass 24-hour recall is a method in which interviewers use several (3 or 5) distinct passes or steps (with multiple cues and opportunities for participants) to collect information about a subject’s food intake over the preceding 24 hours.8 The USDA 5-pass method involves five steps.9 , 10 The first pass is a quick list of all foods or beverages the participant consumed in the previous day. The second pass (termed forgotten list) involves a probe of possible forgotten foods, for example, snacks, sweets, and soft drinks. The third pass (time and occasion) asks the subject to describe the time and situation in which the foods were eaten. The fourth step (the detailed pass) involves probing for detailed information on preparation, ingredients, and portion sizes (use of two-dimensional food models can help subjects estimate portion sizes). The last step (the final review pass) involves reviewing the recalled information and probing for information on any additional food items. The multiple-pass 24-hour recall method has led to improvement in food recalls.9 , 11 However, the approach is still prone to underreporting that typically occurs with self-reported methods, a problem that involves memory lapses and difficulties in estimating portion sizes.

It is well known that a single 24-hour recall does not represent usual intake or reliably rank subjects according to nutrient intakes because of large day-to-day variations. Multiple recalls are required to estimate an individual’s usual diet, however, the optimal number of recalls depends on the nutrients or foods of interest; those with large day-to-day variations will require more recalls than those with smaller day-to-day variations. In large cohort studies, the cost of collecting and processing dietary data from multiple 24-hour recalls is often prohibitive. However, it is possible to do so in a subset of the cohort for validation purposes.

(p.86) Food or Dietary Records

With the food record or food diary method, the subject (or observer) records in detail all foods and beverages consumed on one or more days.3 In most studies, participants are asked to enter information on a standardized hard copy form, but new methods have been developed to collect tape-recorded or bar-coded food consumption information. Whereas weighed food records involve weighing all food and beverages consumed on a small scale, estimated food records require participants to estimate the portion sizes of all foods consumed using household measures or aids (e.g., food models or photographs). Participants should be trained in advance in methods for weighing and recording foods. Ideally, they should weigh and record each portion of food before eating it so that the food records do not depend on memory, although this is not feasible in many situations. In some situations (such as a population with low literacy rates), an observer or field worker is needed to weigh the raw ingredients and cooked dishes to estimate household and individual dietary intakes.3

Food records are typically collected for 3 to 7 days, but multiple 7-day records across different seasons are often required to reflect long-term diet. Multiple 7-day records are often used as the “gold standard” for validating other methods, such as the FFQ. Weighted diet records, if done correctly, have a clear advantage of not relying on a subject’s memory, and of allowing direct and accurate quantification of food intakes. Drawbacks include the need for literate, trained, and highly motivated subjects. Food records place a high burden on participants,8 and the quality of recording declines as the number of days increase. The process itself tends to modify eating habits, with some participants even losing weight. Due to day-to-day variations in diet, short-term (e.g., 3 days) records can misrepresent usual intake, a problem remedied by repeated recording over different time periods and seasons. Food records, as with other self-reported dietary assessment methods, are prone to underreporting bias (see below), especially among the obese.

Food records are most commonly used as a reference method for validating FFQs, however, the cost of collecting and processing the data in large cohort studies has been prohibitive. Nonetheless, some newer cohort studies (e.g., the European Prospective Investigation of Cancer (EPIC) in Norfolk) have simultaneously collected 7-day diet records and FFQ data.12 Diet records can be a cost-effective approach in nested case-control studies for assessment of exposure or for validation purposes. However, it is unclear whether single 7-day diet records can adequately reflect long-term dietary intake.

Food-Frequency Questionnaires

Short-term diet and recall methods fail to represent usual intake, are inappropriate for assessment of past diet, and are costly. Because of these limitations, alternative methods have been developed to measure long-term dietary intake. Among these methods, the FFQ has emerged as the preferred approach. FFQs are easy for participants to complete, can be processed by computer, and are inexpensive—features that make them feasible for use in large prospective studies.

The FFQ is based on two principles—that average long-term diet is conceptually more important than short-term diet, and that relative ranking of individual intakes is more important than absolute intakes in predicting chronic disease risk.1 (Absolute intakes are difficult, if not impossible, to measure precisely in large epidemiologic studies.) The foundation for this framework dates back to the dietary history interview developed in 1947 by Burke (discussed below).13 In the past several decades, the food-frequency method has become the main dietary assessment tool in large nutritional epidemiologic (p.87) studies, and numerous versions of FFQs have been developed for applications in various populations and contexts.

The FFQ asks respondents to report their usual frequency of consumption of each food from a list of foods during a specific period (typically from a few months to a year). The questionnaire consists of a structured listing of individual foods and beverages. For each food item, participants are asked to indicate their average frequency of consumption in terms of a specified serving size by checking one of multiple frequency categories (ranging, for example, from “almost never” to “six or more times a day”). The selected frequency category for each food item can be converted to a daily intake; for example, a response of “two to four per week” converts to 0.43 servings per day (or three times per week). The number or types of food items may vary by study purpose and population. Comprehensive FFQs used in most epidemiologic studies generally include 60 to 180 food items. The questionnaires can be administered by trained personnel in face-to-face interviews, by telephone, or through self-administered postal surveys. FFQs can be optically scanned, which improves the accuracy and efficiency of data entry and makes them suitable for use in large epidemiologic studies.

Collection of portion size information varies according to types of FFQs. In nonquantitative FFQs, portion size information is not collected. Such questionnaires, which cannot provide estimates of nutrient intakes, are typically used for screening purposes. In semiquantitative FFQs, portion sizes are specified as standardized portions or choices. For example, in the FFQ developed by Willett,14 portion size information is included as part of the food item rather than as a separate question. Other questionnaires (such as the Block FFQ) ask respondents to indicate usual portion sizes for each food (e.g., small, medium, or large) (using food models as a unit of reference).15 The NCI Diet History Questionnaire (DHQ) includes an additional question about portion size for each food.16 , 17

Whether adding portion size information to FFQs improves estimation of nutrients is still a matter of debate. In that most of the variation in food intakes is explained by frequency of intake rather than differences in portion sizes,1 available evidence suggests only marginal improvement in the validity of FFQs that include portion size data compared with those that do not. In a Danish study, the mean correlations between food-frequency data, with and without individually estimated portion sizes, and weighed diet records were similar, suggesting that little extra information was obtained by adding questions about portion size.18 Conversely, Subar et al. suggested that portion size information could improve estimates of absolute macronutrient intake,16 but not necessarily the validity of energy-adjusted nutrients.

Carefully developed FFQs offer a conceptual advantage over short-term 24-hour recalls or food records in assessing average long-term diet. They are also relatively inexpensive and impose a low burden on participants, factors that make them feasible for use in large epidemiologic studies. However, FFQs have significant limitations. Because they lack the detail and specificity of diet records or recalls they may not provide accurate estimates of absolute nutrient intakes. Constant changes in food supplies and compositions require that items in the FFQ and nutrient database be updated in a timely manner. In addition, the completion of FFQs involves memory, recall, and cognitive estimation skills. As a result, FFQs, as with other self-reported dietary assessment instruments, are subject to both random and systematic errors (see below).

FFQs need to be specific to individual cultures and populations. In some populations, low education levels may restrict the usefulness of self-administered surveys. These limitations need to be balanced against the strengths of FFQs. As the least expensive and most efficient dietary assessment tool, they have become the method of choice in large epidemiologic studies. However, questions on the validity of nutrient estimates by FFQs (p.88) in epidemiologic studies of obesity and chronic diseases continue to be raised. Later in this chapter, we will discuss the validity and reproducibility of FFQs.

Diet History Method

As discussed earlier, the diet history method was originally described by Burke in 1947.13 It consists of three parts: a detailed face-to-face interview; a cross-check food-frequency list; and a 3-day diet record. The food-frequency list and 3-day records were used by Burke to check the internal consistency of the interview. The purpose of the diet history method is to obtain usual food consumption patterns. The interview typically starts with a 24-hour recall prompted by careful probing of current and past food consumption patterns. Interviews, often lasting 1 to 2 hours, require substantial cooperation from the participants.

Several medium-sized prospective studies and clinical trials have used the diet history method. For example, in the Western Electric Study,19 nutritionists conducted an initial examination followed by a second interview 1 year later. Using standardized interviews and questionnaires, they collected data on usual eating patterns, special diets, and changes in eating habits. To gauge the internal consistency of the interview, they used a 195-item cross-check food-frequency list. Estimates of portion sizes were based on wax models of commonly consumed foods and dishes. Participants’ wives, using mailed questionnaires, provided further information on food preparation, as did neighborhood restaurants and bakeries.

The CARDIA diet history was modeled after the Western Electric dietary history method,20 but the list of foods was expanded from 150 items to approximately 700 to accommodate various populations and ethnic groups. Liu et al.20 reported on the reliability and validity of the CARDIA Diet History in 128 young adults. The reproducibility correlations for the log-transformed nutrient values and calorie-adjusted nutrient values from the two diet histories were generally in the range of 0.50 to 0.80 for Caucasians. For African Americans, the correlations were lower, with a majority in the range of 0.30 to 0.70. The validity correlations between mean daily nutrient intakes from the CARDIA diet history and means from 7 randomly scheduled 24-hour recalls were generally above 0.50.

The Diabetes Control and Complications Trial (DCCT) also used a modified Burketype diet history method. Trained dietitians interviewed participants for approximately 1.5 to 2 hours to collect quantitative and qualitative information on a usual week of dietary intake over the previous year. Schmidt et al.21 found that 1-year reproducibility correlation coefficients ranged from 0.51 for dietary fiber to 0.72 for dietary cholesterol.

Conceptually, the diet history method has advantages over 24-hour recalls and FFQs because it collects more accurate quantitative data on long-term consumption patterns. However, the method is time consuming, expensive, and hard to standardize. Thus, it is not often feasible for use in large epidemiologic studies involving tens of thousands of people. Similar to other dietary assessment methods described earlier, the diet history method is prone to recall bias caused by faulty memory or problems in estimating intake frequencies. It can also be affected by interviewer bias.

Biomarkers

In that the dietary assessment methods discussed earlier are imprecise and subject to bias, the use of biomarkers to assess nutrient intake has been of great interest to the nutritional epidemiology community. Biomarkers offer the advantages of increased reliability (p.89) and objectivity (they do not depend on memory). Although they are not immune to measurement errors, these are not correlated with measurement errors in self-reported dietary assessments. Still, useful biomarkers are not available for all nutrients, and there are no satisfactory and specific biomarkers for intakes of most foods and food groups. The most important requirement for a biomarker is sensitivity to intake, which typically means a dose-response relationship between the biomarker and nutrient intake.22 However, in many situations, there is a threshold effect at very low levels of intake and a plateau effect at very high levels. Another criterion for a useful biomarker is time integration. Because long-term dietary exposure is the main interest of chronic disease epidemiology, a valid biomarker should reflect the cumulative effect of diet over an extended period of time rather than short-term fluctuations. Thus, tissues with longer half-lives (e.g., adipose tissue, erythrocytes, and toenails) can be used to reflect dietary intake over the previous months or years. Below, we describe several key biomarkers that are commonly used in laboratory and field studies.

Doubly Labeled Water

The doubly labeled water (DLW) method is an objective and accurate measure of energy expenditure in free-living subjects.23 Its use as a measure of energy intake is based on the principle that energy expenditure should be equal to energy intake for individuals in energy balance. This method involves the oral administration of a carefully weighed dose of water containing enriched quantities of the stable isotopes deuterium (2H2O) and oxygen-18 (H2 18O) and collection of several urine or plasma samples over the next 15 days. The oxygen-18 is eliminated from the body in the form of both carbon dioxide (C18O2) and water (H2 18O), whereas the deuterium is eliminated in water (2H2O) only. Therefore, the difference in disappearance rate between these two isotopes from the body water pool is a measure of carbon dioxide production from which total energy expenditure can be calculated using standard equations for indirect calorimetry.24 The DLW method has been shown to have high accuracy and precision,25 but because the method is expensive and analysis requires specialized and sophisticated laboratory equipment, it cannot be used in large epidemiologic studies. However, the approach has been widely used in dietary validation studies of total energy intake (see below).

Total energy expenditure consists of three components: the resting metabolic rate (RMR), the thermic effect of food, and energy expended in physical activity.26 Thus, an alternate way to estimate total energy intake is to measure energy expended in physical activity and RMR, which together comprises approximately 90% of the total daily energy expenditure of sedentary persons. RMR can be measured by calorimetry and physical activity energy expenditure can be measured by accelerometers (see Chapter 7). Because the thermic effect of food accounts for approximately 10% of total energy expenditure, the estimated total energy expenditure can be calculated as (RMR + energy expended in physical activity) × 1.10.

Similar to the DLW method, indirect calorimetry is often infeasible in large epidemiologic studies. Several prediction equations have been developed to estimate RMR based on age, sex, height, and weight.27 , 28 These equations provide a rough estimate of minimal energy intake required for an individual’s survival. To estimate total energy expenditure, the RMR is multiplied by an activity factor according to different physical activity levels.29 , 30 The estimated energy expenditure can then be compared with reported energy expenditure by a dietary instrument, which can be used to identify underreporters. For example, a nominal factor of 1.35 (the ratio of estimated to reported total energy intake) has been used to estimate the lowest physiological limit for someone with minimal physical activity.29 , 30

(p.90) 24-Hour Urinary Nitrogen

Urinary nitrogen is commonly used as a biomarker for protein intake. Because most nitrogen intake (>80%) is excreted in the urine and 16% of protein is nitrogen, urinary nitrogen can provide an unbiased marker of protein intake.31 , 32 Bingham and Cummings31 investigated the value of 24-hour urine nitrogen (N) excretion as a way of validating dietary methods of measuring protein intake in four men and four women living in a metabolic ward. Daily N intake and excretion were measured for 28 days. The completeness of the 24-hour urine collections was verified by the use of PABA (p-aminobenzoic acid) taken by the patients with meals. The within-person coefficient of variation (CV%) in dietary intake of protein ranged from 14% to 26%, whereas CV% for urinary N varied from 11% to 18% within individuals. The correlation between 28 days urinary N and 28 days of protein intake was 0.99 with a CV% of 2% for urinary N; the correlation between 8 days urinary N and 28 days diet was 0.95 with a CV% of 5% for urinary N; the correlation between a single-day diet and urinary N was 0.47 with a CV% of 24% for urinary N. This study suggests that multiple-day (at least eight 24-hour collections) urinary nitrogen measurements are needed to provide stable estimates of nitrogen intake. Subsequent studies have found that partial 24-hour urine collection (even repeated overnight) cannot replace full 24-hour urine collection in measuring urea N.33 The collection of multiple 24-hour urine samples poses a major challenge for large epidemiologic studies.

24-Hour Urine Sodium and Potassium

In healthy individuals, blood levels of sodium and potassium do not reflect dietary intakes because of tight homeostatic control. Urine is the major route of excretion of these electrolytes, and thus 24-hour urinary sodium and potassium can serve as valid biomarkers of dietary intake of these nutrients.34 Large day-to-day variations of sodium and potassium intake make it necessary to conduct multiple 24-hour urine collections to obtain stable estimates of these electrolytes, and as with urine N, even repeat overnight collections cannot replace full 24-hour collections.35 Because 77% of dietary potassium is excreted in the urine,36 the (dietary potassium × 0.77)/urinary potassium ratio has been used to identify under- and overreporting of potassium intake assessed by other dietary assessment instruments.37

Total Fat and Dietary Fatty Acids

Dietary fat is probably the most commonly studied dietary exposure variable, yet there is no specific biomarker for total fat intake. This limits the ability to objectively evaluate the validity of total fat intake assessed by dietary instruments. However, it is well established in controlled metabolic studies38 that plasma fasting triglyceride levels are reduced with higher fat intake, and can thus serve as a nonspecific biomarker for fat intake. Willett et al.39 examined the relationship between total fat intake assessed by FFQs and plasma lipid levels among 185 women in the Nurses’ Health Study (NHS) and 269 men in the Health Professionals Follow-up Study (HPFS). In a multiple regression analysis adjusted for age, smoking, alcohol consumption, physical activity, body mass index (BMI), and intakes of protein, dietary fiber, and total energy, total fat intake was inversely associated with fasting triglycerides (a fat increase of 1% of energy, lowered triglyceride levels by 2.5% [95% CI: −3.7 to −1.3%, P = .0002]). For reported fat intake of 20% or less of energy, the geometric mean fasting triglyceride level was 179, and for more than 40% of energy it was 102 mg/dL (Fig. 6.1). This relationship was actually stronger than what (p.91)

                      Dietary Assessment Methods

Figure 6.1 Adjusted geometric mean fasting triglyceride (TG) levels by category of total fat intake among men in the Health Professionals’ Follow-Up Study (HPFS) (1994) and women in the Nurses’ Health Study (NHS) (1990). Values for men were adjusted for age in 1994, smoking, alcohol consumption, physical activity, BMI, total energy intake, and total protein intake. Values for women were adjusted for age; age at menarche; age at menopause; smoking status; BMI at age 18 years; intakes of total energy, fiber, protein, and alcohol; physical activity; history of breast cancer; history of benign breast disease; parity age at first birth; laboratory batch; and time of day of phlebotomy. Reproduced with permission from Willett W, Stampfer M, Chu NF, Spiegelman D, Holmes M, Rimm E. Assessment of questionnaire validity for measuring total fat intake using plasma lipid levels as criteria. Am J Epidemiol. 2001;154:1107–1112.39

would be predicted by the equations derived from metabolic studies,38 probably because the participants in the cohorts were middle-aged and older, and thus had greater body fat and insulin resistance than subjects typically enrolled in metabolic studies. These findings indicate that biomarkers sensitive to (but not necessarily specific for) the dietary factor being evaluated are of value for assessing the validity of dietary questionnaires or evaluation of compliance in intervention studies.

Concentrations of fatty acids in tissue can be used as biomarkers for the intake of different types of fatty acids. These tissues include plasma, erythrocytes, platelets, adipose tissue, various lipoprotein subfractions, and others.40 Useful fatty acid biomarkers are those that cannot be endogenously synthesized, such as polyunsaturated fatty acids (n-3 and n-6), trans fatty acids, and odd-numbered saturated fatty acids (e.g., 15:0, 17:0). Endogenously synthesized saturated (lauric, myristic, palmitic, and stearic acids) and monounsaturated fatty acids in tissue are not considered good biomarkers of intake. For saturated and monounsaturated fat intake, however, changes in nonspecific biomarkers (e.g., serum cholesterol and triglycerides) can be used as markers of intake. Controlled metabolic trials have shown that substituting saturated fat for carbohydrates significantly increases low-density lipoprotein (LDL) and high-density lipoprotein (HDL) cholesterol and decreases triglycerides, whereas substituting mono- or polyunsaturated fat for carbohydrates significantly decreases LDL cholesterol and triglycerides and increases HDL cholesterol.38 These effects can be predicted by well-established equations derived from metabolic studies.38 , 41 , 42

There has been considerable interest in using plasma levels of fatty acids as biomarkers of intake. Baylin and Campos40 summarized the quantitative effect of fatty acid substitution in the diet on changes in fatty acid composition in plasma cholesterol esters (p.92) (CEs), plasma triglycerides, and phospholipids from dietary metabolic trials. There was a clear dose-response relationship between increasing dietary linoleic acid intake and observed changes in serum CE or triglyceride linoleic acid concentrations. However, the dose-response relationship between serum phospholipids and dietary linoleic acid was much weaker, suggesting that more tightly regulated tissues (in particular, phospholipids of membranes) may not adequately reflect long-term intake.

Sun et al.43 compared fatty acid content of erythrocytes to that of plasma with respect to their abilities to reflect usual dietary fatty acid intake as measured by a FFQ. Docosahexaenoic acid (DHA, 22:6n-3) in plasma and erythrocytes provided the strongest correlations with its dietary intake, but erythrocytes (r = .56) were better than plasma (r = .48) as a biomarker. Similarly, total trans fatty acids (r = .43) and total 18:1 trans isomers (r = .42) in erythrocytes were more strongly correlated with dietary intake than plasma markers (r = .30 and r = .29, respectively). In addition, use of repeated measures of diet further improved these correlation coefficients.

Baylin et al.44 evaluated whole blood as a biomarker of intake. The diet-whole blood correlations were 0.43 for linoleic acid, 0.38 for alpha-linolenic acid, and 0.26 for 18:2 trans fatty acids. These results show that whole blood was a reasonable alternative for plasma for the assessment of fatty acid intake.

Adipose tissue is considered the best choice to assess long-term fatty acid intake because of its slow turnover rate. In a secondary prevention trial of coronary disease by substituting unsaturated fat for saturated fat, Dayton et al.45 observed that adipose tissue linoleic acid increased from 11% in the first year to 32% in year 5, suggesting excellent compliance with the intervention. In epidemiologic studies, intakes of linoleic and trans fatty acids estimated by FFQs are reasonably correlated with corresponding lipids in adipose tissue (with correlations in the range of 0.40 to 0.50).22 However, these correlations are only modestly higher than those for plasma markers.46

Adipose tissue levels of pentadecanoic acid (15:0) (PDA) and heptadecanoic acid (17:0) (HAD) can be used to reflect average long-term dairy fat consumption in free-living subjects. In a study of 81 healthy women aged 30 to 77 years in Sweden, Wolk et al.47 found a Pearson correlation coefficient of 0.63 between the 15:0 content in adipose tissue and intake from dairy foods from diet records, with a somewhat lower correlation for 17:0 (r = .42). Baylin et al.48 reported a correlation of 0.31 between adipose tissue content of 15:0 (also 17:0) and dairy product intake in Costa Rican men and women. Sun et al.49 reported correlation coefficients between 15:0 content and average dairy fat intake in 1986–1990 were 0.36 for plasma and 0.30 for erythrocytes. Trans 16:1n-7 in plasma (r = .30) and erythrocytes (r = .32) were also correlated with dairy fat intake.

Because of their high cost, laboratory measurements of fatty acids are most suitable for use in nested case-control or case-cohort studies, or as a reference method in validation studies. In addition, because they are usually expressed as a percentage of total fatty acids, they only reflect relative intake, with no measure of absolute fatty acid intake.40 Thus, changes in one fatty acid affect the distributions of the others.

Carbohydrates and Quality of Carbohydrates

As with total fat, there is no specific biomarker for total carbohydrate intake. However, plasma triglycerides rise in response to increasing carbohydrate intake (and decreasing fat intake),38 and can be used as a sensitive but nonspecific marker of carbohydrate intake. Because both the amount and quality of carbohydrates are important determinants of fasting plasma triacylglycerol concentrations, glycemic load has been used as a measure that incorporates the quantity as well as the quality of dietary carbohydrates consumed.50 (p.93) The glycemic load of a specific food—calculated as the product of that food’s carbohydrate content and its glycemic index value—has direct physiologic meaning in that each unit can be interpreted as the equivalent of 1 g of carbohydrate from white bread (or glucose depending on the reference used in determining the glycemic index). Liu et al.50 found a positive relationship between serum fasting triglycerides and total carbohydrate intake, the overall dietary glycemic index, and the dietary glycemic load among postmenopausal women, with the strongest relationship for glycemic load. The association between triglycerides and glycemic load appears to be stronger for overweight women than those who are not overweight, implying a biological interaction between underlying insulin resistance and carbohydrate metabolism.

Biomarkers of One-Carbon (Methyl) Metabolism

Folate, B12, B6, B2, choline, methionine, and betaine play key roles in one-carbon metabolism that involve transfer and utilization of one-carbon groups from one compound to another.51 , 52 One of folate’s main biological functions is remethylation of homocysteine to methionine. Folate status can be measured in serum (or plasma) and red cells. Serum folate reflects short-term folate status (within the past few days), while a concentration of red cell folate represents longer term and integrated folate intake because the half-life of red cells is approximately 4 months.53 Thus, red cell folate more closely reflects tissue folate status. Vitamin B12 status is measured by the serum cobalamin assay, and vitamin B6 status is indicated by the circulating concentration of pyridoxial-5′-phosphate (PLP). Elevated serum or plasma homocysteine is a sensitive and nonspecific marker of both folate and vitamin B12 deficiency, but it is not a reliable indicator of vitamin B6 status.53 In a recent study, Cho et al.54 demonstrated a significant association between dietary intake of choline plus betaine (assessed by FFQ in the Framingham Offspring Study) and lower homocysteine concentrations. Main sources of choline in the diet included red meat, poultry, milk, eggs, and fish, while main sources of betaine included spinach, pasta, white bread, cold breakfast cereal, and English muffins, bagels, or rolls.

Dietary folate assessed by FFQ is well correlated with biomarkers of folate status, including serum or plasma folate: r = .56 among 385 participants in the Framingham Heart Study,55 and r = .63 among 139 Boston-area participants.56 For red cell folate, the r was .42.57 The measurement of genomic DNA methylation in blood mononuclear cells may also serve as a useful biomarker for dietary folate intake.53

Biomarkers of Isoflavones and Lignans

Isoflavones and lignans are naturally occurring plant-derived phytoestrogens that may have biologically active properties.58 Lignans are present in grains, beans, green vegetables, fruits, nuts, and grasses, whereas isoflavones are concentrated in soybeans and soy foods. Common dietary isoflavonoids and metabolites include genistein, daidzein, dihydrodaidzein, O-desmethylangolensin, and equol; common lignans and metabolites include enterolactone, enterodiol, matairesinol, and secoisolariciresinol. In a typical Western diet, the daily intake of phytoestrogens is very low (<1 mg/day).59

Biochemical indicators of isoflavones and lignans can be measured in urine and blood specimens. However, these measurements often reflect only short-term intake, that is, several to 24 hours before a blood draw.22 , 60 , 61 Isoflavone excretion is substantial in the urine of Asian populations, which have high soy intake.61 Populations that consume a typical Western diet have very low blood concentrations or urinary excretion of isoflavone, and large within-person variations.62

(p.94) Urinary or plasma isoflavone and lignan concentrations can be used as measures of adherence for dietary intervention trials of soy or isoflavone supplementation.62 Cross-sectional studies have demonstrated good correlations between dietary intakes of soy and urinary concentrations of isoflavone in several Asian populations.63 , 64 Because these biomarkers reflect short-term intake, their usefulness in predicting long-term disease risk, especially in populations consuming a Western diet, is unclear.

Biomarkers of Trace Minerals

Hambidge65 provided a comprehensive review of biomarkers of trace mineral intake and status. Although there are no reliable biomarkers of iron intake, plasma ferritin is considered the best marker of body iron stores in the absence of acute inflammation. A higher ferritin concentration has been associated with increased consumption of heme iron and iron supplementation assessed by a FFQ.66 However, serum ferritin is also influenced by nondietary determinants, such as age, postmenopausal hormone use, obesity, physical activity, aspirin use, gastrointestinal ulcer, and genetic polymorphisms. Thus, ferritin is a nonspecific marker of iron stores. Plasma soluble transferrin receptor concentration (sTfR) is considered a sensitive and specific marker of early iron deficiency.

Selenium can be measured in plasma, red cells, and toenails. Selenium intake calculated from duplicate meals correlated well with serum (r = .63), whole blood (r = .62), and toenail (r = .59) selenium concentrations.67 Plasma selenium levels appear to be sensitive to short-term changes in dietary intake of selenium, whereas erythrocyte selenium can reflect relatively long-term exposure (e.g., several months). Toenail selenium levels are considered the best time-integrated biomarker of long-term selenium intake because toenails have a slow turnover rate. Hunter et al.68 showed a dose-response relationship between selenium supplementation and toenail selenium levels in free-living women. Longnecker et al.69 conducted an intervention study in which 12 males were fed high-dose (4.91 μmol Se/d), medium-dose (2.61 μmol Se/d), or control (0.41 μmol Se/d) whole wheat bread for 1 year, with the concentration of selenium measured in toenail clippings collected every 12 weeks for 2 years. Toenail selenium concentration was unaffected by dietary intake in the first 3 months and appeared to provide a time-integrated measure of intake over a period of 26–52 weeks. Thus, toenail concentration of selenium is a useful marker of long-term average intake. This is important because highly variable selenium concentrations in different samples of the same food make it difficult to calculate dietary selenium intake accurately. Other trace elements (e.g., chromium, magnesium, zinc, and copper) can also be measured in toenails using the same procedure as that used to measure selenium (instrumental neutron-activation analysis).70 Whether these biomarkers reflect long-term dietary intakes needs to be studied further.

Use of Biomarkers in Obesity Epidemiologic Studies

Nutritional biomarkers are used in several important ways in obesity epidemiologic studies. Because measurement errors of biomarkers are essentially uncorrelated with errors in any dietary assessment methods, they can be used as a reference method for validating self-reported dietary instruments. DLW, considered the gold standard for measuring energy intake (in the absence of weight change), is now widely used in validation studies of total energy intake measured by self-reported methods (see below). Similarly, urinary nitrogen has been used as a reference marker of dietary protein intake. Sensitive but nonspecific biomarkers, such as HDL cholesterol and triglycerides, can be used to validate long-term intakes of dietary fat and carbohydrates. These biomarkers are also useful in monitoring dietary compliance in weight loss trials.

(p.95) Some biomarkers can serve as surrogate indicators of long-term dietary intake in studies on dietary predictors of obesity. For example, essential fatty acid composition in adipose tissue can be used to predict long-term weight gain and obesity risk. In nutritional epidemiologic studies, the relationship between biomarkers of nutritional intake and status and risk of chronic disease incidence and mortality are typically investigated in a nested case-control design. The same design can be used to study the associations between nutrient biomarkers and onset of obesity. Hunter22 provided a comprehensive list of biochemical markers that are used in epidemiologic studies along with representative values of their reproducibility and validity.

Sufficiently valid biomarkers for intakes of food and food groups can also be useful in dietary intervention trials or in studies of dietary determinants of obesity and obesity- related chronic diseases. For example, plasma carotenoids are known to be useful biomarkers of vegetable and fruit intake.71 As discussed earlier, urinary or plasma concentrations of isoflavones can reflect soy-rich diets and adipose tissue levels of 15:0 and 17:0 are reasonable markers of dairy fat. However, biomarkers cannot substitute for self-reported dietary assessment methods for several reasons. First, not all nutrients have sensitive biomarkers that are practical to measure, and most foods and food groups have no useful biomarkers. Second, many biomarkers reflect short-term intake rather than usual diet. This limits their usefulness in validation studies of usual diet, and in investigations of the relationship between diet and long-term risk of obesity and obesity-related conditions.

Biochemical indicators are not influenced by dietary intake alone because individuals generally differ to some degree in the absorption and metabolism of most nutrients. Other sources of physiologic or genetic variations, such as levels of binding proteins or diurnal or menstrual cycles, may also influence the biochemical levels of nutrients and their metabolites.22 Finally, measurements of biomarkers are prone to many sources of laboratory errors. Thus, careful attention to specimen collection, storage, and assays, and sound epidemiologic design are critical in studies involving biomarkers.

Validation of Dietary Assessment Methods

Validation studies are designed to evaluate the reproducibility and validity of dietary measurements against one or more “reference methods.” Reproducibility refers to “consistency of questionnaire measurements on more than one administration to the same person at different times”; validity refers to “the degree to which the questionnaire actually measures the aspect of diet that it was designed to measure.”14 A precise reproducible instrument shows good agreement in repeated administrations, while a valid instrument is accurate in measuring unbiased true intake (typically the usual diet over a period of time). Ideally, an instrument should be precise as well as accurate. However, random and systematic errors can lead to inaccuracy and imprecision (Fig. 6.2).72

In validation studies, the choice of the reference method is a critical issue. One dilemma facing nutritional epidemiologists is the lack of a true gold standard against which to assess habitual dietary intakes. Nelson73 discussed limitations of reference methods that are commonly used in validation studies of test instruments (Table 6.1). For example, repeat 24-hour recalls and multiple diet records can provide a quantitative assessment of food consumption, but the number of recalls or records required to reflect usual diet is large (depending on the magnitude of the intraindividual variations of the nutrients). Underreporting of energy intake is common for both instruments. Objective methods (e.g., DLW, urinary nitrogen excretion, adipose tissue fatty acid composition, and (p.96)

                      Dietary Assessment Methods

Figure 6.2 Visual representation of accuracy (validity) and precision (repeatability). True average (solid lines), repeat measurements (filled circle) and measured average (dashed lines). Reproduced with permission from Livingstone MB, Black AE. Markers of the validity of reported energy intake. J Nutr. 2003;133(Suppl 3):895S–920S.72

toenail concentrations of trace minerals) have the advantage of measurement errors that are independent of test instrument errors. However, these methods can be used to validate only one or a few nutrients at a time, and some methods, such as DLW and urinary nitrogen excretion, may not be sufficient to reflect long-term energy and protein intake unless multiple assessments are done over a prolonged period of time.

Because of drawbacks associated with reference methods, validation studies of nutrient intake often rely on alloyed standards. Diet records are the standard that is most commonly used to evaluate test methods, especially FFQs. A major advantage of diet records is that they do not depend on memory; similarly, weighed records do not depend on perceptions of portion size or amount of foods consumed. Diet records should be kept for a sufficient number of days over a prolonged period of time (e.g., 1 year) to represent long-term average intake. Multiple 24-hour recalls are also a popular choice as a reference method. Unlike diet records, repeat 24-hour recalls do not alter participants’ regular eating habits. In the past two decades, numerous validation studies have been conducted to examine the validity of FFQs developed for different populations and cultures. Since Willett’s summary of validation studies published through 1997,1 the literature on the validity of FFQs has continued to grow.

The validity of FFQs can vary considerably across different populations or cultures. However, average correlations between nutrients assessed by FFQs and reference methods (e.g., one- or multiple-week diet records or repeat 24-hour recalls), when adjusted for total energy intake, are generally in the range of 0.4 to 0.7.14 For example, in a validation study of the 136-item Willett questionnaire in the NHS, (p.97)

Table 6.1 Limitations of Reference Methods Appropriate for Validation of Dietary Assessment Measures

Reference Method

Limitations

Doubly labeled water

Energy only

Assumptions of model regarding water partitioning may not apply in cases of gross obesity or high alcohol intake

Very expensive

Urinary nitrogen (completeness of samples confirmed using PABA)

Protein only

Requires multiple 24-h urine collections

PABA analysis affected by paracetamol and related products

Urinary nitrogen only

Protein only

Danger of incomplete samples

Weighed records or household measures

Underreporting

Not representative of usual diet due to insufficient number of days

Distortion of food habits due to recording process

Diet history

Interviewer bias

Inaccuracy of portion size reporting due to conceptualization and memory errors

Errors in reporting of frequency, especially overreporting of related foods listed separately (e.g., individual fruits and vegetables)

Requires regular eating habits

Repeat 24-h recalls

Under- or overreporting of foods due to reporting process (e.g., alcohol and fruit)

Not representative of usual diet due to insufficient number of days

Inaccuracy of portion size reporting due to conceptualization and memory errors

Biochemical measurements of nutrients in blood or other tissues

Complex relationship with intake mediated by digestion, absorption, uptake, utilization, metabolism, excretion, and homeostatic mechanisms

Cost and precision of assays

Invasive

Sensitive biomarkers do not exist for many nutrients

Adapted from Nelson M. The validation of dietary assessment. In: Margetts BM, Nelson MC, eds. Design Concepts in Nutritional Epidemiology. 2nd ed. New York: Oxford University Press;1997;241–272;Chapter 8.73

the mean deattenuated correlation coefficient for nutrient intakes (corrected for within-person variation) between FFQs and diet records was 0.62. A similar FFQ was also evaluated in a sample of 127 male participants in the HPFS.74 During a 1-year interval, men completed two 1-week diet records spaced approximately 6 months apart. Intraclass correlation coefficients for nutrient intakes assessed by questionnaires 1 year apart ranged from 0.47 for vitamin E without supplements to 0.80 for vitamin C with supplements. Correlation coefficients between the energy-adjusted nutrient intakes measured by diet records and the second questionnaire (which asked about diet during the year encompassing the diet records) ranged from 0.28 for iron without supplements to 0.86 for vitamin C with supplements (mean r = .59). These correlations were higher after adjusting for week-to-week variation in diet record intakes (mean r = .65). Food-based analyses reported an average (p.98) correlation coefficient of >0.60 comparing the FFQ and dietary record for foods in dietary questionnaires after correcting for within-person variation in both men75 and women.76 These data indicate that the FFQ provides a useful measure of intake for many nutrients and foods over a 1-year period.

In a validation study of FFQs against repeated 24-hour recalls in the European Prospective Investigation into Cancer and Nutrition (EPIC), Kroke et al.77 also obtained energy-adjusted correlation coefficents ranging from 0.54 for dietary fiber to 0.86 for alcohol. Similarly, in a validation study of the FFQ used in the Shanghai Women’s Health Study,78 nutrient and food intake assessed by the FFQ and the multiple 24-hour dietary recall were reasonably correlated, with the energy-adjusted correlation coefficients ranging from 0.59 to 0.66 for macronutrients, from 0.41 to 0.59 for micronutrients, and from 0.41 to 0.66 for major food groups.

Several studies have compared different versions of FFQs commonly used in nutritional epidemiologic studies. Caan et al.79 compared the performance of the Block and Willett FFQs with a longer, interviewer-administered diet history in two separate subsamples of participants. Although both questionnaires generally provided lower absolute intake estimates than the diet history, the ability to rank or classify individuals was very similar, and comparable to that of the diet history. In a comparison of the Block and Willett FFQs against multiple 24-hour recalls, Wirfalt et al.80 found different performance characteristics for the two FFQs with respect to categorizing of individuals according to different nutrients. Subar and colleagues conducted a detailed study to compare the validity of the Block, Willett, and NCI DHQs against four 24-hour recalls completed over a 1-year period.16 They found that the crude validity correlation coefficients tended to be lower for the Willett FFQ, but the energy-adjusted correlation coefficients were similar across the different questionnaires.

Numerous validation studies have employed biomarkers as the reference method. As discussed earlier, sensitive and specific biomarkers that represent long-term dietary intake are only available for a limited number of nutrients, and no specific biomarkers exist for most nutrients and foods. However, sensitive though nonspecific biomarkers can still be useful in evaluating the validity of nutrient measures. For example, because plasma triglycerides are responsive to an increase in carbohydrates, they can serve as an indirect biomarker for changes in fat and carbohydrate composition in the diet. Plasma HDL cholesterol level, another nonspecific biomarker, is not only responsive to changes in dietary fat and carbohydrates, but is also influenced by habitual alcohol consumption.81 For this reason, HDL cholesterol has been used as a biomarker to validate long-term alcohol consumption assessed by FFQs or other dietary assessment instruments.82 The urinary excretion of 5-hydroxytryptophol (5-HTOL):5-hydroxyindole-3-acetic acid (5-HIAA) ratio has been shown to be a sensitive marker of recent alcohol intake based on a single 24-hour recall.83

For most nutrients, correlations with biomarkers are in the range of 0.3 to 0.5. These moderate correlations result from imperfections in dietary assessment instruments as well as technical errors in measurement of biomarkers. In many situations, they are also due to homeostatic control of biomarker metabolism and nondietary determinants of biomarkers. Nonetheless, these correlations provide useful and objective evidence for the validity of nutrient intake assessed by a dietary instrument.

Method of Triads

Biomarkers and a reference dietary method (e.g., repeat 24-hour recalls or food records) provide complementary information on the validity of a FFQ. These measures can also be used to estimate the correlation between FFQs and the true measure of long-term (p.99) diet by using the method of triads.84 Kabagambe et al.85 employed this method to assess the validity of a FFQ used in a Hispanic population. Seven 24-hour dietary recalls and two FFQ interviews (12 months apart) were conducted to estimate dietary intake during the past year. Plasma and adipose tissue samples were collected from all subjects. The validity coefficient (VC) was estimated from three pair wise correlations between the FFQ, the reference method, and the biomarker (Fig. 6.3). Suppose Q, R, and M are the measurements from the FFQ, the reference method, and the biomarker, respectively, VCs for the reference method and the biomarker can be estimated as follows:

                      Dietary Assessment Methods
and
                      Dietary Assessment Methods
where r is the correlation corrected for within-subject variation and T is a latent variable representing the true but unknown long-term dietary intake.

Kabagambe et al.85 found that the median VCs for tocopherols and carotenoids estimated by repeat 24-hour recalls, the average of the two FFQs, and plasma were 0.71, 0.60, and 0.52, respectively. Plasma measures were better biomarkers for carotenoids and tocopherols than adipose tissue. However, adipose tissue appeared to be a better biomarker for polyunsaturated fatty acids (VC, 0.45 to 1.00) and lycopene (VC, 0.51). In general, the biomarkers did not perform better than the FFQs, and thus, the authors concluded that biomarkers should be used to complement the FFQ rather than substitute for it.

The method of triads has also been used in several other studies.86 88 Though useful, it requires the assumption that the errors between the methods are independent. This method has an important drawback. In some situations, VCs are inestimable, or >1, a condition referred to as Heywood case.84

                      Dietary Assessment Methods

Figure 6.3 Diagrammatic representation of the method of triads used to estimate the correlation between true long-term nutrient intake and intake estimated using dietary assessment methods. Reproduced with permission from Kabagambe EK, Baylin A, Allan DA, Siles X, Spiegelman D, Campos H. Application of the method of triads to evaluate the performance of food-frequency questionnaires and biomarkers as indicators of long-term dietary intake. Am J Epidemiol. 2001;154:1126–1135.85

(p.100) Improving the Validity of FFQs through Repeated Measures

As discussed earlier, in most validation studies the correlations between nutrients assessed by FFQs and diet records or repeat 24-hour recalls range from 0.4 to 0.7. This ceiling effect results not only from the inability of FFQs to capture the complexity of diets fully but also from lack of a true gold standard for comparison.89 Use of repeated measures from FFQs in prospective studies can potentially improve the validity of dietary measures by reducing random errors. Repeated measures are also more likely to represent long-term dietary intake.

As part of the evaluation of the expanded FFQ used in the NHS in 1986, we included participants from a similar study done 6 years earlier (in 1980). The series of diet records from the same women after a 6-year interval allowed us to assess the ability of repeated measurement of intakes by FFQs to represent long-term diet. In the validation study of the original FFQ in 1980, participants were asked to complete four 1-week diet records over a 1-year period. In the validation study of the expanded FFQ, participants were asked to complete two 7-day diet records approximately 6 months apart. To capture seasonal changes in diet, the first set of records was completed in the winter or spring, and the second set in the summer or fall of 1987 (Fig. 6.4).

To assess the ability of FFQs to reflect long-term diet, we compared the averages of nutrient intakes from the 1980 and 1986 diet record means with nutrient intakes assessed by FFQs in 1980, 1984, and 1986, separately, and the average intakes from the 1980, 1984, and 1986 FFQs. In both the 1980 and 1986 validation studies, FFQs were administered twice, before and after completing the diet records. We used the second FFQ for this analysis because it covered the time period in which the diet records were completed. Rather than using only one set of diet records, we used two sets (i.e., 1980 and 1986) as the comparison method. To compensate for attenuation of correlation coefficients, we used the within- and between-person components of variation in diet records (treating the two sets of diet records as two random units of observation) to deattenuate correlation coefficients for intakes of macronutrients (see below). This approach provided an estimate of the correlation that would have been observed had we collected diet records for each year during the 6-year period.

Pearson correlations between nutrient intakes from the 1980 and 1986 diet records ranged from 0.42 for saturated fat to 0.74 for carbohydrates (mean = 0.55). For the FFQs, the mean reproducibility coefficients were 0.37 between 1980 and 1984, 0.53 between 1984 and 1986, and 0.34 between 1980 and 1986. When the averages of nutrients from the 1980 and 1986 diet records were compared with the questionnaires (Table 6.2), the mean correlations for the above macronutrients (after correction for within-person variability in diet records) were 0.57 for the 1980 questionnaire, 0.65

                      Dietary Assessment Methods

Figure 6.4 An outline of the time frame for the 1980 and 1986 FFQ validation studies in the Nurses’ Health Study.

(p.101)

Table 6.2 Pearson Correlation Coefficients (Deattenuated) for Energy-Adjusted Macronutrient Intakes Assessed by FFQs and the Average Intakes Assessed by 1980 and 1986 Diet Records

1980 FFQ vs. Average Diet Records

1984 FFQ vs. Average Diet Records

1986 FFQ vs. Average Diet Records

Average of 1980, 1984, 1986 FFQs vs. Average Diet Records

Total fat

0.44

0.47

0.62

0.64

(0.57)

(0.61)

(0.81)

(0.83)

Saturated fat

0.50

(0.70)

0.49

(0.68)

0.64

(0.90)

0.68

(0.95)

Cholesterol

0.52

(0.69)

0.61

(0.82)

0.58

(0.78)

0.67

(0.90)

Protein

0.39

(0.48)

0.38

(0.46)

0.50

(0.61)

0.56

(0.68)

Carbohydrates

0.37

(0.43)

0.59

(0.68)

0.69

(0.79)

0.68

(0.78)

Mean

0.44

0.51

0.61

0.65

(0.57)

(0.65)

(0.78)

(0.83)

for the 1984 questionnaire, and 0.78 for the 1986 questionnaire. The mean correlation increased to 0.83 when the averages of nutrients from the three questionnaires were compared with the averages from the 1980 and 1986 diet records. This correlation was appreciably greater than the average correlations derived from evaluations of single comprehensive FFQs, which range from 0.40 to 0.70. These data indicate that repeated measures of diet by FFQs provide useful measures of long-term average dietary intakes over a 6-year period. This finding is critically important in studies of chronic diseases that develop over a period of many years, such as cancers or atherosclerosis. Consistent with these findings, we found in a previous analysis that use of the cumulative average of dietary intakes of fatty acids was more predictive of coronary heart disease (CHD) risk than use of only baseline diet.90

Underreporting and Adjustment for Total Energy Intake

Underreporting of total energy intake by dietary instruments is widely recognized. Livingstone and Black72 conducted a comprehensive review of studies in which energy intake (EI) was reported and energy expenditure (EE) was measured using the DLW method. Under the condition of stable weight, EI should equal EE. Under- and overreporters were identified by using EI:EE <0.82 or >1.18. Figure 6.5 shows EI:EE from 43 studies of adults (73 subgroups), with a mean ± SD EI:EE of 0.83 ± 0.14. In 29% of subgroups, EI and EE agreed to within ±10%, but 69% of the subgroups had a reported mean EI >10% below the mean EE, whereas less than 3% had a mean EI >10% above the mean EE. There were no significant differences between different dietary assessment methods (Table 6.3). However, in the Observing Protein and Energy Nutrition (OPEN) Study, Subar et al.91 found a greater degree of underreporting with FFQs than with 24-hour recalls. The likelihood of underreporting appears to be strongly related to participants’ weight status. In most of the studies, underreporting is more common in the obese than in normal weight subjects.

(p.102)

                      Dietary Assessment Methods

Figure 6.5 Frequency distribution of the ratio of energy intake to energy expenditure (EI:EE) by sex in 43 doubly labeled water (DLW) energy expenditure (DLW-EE) studies of adults comprising 77 subgroups (men and women separately). Reproduced with permission from Livingstone MB, Black AE. Markers of the validity of reported energy intake. J Nutr. 2003;133(Suppl 3):895S–920S.72

Table 6.3 Comparison of Reported Energy Intake by Dietary Assessment Method with Energy Expenditure Measured by Doubly Labeled Water

Dietary Method

N (# Studies)

Mean (EI/EE Ratio)

SD

Observation

5

1.06

0.09

Weighed records*

22

0.84

0.11

Estimated records*

25

0.84

0.10

Diet history

4

0.84

0.14

Twenty-four-hour recall (single or multiple)

6

0.84

0.08

Food-frequency questionnaire

6

0.87

0.12

All

68

0.86

0.13

(*) Excluding studies on subjects recruited as obese or as large or small eaters.

Reproduced with permission from Livingstone MB, Black AE. Markers of the validity of reported energy intake. J Nutr. 2003;133(Suppl 3):895S–920S.72

The impact of underreporting energy intake on the analysis and interpretation of diet and disease relationships is uncertain. The main interest in nutritional epidemiology is the composition of the diet, as represented by energy-adjusted nutrient intakes (see below), rather than simply increasing or decreasing energy-bearing nutrient intakes.1 In addition, in epidemiologic studies, factors that may be related to underreporting of energy intakes, (p.103) such as age, sex, and BMI, are often adjusted for in the analyses. Total energy intake is seldom used as an exposure or outcome variable because it is difficult to measure and interpret. In free-living populations, between-person variations in total energy intake are primarily determined by individual differences in physical activity, body size, and metabolic efficiency; energy balance over a period of time is primarily reflected in body weight change. After controlling for total or lean body mass, the variation in total energy intake appears to be primarily determined by physical activity levels. The substantial contribution of physical activity to between-person variations in total energy intake may explain a positive association between total energy intake and physical activity levels observed in some epidemiologic studies.92 Although there are individual differences in metabolic efficiency, it is infeasible to measure them in epidemiologic studies.

The best way to deal with underreporting in analyses is unclear. Some studies exclude underreporters from the data set. This approach not only reduces power, but may also introduce selection bias because subjects with high BMI levels are more likely to be excluded.72 Nonetheless, this approach can be used in sensitivity or secondary analyses. For example, in the analysis of the relationship between dietary fat and CHD, we calculated the ratio of reported caloric intake to predicted caloric intake for each participant using their age and weight. Excluding women with the greatest likelihood for underreporting (the lowest quintile of the ratio) did not change the associations.90 It is possible that simultaneous adjustment for BMI and total energy in the overall analyses may have already taken care of potential biases caused by underreporting.

These analyses demonstrate the importance of measuring and adjusting for total energy intake in epidemiologic studies. Adjustment for total energy intake in data analyses has several conceptual and practical advantages.14 First, control for total energy in epidemiologic studies mimics isocaloric substitution of one macronutrient (e.g., fat) for another (e.g., carbohydrates) in controlled experimental studies. In most situations, dietary composition rather than absolute intake is the primarily interest in nutritional epidemiologic studies. Absolute increase or decrease in nutrient or food intakes can often lead to changes in total energy intake. Unless physical activity levels are also changed, the changes in food intake will, theoretically, lead to weight gain or loss that make it difficult to interpret the nutrient-disease association. Second, because measurement errors for energy and nutrient intakes are correlated, they tend to cancel each other in energy-adjusted nutrients. The correlated errors between nutrients and energy typically result from overreporting or underreporting of specific foods. In many validation studies, the correlation coefficients between nutrient intakes calculated by the FFQs and reference methods improve after adjusting for energy intake,1 which can be largely attributed to reduced measurement errors. Third, energy adjustment also removes “extraneous variation” that results from the differences in energy requirements among individuals of different body sizes and physical activity levels. It should be noted that because energy adjustment leads to a reduction in between-person variations in nutrient intakes, it can sometimes reduce the correlations between energy-adjusted nutrient intakes calculated by the FFQ and diet records.

Another reason to adjust for total energy intake is to control for confounding in cases where total energy is associated with disease risk, and a spurious association between nutrient intake and disease may occur because of confounding by total energy intake. Using an energy-adjusted nutrient instead of absolute intake should eliminate such confounding because this variable is, by definition, not correlated with total energy intake.

The most commonly used method to adjust for total energy is to calculate nutrient density (i.e., percentage of calories contributed by a macronutrient). Public health recommendations are generally expressed in these units. For nonenergy contributing nutrients, density (p.104) can be expressed as absolute intake per 1,000 cal. The major limitation of using nutrient densities in epidemiologic studies is that it does not control adequately for confounding by total energy intake. An alternative method proposed by Willett and Stampfer93 is to calculate nutrient residuals. In this method, energy-adjusted nutrient intakes are computed as the residuals from a regression model, with total energy intake as the independent variable and absolute nutrient intake as the dependent variable (Fig. 6.6). Thus, the nutrient residuals, by definition, provide a measure of nutrient intake uncorrelated with total energy. Because the residuals have a mean of zero and include negative values, a constant is added to make them more interpretable. Typically, the predicted nutrient intake for the mean energy intake of the study population is added to the residuals. In multivariate analyses of nutrient-disease relationships, several statistical models are available for energy adjustments (Table 6.4): (a) the standard multivariate model; (b) the nutrient residual model; (c) the energy-partition model; and (d) the multivariate nutrient density model. The standard multivariate model includes total energy along with absolute intake of the nutrient of interest. The nutrient residual model includes the nutrient residuals obtained by regressing

                      Dietary Assessment Methods

Figure 6.6 Calorie-adjusted intake = a + b, where a = residual for subject from regression model with nutrient intake as the dependent variable and total caloric intake as the independent variable and b = the expected nutrient intake for a person with mean caloric intake. Reproduced with permission from Willett W, Stampfer MJ. Total energy intake: implications for epidemiologic analyses. Am J Epidemiol. 1986;124:17–27.93

Table 6.4 Statistical Models for Adjusting for Total Energy Intake in Epidemiologic Analyses

Model

Relation Expressed

Model 1A (standard multivariate)

Disease risk = β1 nutrient + α (total energy)

Model 1B (residual nutrient)

Disease risk = β1 nutrient residual* + β2 total energy

Model 1C (energy partition)

Disease risk = (α + β1) nutrient + α (energy from nonnutrient sources)

Model 2 (multivariate nutrient density)

Disease risk = β3 nutrient/total energy + β4 total energy

(*) “Nutrient residual” is the residual from the regression of a specific nutrient on energy.

From Willett WC. Nutritional Epidemiology. 2nd ed. New York: Oxford University Press; 1998.1

(p.105) nutrient intake as an independent variable on total energy intake (see above); total energy intake is also included as a covariate. In the energy-partition model, total energy is partitioned into contributions by different energy sources. The coefficient for one type of energy source (e.g., fat) represents the effect of increasing absolute intake of this nutrient while holding other energy sources (i.e., protein and carbohydrates) constant. Therefore, it represents the effect of adding fat, which includes both its energy and nonenergy effect. This would be analogous to conducting a trial holding protein or carbohydrate constant, while varying the amount of fat as well as total calories. The multivariate nutrient density model includes the nutrient densities (percentages of energy) from the macronutrients of interest (e.g., fat) as exposure variables with total energy included as a covariate. The coefficients from this model also have an isocaloric interpretation. It represents the substitution of fat for an equal amount of energy from carbohydrate (if percentage of energy from protein is included in the model), but in units of the percentage of energy.

Although the first three methods are mathematically equivalent when the nutrients are in their continuous form,94 the interpretation of the coefficients varies. For the standard, residual, and nutrient density models, the estimated associations all have the “isocaloric substitution” interpretation. However, interpretations of the associations for total energy are different. In the standard model, the associations should be interpreted as the effect of the sources of energy that are not included in the model, whereas the meaning of total energy intake is retained in the residual and nutrient density models. In the energy-partition model, total energy is not held constant. Therefore, an increase in absolute intake of one macronutrient will lead to an increase in total energy intake. As discussed earlier, the consequence of altered energy intake on body weight can complicate the interpretation of the nutrient-disease association.

In many circumstances, dietary variables are categorized according to quartiles or quintiles to avoid incorrect specification of the model, and to reduce the influence of outliers. Unfortunately, statistical equivalence among the standard, residual, and energy-partition models ceases to exist once a nutrient has been categorized. In a comparison of the standard multivariate and residual methods, Brown et al.94 found the latter more powerful for detecting linear trends in associations, and more robust to residual confounding when the adjustment variable was categorized. We have also reported that when dietary exposures were categorized into quantiles, the residual and nutrient density models yielded stronger associations and narrower 95% confidence intervals for the associations between polyunsaturated and trans fats and risk of CHD than did the standard multivariate and energy-partition models.95

Adjusting for Total Energy Intake in Obesity Epidemiologic Studies

Because weight gain results from positive energy balance, total energy intake can be interpreted as an intermediate end point between higher consumption of energy-bearing nutrients, foods, or beverages and subsequent weight gain. Thus, in studies on determinants of obesity and weight gain, total energy intake should be treated as a mediator rather than a confounder. For example, in analyzing the relationship between macronutrient intake (e.g., dietary fat) and weight gain, it is best to use percentage of calories from fat as the exposure variable. Not controlling for total calories in the multivariate model allows testing of the hypothesis that higher consumption of dietary fat may lead to obesity through passive overconsumption of calories. In an alternative analysis, total calories can be added to the model to see whether the association between the macronutrient and weight gain is mediated through excess energy intake.

Schulze et al.96 examined the relationship between sugar-sweetened soft drink consumption, calorie intake, and weight gain in a large cohort of younger and middle-aged women. Women who increased their sugar-sweetened soft drink consumption between 1991 and (p.106)

                      Dietary Assessment Methods

Figure 6.7 Mean change in energy consumption according to time trends in sugar-sweetened soft drink consumption between 1991 and 1995 in 51,603 women in the Nurses’ Health Study II. Reproduced with permission from Schulze MB, Manson JE, Ludwig DS, et al. Sugar-sweetened beverages, weight gain, and incidence of type 2 diabetes in young and middle-aged women. JAMA. 2004;292:927–34.96

1995 from low (≤1 per week) to high (≥1 per day) significantly increased their reported total energy intake by an average of 358 kcal/day (Fig. 6.7), whereas women who reduced their sugar-sweetened soft drink consumption between 1991 and 1995 also reduced their total energy consumption by a mean of 319 kcal/day. Changes in energy intake from food sources other than sugar-sweetened soft drinks accounted for about one third of the changes in total energy intake. These findings suggest that intake of sugar-sweetened beverages is an important source of excess calories, thereby resulting in a positive caloric balance and development of obesity. In this situation, to estimate the association between soft drink consumption and weight gain, the increase in total calories should be treated as an intermediate variable.

Measurement Error Corrections

All dietary assessment methods inevitably lead to measurement errors, which include random errors due to day-to-day variations in food intakes and systematic errors arising from inaccurate assessments of food intake frequency and portion sizes, errors in food composition tables, and selective underreporting or overreporting of consumption of certain foods.1 In epidemiologic research, daily variations in intakes of specific nutrients have been studied extensively using the analysis of variance technique.97 99 Beaton et al.98 observed that ratios of the within-person and between-person components of variance differed tremendously across nutrients, ranging from 1.0 for calories to >100 for vitamin A in men, and 1.4 for calories to 47.6 for vitamin A in women. Similarly, Willett reported that the ratios ranged from 1.9 for calories to 11.7 for vitamin A among (p.107) 173 women.1 In that total energy intake is quite well regulated by physiologic mechanisms, there is relatively low day-to-day variation for total calories. In contrast, the concentration of vitamin A and other micronutrients in certain foods can cause intake to vary considerably from day to day, depending on food choices and seasonable availability of foodstuffs.

Because of day-to-day variations in diet, a single 24-hour recall provides a poor estimate of a person’s usual diet. However, repeated measures can be used to improve the estimate. In validation studies, repeated measures of the reference method are commonly used to correct for within-person variations in dietary intakes. For example, deattenuated correlation coefficients for the nutrient of interest between a FFQ and weighed diet records were corrected for week-to-week variation in diet records by using the following formula.99

(6.1)                       Dietary Assessment Methods
where r t is the corrected correlation between the dietary pattern scores derived from the FFQ and diet records, r o is the observed correlation, γ is the ratio of estimated within-person and between-person variation in nutrient intakes derived from the two 1-week diet records, and k is the number of repeated observations of diet records. Conceptually, these corrected correlations provide an estimate of the correlations between the FFQ and true intake whereby each person’s intake was measured by a very large number of diet records. Rosner and Willett100 provided an estimate of the standard error for the corrected or deattenuated correlation coefficient and an associated 100% × (1 − α) confidence interval.

In epidemiologic studies of diet and disease risk, the regression calibration approach can be used to correct for both random and systematic errors, but this approach requires a validation study in a subsample of the cohort. Rosner et al.101 developed a method to correct odds ratio estimates from logistic regression models for measurement errors in continuous exposures within cohort studies; these errors could be systematic or due to random within-person variation. Let X denote true dietary intake by a reference method and Z denote surrogate exposure by a FFQ. Ignoring measurement error, the logistic regression model for regressing a dichotomous disease variable D on Z,

(6.2)                       Dietary Assessment Methods

True intake (X) is estimated as a function of observed surrogate intake (Z) from a regression derived from validation study data (X = α′ + λZ + ε). The corrected β* is obtained by

(6.3)                       Dietary Assessment Methods
where β is the uncorrected logistic regression coefficient of D on Z from the main study (from equation 6.2), and λ is the estimated regression slope of X on Z from the validation study.

Koh-Banerjee et al.102 extended the regression calibration method to estimate regression coefficients adjusted for measurement error in an analysis of the relationship between changes in diet and 9-year gain in waist circumference. Such an analysis requires validation studies conducted at two separate time points.

  • Let: X 1 represent true dietary intake (diet record) in time 1; X 2 represent true dietary intake (diet record) in time 2.

    • Z 1 represent dietary intake measured by a surrogate (FFQ) in time 1.

    • Z 2 represent dietary intake measured by a surrogate (FFQ) in time 2.

(p.108) Using the same regression calibration approach discussed earlier, the change in true dietary intakes (X 2X 1) is estimated as a function of the change in surrogate intakes (Z 2Z 1) derived from the validation study data:

(6.4)                       Dietary Assessment Methods

In a linear regression with the amount of weight or waist change as the outcome, the corrected point estimate for the exposure measure (i.e., difference in observed dietary intake over time) is:

                      Dietary Assessment Methods
where β is the estimated (or uncorrected) linear regression coefficient from the main study, and γ is the estimated regression slope of changes in X on changes in Z from the validation studies.

Using this method, Koh-Banerjee et al.102 estimated that after error correction, the substitution of trans fats as 2% of energy for polyunsaturated fats was associated with a 2.7 cm increase in waist circumference over 9 years (P < .001) (as compared with a 0.77 cm waist gain, uncorrected). An increase of 12 g fiber/day (r = .68 between FFQs and diet records) was associated with a 2.21 cm reduction in waist circumference after error correction (P < .001) (0.63 cm waist gain, uncorrected). The same method was employed by Liu et al.103 to correct for measurement error in the analyses of changes in dietary fiber intake and weight gain during 12 years of follow-up in the NHS. After further correction for measurement errors in changes in dietary fiber intake, they estimated that an increase of 12 g in dietary fiber intake was associated with ≈3.5 kg (8 lb) less weight gain in 12 years.

Dietary Pattern Analyses

A growing interest in the study of overall dietary patterns in relation to obesity and chronic diseases104 has been spurred, in part, by several conceptual and methodological challenges associated with the traditional approach of examining individual nutrients and foods. These include high levels of intercorrelations among nutrients and foods, lack of consideration of synergistic or cumulative effects of multiple nutrients, multiple comparison problems, and confounding by other dietary components. Patterns are characterized based on similarity of habitual food use, which minimizes confounding by other foods or nutrient. Thus, in dietary pattern analysis, the collinearity of nutrients and foods can be used to advantage. Classifying individuals according to their overall eating pattern (i.e., by considering how foods and nutrients are consumed in combination) can yield a larger contrast between exposure groups than analyses based on single nutrients. Because overall patterns of dietary intake might be easy for the public to interpret or translate into diets, research on dietary patterns could have important public health implications.

Several methods have been commonly used to characterize dietary patterns using collected dietary information, including factor analysis, cluster analysis, and dietary indices. Factor analysis, as a generic term, includes both principal component analysis (PCA) and common factor analysis. PCA is commonly used to define dietary patterns because the principal components are expressed by certain mathematical functions of the observed consumption of food items.105 The method aggregates specific food items or food groups based on the degree to which food items in the data set are correlated with one another. A summary score for each pattern is then derived and can be used in (p.109) either correlation or regression analysis to examine relationships between various eating patterns and outcomes of interest, such as weight gain106 and chronic diseases.107 , 108 In a validation study, we found that two major patterns (the prudent and Western patterns) identified through PCA of food consumption data assessed by FFQs were reproducible over time and correlated reasonably well with the patterns identified from two 1-week diet records.109

Cluster analysis is another multivariate method for characterizing dietary patterns. In contrast to factor analysis, cluster analysis aggregates individuals into relatively homogeneous subgroups (clusters) with similar diets. Individuals have been clustered on the basis of the frequency of food consumed; the percentage of energy contributed by each food or food group, the average grams of food intakes, or standardized nutrient intakes.110 112 When the cluster procedure is completed, further analyses (e.g., comparing dietary profiles across clusters) are necessary to interpret the identified patterns.113 Earlier studies have examined the relationships between dietary patterns identified by cluster analysis and weight gain.114

A variety of dietary indices have been proposed to assess overall diet quality.115 , 116 These are typically constructed on the basis of dietary recommendations or existing dietary patterns. Commonly used dietary indices include the diet quality index-revised (DQI),117 the USDA Healthy Eating Index (HEI),118 the Recommended Food Score (RFS),119 and the Mediterranean Diet Index (MDI).120 , 121 We have created an alternate HEI (AHEI) by incorporating components regarding trans fat, polyunsaturated to saturated fat ratio, moderate alcohol use, and multivitamin use.122 Several studies have examined the relationships between various dietary indices and risk of obesity and weight gain.123

Both factor and cluster analyses are considered a posteriori because the eating patterns are derived through statistical modeling of dietary data at hand.124 The dietary index approach, in contrast, is a priori, because the indices are created based on previous knowledge of a healthy diet. Because the factor and cluster analysis approaches generate patterns based on available empirical data without a priori hypotheses, they do not necessarily represent optimal patterns for disease prevention. On the other hand, the dietary index approach is limited by current knowledge and understanding of diet-disease relationships, and can also be fraught with uncertainties in selecting individual components of the score and subjectivity in defining cutoff points. Typically, dietary indices are constructed on the basis of prevailing dietary recommendations, some of which may not represent the best available evidence.104

A newly developed approach that bridges the gap between the two major dietary pattern approaches is Reduced Rank Regression (RRR).125 127 The RRR method takes into account the biological pathways from diet to outcome by identifying dietary patterns associated with biomarkers of a specific disease, then uses the identified patterns to predict disease occurrence. Unlike analyses using PCA, which derives dietary patterns based on observed covariance among food groups, the RRR method utilizes information on biomarkers to derive dietary patterns (Fig. 6.8). Thus, this approach is considered a combination of a priori and a posteriori methods.

Because RRR patterns are derived based on biological pathways rather than patterns of eating behavior, they could be more predictive of disease risk. The method has been applied to studies of CHD128 and type 2 diabetes.129 The RRR approach requires response (biomarker) information. However, such data may not be available in many studies. Also, the biomarker information available in a study may not reflect the current status of knowledge. Nonetheless, the hypothesis-driven nature of the RRR method is considered complementary to the traditional factor analysis approach.

(p.110)

                      Dietary Assessment Methods

Figure 6.8 Approaches to define dietary patterns in epidemiologic studies. Adapted from Schulze MB, Hoffmann K. Methodological approaches to study dietary patterns in relation to risk of coronary heart disease and stroke. Br J Nutr. 2006 May;95:860–869.127

Summary

Despite many advances in dietary assessment methodologies in the past two decades, it remains a major challenge in epidemiologic studies to accurately quantify dietary intakes in free-living populations. Sufficiently valid dietary assessment is particularly important for studying dietary determinants of obesity. However, such studies are made exceedingly complex by combinations of large day-to-day variations in nutrient and food intakes, biased reports associated with obesity status, and difficulties in controlling for confounding variables. The same methodological problems also apply to nutritional epidemiologic studies on dietary determinants of chronic disease risk. These require careful choice of the most appropriate dietary assessment methods, and rigorous validation studies and statistical analyses and interpretation.

Because there is no perfect dietary assessment method, choice of a dietary instrument has to balance strengths and limitations of various methods in specific research settings. FFQs offer low cost and the ability to assess usual diet, the main interest in most epidemiologic studies of obesity and related chronic diseases. In the past two decades, the FFQ has become the method of choice for large epidemiologic studies involving hundreds and thousands of people. However, the validity of a FFQ is population- and culture specific; it is crucial to consider population characteristics, such as age, sex, education/literacy, and cultural characteristics when developing a FFQ.

One emerging trend in nutritional epidemiologic studies is to assess overall dietary patterns in relation to obesity and chronic diseases. Such analyses take advantage of a FFQ’s ability to assess habitual diet and examine cumulative effects of overall diet. Dietary pattern analysis will certainly not replace nutrient or food analysis, but instead, be complementary to more traditional analysis. Evidence is enhanced when the results from multiple lines of research (i.e., biomarkers of nutrient intake, nutrients, foods, and dietary patterns) are consistent.

Another emerging trend in large epidemiologic studies is to collect repeated measures of diet during follow-up. These can be used to reduce measurement error and best represent long-term diet. They can be useful in correcting for both random and systematic measurement errors in analyses of changes in diet, body weight, or waist circumference over time.

(p.111) A recent development in large epidemiologic studies is to combine different types of dietary assessment methods. For example, NHANES III added a FFQ to its 24-hour recall.130 In the EPIC-Norfolk study, 7-day diet records and FFQ data were collected simultaneously from subjects who were willing to provide food records.131 Although the combined approach may provide a more complete picture on the complexity of diet, different data from the various methods may create a dilemma in the interpretation of the findings. In addition, the cost of collecting food record data from hundreds of thousands of people is often prohibitive for most cohort studies, especially if repeated measures of diet are considered.

There is a general consensus that adjusting for total energy intake when estimating individual nutrient intake from FFQs can reduce correlated errors and improve estimates in validation studies. Thus, in most situations, nutrient density (percentage of calories from specific macronutrients) or nutrient residuals should be used as the primary exposure variable in studying dietary determinants of obesity or chronic diseases. Total energy is typically adjusted in multivariate analyses of diet and incidence of chronic diseases (e.g., heart disease or cancer) to simulate isocaloric substitution of one macronutrient (e.g., fat) for another (e.g., carbohydrates). Whether such adjustment should be done in studies on dietary determinants of obesity and weight gain is more complicated because total energy is considered an intermediate biological variable between macronutrient intake and body weight. If the main interest is weight gain in relation to changes in dietary composition (without change in energy intake), then total energy intake should be adjusted in the model. On the other hand, if one hypothesizes that increased consumption of a particular nutrient or food may lead to an increase in subsequent energy intake, then total energy should not be controlled for in the model.

It has become critically important for large cohort studies of obesity and chronic diseases to collect and store biological samples such as plasma, red blood cells, toenails, and DNA for biomarker analyses. Effective biomarkers of nutrient intake and status can provide valuable data on the validity of dietary assessment methods, and also serve as an objective and time-integrated dietary exposure for some nutrients that are difficult to assess through self-reports. However, biomarkers are a complement to, rather than a replacement for, dietary assessment methods in epidemiologic studies. The combination of repeated measures of FFQs and biomarker data is likely to provide reasonably accurate measures of long-term diet in large cohort studies of obesity and chronic diseases.

References

Bibliography references:

1. Willett WC. Nutritional Epidemiology. 2nd ed. New York: Oxford University Press; 1998.

2. Gibson RS. Principles of Nutritional Assessment. 2nd ed. New York: Oxford University Press; 2005.

3. Buzzard M. 24-hour dietary recall and food record methods. In: Willet WC, ed. Nutritional Epidemiology. 2nd ed. New York: Oxford University Press; 1998;50–73.

4. Posner BM, Borman CL, Morgan JL, Borden WS, Ohls JC. The validity of a telephone-administered 24-hour dietary recall methodology. Am J Clin Nutr. 1982;36:546–553.

5. Buzzard IM, Faucett CL, Jeffery RW, et al. Monitoring dietary change in a low-fat diet intervention study: advantages of using 24-hour dietary recalls vs food records. J Am Diet Assoc. 1996;96:574–579.

6. Feskanich D, Sielaff BH, Chong K, Buzzard IM. Computerized collection and analysis of dietary intake information. Comput Methods Programs Biomed. 1989;30:47–57.

7. Slimani N, Ferrari P, Ocke M, et al. Standardization of the 24-hour diet recall calibration method used in the European prospective investigation into cancer and nutrition (EPIC): general concepts and preliminary results. Eur J Clin Nutr. 2000;54:900–917.

(p.112) 8. Johnson RK. Dietary intake—how do we measure what people are really eating? Obes Res. 2002;10(Suppl 1):63S–68S.

9. Conway JM, Ingwersen LA, Moshfegh AJ. Accuracy of dietary recall using the USDA five-step multiple-pass method in men: an observational validation study. J Am Diet Assoc. 2004;104:595–603.

10. Conway JM, Ingwersen LA, Vinyard BT, Moshfegh AJ. Effectiveness of the US Department of Agriculture 5-step multiple-pass method in assessing food intake in obese and nonobese women. Am J Clin Nutr. 2003;77:1171–1178.

11. Johnson RK, Driscoll P, Goran MI. Comparison of multiple-pass 24-hour recall estimates of energy intake with total energy expenditure determined by the doubly labeled water method in young children. J Am Diet Assoc. 1996;96:1140–1144.

12. Bingham SA, Welch AA, McTaggart A, et al. Nutritional methods in the European Prospective Investigation of Cancer in Norfolk. Public Health Nutr. 2001;4:847–858.

13. Burke BS. The dietary history as a tool in research. J Am Diet Assoc. 1947;23:1041–1046.

14. Willett WC. Food-Frequency Methods. Nutritional Epidemiology. 2nd ed. New York: Oxford University Press; 1998;74–100.

15. Block G, Thompson FE, Hartman AM, Larkin FA, Guire KE. Comparison of two dietary questionnaires validated against multiple dietary records collected during a 1-year period. J Am Diet Assoc. 1992;92:686–693.

16. Subar AF, Thompson FE, Kipnis V, et al. Comparative validation of the Block, Willett, and National Cancer Institute food-frequency questionnaires: the Eating at America’s Table Study. Am J Epidemiol. 2001;154:1089–1099.

17. Subar AF. Developing dietary assessment tools. J Am Diet Assoc. 2004;104:769–770.

18. Tjonneland A, Haraldsdottir J, Overvad K, Stripp C, Ewertz M, Jensen OM. Influence of individually estimated portion size data on the validity of a semiquantitative food-frequency questionnaire. Int J Epidemiol. 1992;21:770–777.

19. Orencia AJ, Daviglus ML, Dyer AR, Shekelle RB, Stamler J. Fish consumption and stroke in men. 30-year findings of the Chicago Western Electric Study. Stroke. 1996;27:204–209.

20. Liu K, Slattery M, Jacobs D Jr, et al. A study of the reliability and comparative validity of the cardia dietary history. Ethn Dis. 1994;4:15–27.

21. Schmidt LE, Cox MS, Buzzard IM, Cleary PA. Reproducibility of a comprehensive diet history in the Diabetes Control and Complications Trial. The DCCT Research Group. J Am Diet Assoc. 1994;94:1392–1397.

22. Hunter D. Biochemical indicators of dietary intake. In: Willet WC, ed. Nutritional Epidemiology. 2nd ed. New York: Oxford University Press; 1998.

23. Speakman JR. Doubly Labeled Water: Theory and Practice. London: Chapman & Hall; 1997.

24. Trabulsi J, Schoeller DA. Evaluation of dietary assessment instruments against doubly labeled water, a biomarker of habitual energy intake. Am J Physiol Endocrinol Metab. 2001;281:E891–E899.

25. Schoeller DA, Schoeller DA. Validation of habitual energy intake. Public Health Nutr. 2002;5:883–888.

26. Horton ES. Introduction: an overview of the assessment and regulation of energy balance in humans. Am J Clin Nutr. 1983;38:972–977.

27. Schofield WN. Predicting basal metabolic rate, new standards and review of previous work. Hum Nutr Clin Nutr. 1985;39(Suppl 1):5–41.

28. Mifflin MD, St Jeor ST, Hill LA, Scott BJ, Daugherty SA, Koh YO. A new predictive equation for resting energy expenditure in healthy individuals. Am J Clin Nutr. 1990;51:241–247.

29. Goldberg GR, Black AE, Jebb SA, et al. Critical evaluation of energy intake data using fundamental principles of energy physiology: 1. Derivation of cut-off limits to identify under-recording. Eur J Clin Nutr. 1991;45:569–581.

(p.113) 30. Black AE. Critical evaluation of energy intake using the Goldberg cut-off for energy intake: basal metabolic rate. A practical guide to its calculation, use and limitations. Int J Obes Relat Metab Disord. 2000;24:1119–1130.

31. Bingham SA, Cummings JH. Urine nitrogen as an independent validatory measure of dietary intake: a study of nitrogen balance in individuals consuming their normal diet. Am J Clin Nutr. 1985;42:1276–1289.

32. Matthews DE. Proteins and amino acids. In: Shils ME, Olson JA, Shike M, et al., eds. Modern Nutrition in Health and Disease. 9th ed. Baltimore, MD: Williams & Wilkins Company; 1999:11–48.

33. Margetts BM, Nelson M. Design Concepts in Nutritional Epidemiology. 2nd ed. New York: Oxford University Press; 1997.

34. Kesteloot H, Joossens JV. The relationship between dietary intake and urinary excretion of sodium, potassium, calcium, and magnesium: Belgian Interuniversity Research on Nutrition and Health. J Hum Hypertens. 1990;4:527–533.

35. Liu K, Stamler J. Assessment of sodium intake in epidemiological studies on blood pressure. Ann Clin Res. 1984;16(Suppl):49–54.

36. Holbrook JT, Patterson KY, Bodner JE, et al. Sodium and potassium intake and balance in adults consuming self-selected diets. Am J Clin Nutr. 1984;40:786–793.

37. Zhang J, Temme EH, Sasaki S, Kesteloot H. Under- and overreporting of energy intake using urinary cations as biomarkers: relation to body mass index. Am J Epidemiol. 2000;152:453–462.

38. Mensink RP, Katan MB. Effect of dietary fatty acids on serum lipids and lipoproteins. A meta-analysis of 27 trials. Arterioscler Thromb. 1992;12:911–919.

39. Willett W, Stampfer M, Chu NF, Spiegelman D, Holmes M, Rimm E. Assessment of questionnaire validity for measuring total fat intake using plasma lipid levels as criteria. Am J Epidemiol. 2001;154:1107–1112.

40. Baylin A, Campos H. The use of fatty acid biomarkers to reflect dietary intake. Curr Opin Lipidol. 2006;17:22–27.

41. Keys A. Serum cholesterol response to dietary cholesterol. Am J Clin Nutr. 1984;40:351–359.

42. Hegsted DM, Ausman LM, Johnson JA, Dallal GE. Dietary fat and serum lipids: an evaluation of the experimental data. Am J Clin Nutr. 1993;57:875–883.

43. Sun Q, Ma J, Campos H, Hankinson SE, Hu FB. Comparison between plasma and erythrocyte fatty acid content as biomarkers of fatty acid intake in U.S. women. Am J Clin Nutr. 2007;86(1):74–81.

44. Baylin A, Kim MK, Donovan-Palmer A, et al. Fasting whole blood as a biomarker of essential fatty acid intake in epidemiologic studies: comparison with adipose tissue and plasma. Am J Epidemiol. 2005;162:373–381.

45. Dayton S, Hashimoto S, Dixon W, Pearce ML. Composition of lipids in human serum and adipose tissue during prolonged feeding of a diet high in unsaturated fat. J Lipid Res. 1966;7:103–111.

46. Ma J, Folsom AR, Shahar E, Eckfeldt JH. Plasma fatty acid composition as an indicator of habitual dietary fat intake in middle-aged adults. The Atherosclerosis Risk in Communities (ARIC) Study Investigators. Am J Clin Nutr. 1995;62:564–571.

47. Wolk A, Vessby B, Ljung H, Barrefors P. Evaluation of a biological marker of dairy fat intake. Am J Clin Nutr. 1998;68:291–295.

48. Baylin A, Kabagambe EK, Siles X, Campos H. Adipose tissue biomarkers of fatty acid intake. Am J Clin Nutr. 2002;76:750–757.

49. Sun Q, Ma J, Campos H, Hankinson S, Hu FB. Plasma and erythrocyte biomarkers of dairy fat intake and risk of coronary heart disease. Am J Clin Nutr. 2007;86:929–937.

50. Liu S, Manson JE, Stampfer MJ, et al. Dietary glycemic load assessed by food-frequency questionnaire in relation to plasma high-density-lipoprotein cholesterol and fasting plasma triacylglycerols in postmenopausal women. Am J Clin Nutr. 2001;73:560–566.

(p.114) 51. Selhub J. Folate, vitamin B12 and vitamin B6 and one carbon metabolism. J Nutr Health Aging. 2002;6:39–42.

52. Stanger O. Physiology of folic acid in health and disease. Curr Drug Metab. 2002;3:211–223.

53. Mason JB. Biomarkers of nutrient exposure and status in one-carbon (methyl) metabolism. J Nutr. 2003;133:941S–947S.

54. Cho E, Zeisel SH, Jacques P, et al. Dietary choline and betaine assessed by food-frequency questionnaire in relation to plasma total homocysteine concentration in the Framingham Offspring Study. Am J Clin Nutr. 2006;83:905–911.

55. Selhub J, Jacques PF, Wilson PW, Rush D, Rosenberg IH. Vitamin status and intake as primary determinants of homocysteinemia in an elderly population. JAMA. 1993;270:2693–2698.

56. Jacques PF, Sulsky SI, Sadowski JA, Phillips JC, Rush D, Willett WC. Comparison of micronutrient intake measured by a dietary questionnaire and biochemical indicators of micronutrient status. Am J Clin Nutr. 1993;57:182–189.

57. Green TJ, Allen OB, O’Connor DL. A three-day weighed food record and a semiquantitative food-frequency questionnaire are valid measures for assessing the folate and vitamin B-12 intakes of women aged 16 to 19 years. J Nutr. 1998;128:1665–1671.

58. van der Schouw YT, de Kleijn MJ, Peeters PH, Grobbee DE. Phyto-oestrogens and cardiovascular disease risk. Nutr Metab Cardiovasc Dis. 2000;10:154–167.

59. de Kleijn MJ, van der Schouw YT, Wilson PW, et al. Intake of dietary phytoestrogens is low in postmenopausal women in the United States: the Framingham study(1–4). J Nutr. 2001;131:1826–1832.

60. Karr SC, Lampe JW, Hutchins AM, Slavin JL. Urinary isoflavonoid excretion in humans is dose dependent at low to moderate levels of soy-protein consumption. Am J Clin Nutr. 1997;66:46–51.

61. Adlercreutz H, Fotsis T, Lampe J, et al. Quantitative determination of lignans and isoflavonoids in plasma of omnivorous and vegetarian women by isotope dilution gas chromatography-mass spectrometry. Scand J Clin Lab Invest Suppl. 1993;215:5–18.

62. Lampe JW. Isoflavonoid and lignan phytoestrogens as dietary biomarkers. J Nutr. 2003;133(Suppl 3):956S–964S.

63. Seow A, Shi CY, Franke AA, Hankin JH, Lee HP, Yu MC. Isoflavonoid levels in spot urine are associated with frequency of dietary soy intake in a population-based sample of middle-aged and older Chinese in Singapore. Cancer Epidemiol Biomarkers Prev. 1998;7:135–140.

64. Chen Z, Zheng W, Custer LJ, et al. Usual dietary consumption of soy foods and its correlation with the excretion rate of isoflavonoids in overnight urine samples among Chinese women in Shanghai. Nutr Cancer. 1999;33:82–87.

65. Hambidge M. Biomarkers of trace mineral intake and status. J Nutr. 2003;133(Suppl 3):948S–955S.

66. Liu JM, Hankinson SE, Stampfer MJ, Rifai N, Willett WC, Ma J. Body iron stores and their determinants in healthy postmenopausal US women. Am J Clin Nutr. 2003;78:1160–1167.

67. Swanson CA, Longnecker MP, Veillon C, et al. Selenium intake, age, gender, and smoking in relation to indices of selenium status of adults residing in a seleniferous area. Am J Clin Nutr. 1990;52:858–862.

68. Hunter DJ, Morris JS, Chute CG, et al. Predictors of selenium concentration in human toenails. Am J Epidemiol. 1990;132:114–122.

69. Longnecker MP, Stampfer MJ, Morris JS, et al. A 1-y trial of the effect of high-selenium bread on selenium concentrations in blood and toenails. Am J Clin Nutr. 1993;57:408–413.

70. Garland M, Morris JS, Colditz GA, et al. Toenail trace element levels and breast cancer: a prospective study. Am J Epidemiol. 1996;144:653–660.

(p.115) 71. Martini MC, Campbell DR, Gross MD, Grandits GA, Potter JD, Slavin JL. Plasma carotenoids as biomarkers of vegetable intake: the University of Minnesota Cancer Prevention Research Unit Feeding Studies. Cancer Epidemiol Biomarkers Prev. 1995;4:491–496.

72. Livingstone MB, Black AE. Markers of the validity of reported energy intake. J Nutr. 2003;133(Suppl 3):895S–920S.

73. Nelson M. The validation of dietary assessment. In: Margetts BM, Nelson MC, eds. Design Concepts in Nutritional Epidemiology. 2nd ed. New York: Oxford University Press; 1997;241–272;Chapter 8.

74. Rimm EB, Giovannucci EL, Stampfer MJ, Colditz GA, Litin LB, Willett WC. Reproducibility and validity of an expanded self-administered semiquantitative food-frequency questionnaire among male health professionals. Am J Epidemiol. 1992;135:1114.

75. Feskanich D, Rimm EB, Giovannucci EL, et al. Reproducibility and validity of food intake measurements from a semiquantitative food-frequency questionnaire. J Am Diet Assoc. 1993;93:790–796.

76. Salvini S, Hunter DJ, Sampson L, et al. Food-based validation of a dietary questionnaire: the effects of week-to-week variation in food consumption. Int J Epidemiol. 1989;18:858–867.

77. Kroke A, Klipstein-Grobusch K, Voss S, et al. Validation of a self-administered food-frequency questionnaire administered in the European Prospective Investigation into Cancer and Nutrition (EPIC) Study: comparison of energy, protein, and macronutrient intakes estimated with the doubly labeled water, urinary nitrogen, and repeated 24-h dietary recall methods. Am J Clin Nutr. 1999;70:439–447.

78. Shu XO, Yang G, Jin F, et al. Validity and reproducibility of the food-frequency questionnaire used in the Shanghai Women’s Health Study. Eur J Clin Nutr. 2004;58:17–23.

79. Caan BJ, Slattery ML, Potter J, Quesenberry CP Jr, Coates AO, Schaffer DM. Comparison of the Block and the Willett self-administered semiquantitative food-frequency questionnaires with an interviewer-administered dietary history. Am J Epidemiol. 1998;148:1137–1147.

80. Wirfalt AK, Jeffery RW, Elmer PJ. Comparison of food-frequency questionnaires: the reduced Block and Willett questionnaires differ in ranking on nutrient intakes. Am J Epidemiol. 1998;148:1148–1156.

81. Rimm EB, Klatsky A, Grobbee D, Stampfer MJ. Review of moderate alcohol consumption and reduced risk of coronary heart disease: is the effect due to beer, wine, or spirits. BMJ. 1996;312:731–736.

82. Giovannucci E, Colditz G, Stampfer MJ, et al. The assessment of alcohol consumption by a simple self-administered questionnaire. Am J Epidemiol. 1991;133:810–817.

83. Kroke A, Klipstein-Grobusch K, Hoffmann K, Terbeck I, Boeing H, Helander A. Comparison of self-reported alcohol intake with the urinary excretion of 5-hydroxytryptophol:5-hydroxyindole-3-acetic acid, a biomarker of recent alcohol intake. Br J Nutr. 2001;85:621–627.

84. Kaaks RJ. Biochemical markers as additional measurements in studies of the accuracy of dietary questionnaire measurements: conceptual issues. Am J Clin Nutr. 1997;65(4 Suppl):1232S–1239S.

85. Kabagambe EK, Baylin A, Allan DA, Siles X, Spiegelman D, Campos H. Application of the method of triads to evaluate the performance of food-frequency questionnaires and biomarkers as indicators of long-term dietary intake. Am J Epidemiol. 2001;154:1126–1135.

86. Ocke MC, Kaaks RJ. Biochemical markers as additional measurements in dietary validity studies: application of the method of triads with examples from the European Prospective Investigation into Cancer and Nutrition. Am J Clin Nutr. 1997;65(4 Suppl):1240S–1245S.

(p.116) 87. Bhakta D, dos Santos Silva I, Higgins C, et al. A semiquantitative food-frequency questionnaire is a valid indicator of the usual intake of phytoestrogens by south Asian women in the UK relative to multiple 24-h dietary recalls and multiple plasma samples. J Nutr. 2005;135:116–123.

88. McNaughton SA, Marks GC, Gaffney P, Williams G, Green A. Validation of a food-frequency questionnaire assessment of carotenoid and vitamin E intake using weighed food records and plasma biomarkers: the method of triads model. Eur J Clin Nutr. 2005;59:211–218.

89. Willett W. Invited commentary: a further look at dietary questionnaire validation. Am J Epidemiol. 2001;154:1100.

90. Hu FB, Stampfer MJ, Manson JE, et al. Dietary fat intake and the risk of coronary heart disease in women. N Engl J Med. 1997;337:1491–1499.

91. Subar AF, Kipnis V, Troiano RP, et al. Using intake biomarkers to evaluate the extent of dietary misreporting in a large sample of adults: the OPEN study. Am J Epidemiol. 2003;158:1–13.

92. Saltzman E, Roberts SB. The role of energy expenditure in energy regulation: findings from a decade of research. Nutr Rev. 1995;53:209–220.

93. Willett W, Stampfer MJ. Total energy intake: implications for epidemiologic analyses. Am J Epidemiol. 1986;124:17–27.

94. Brown CC, Kipnis V, Freedman LS, Hartman AM, Schatzkin A, Wacholder S. Energy adjustment methods for nutritional epidemiology: the effect of categorization. Am J Epidemiol. 1994;139:323–338.

95. Hu FB, Stampfer MJ, Rimm E, et al. Dietary fat and coronary heart disease: a comparison of approaches for adjusting for total energy intake and modeling repeated dietary measurements. Am J Epidemiol. 1999;149:531–540.

96. Schulze MB, Manson JE, Ludwig DS, et al. Sugar-sweetened beverages, weight gain, and incidence of type 2 diabetes in young and middle-aged women. JAMA. 2004;292:927–934.

97. Beaton GH, Milner J, Corey P, et al. Sources of variance in 24-hour dietary recall data: implications for nutrition study design and interpretation. Am J Clin Nutr. 1979;32:2546–2559.

98. Beaton GH, Milner J, McGuire V, Feather TE, Little JA. 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. 1983;37:986–995.

99. 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.

100. Rosner B, Willett WC. Interval estimates for correlation coefficients corrected for within-person variation: implications for study design and hypothesis testing. Am J Epidemiol. 1988;127:377–386.

101. Rosner B, Willett WC, Spiegelman D. Correction of logistic regression relative risk estimates and confidence intervals for systematic within-person measurement error. Stat Med. 1989;8:1051–1069.

102. Koh-Banerjee P, Chu NF, Spiegelman D, et al. Prospective study of the association of changes in dietary intake, physical activity, alcohol consumption, and smoking with 9-y gain in waist circumference among 16 587 US men. Am J Clin Nutr. 2003;78:719–727.

103. Liu S, Willett WC, Manson JE, Hu FB, Rosner B, Colditz G. Relation between changes in intakes of dietary fiber and grain products and changes in weight and development of obesity among middle-aged women. Am J Clin Nutr. 2003;78:920–927.

104. Hu FB. Dietary pattern analysis: a new direction in nutritional epidemiology. Curr Opin Lipidol. 2002;13:3–9.

105. Kleinbaum DG, Kupper LL, Muller KE. Chapter 24. Variable reduction and factor analysis. Applied Regression Analysis and Other Multivariate Methods. Boston: PWS-KENT Publishing Company; 1988:595–640.

(p.117) 106. Newby PK, Muller D, Hallfrisch J, Andres R, Tucker KL. Food patterns measured by factor analysis and anthropometric changes in adults. Am J Clin Nutr. 2004;80:504–513.

107. van Dam RM, Rimm EB, Willett WC, Stampfer MJ, Hu FB. Dietary patterns and risk for type 2 diabetes mellitus in U.S. men. Ann Intern Med. 2002;136:201–209.

108. Fung TT, Hu FB, Holmes MD, et al. Dietary patterns and the risk of postmenopausal breast cancer. Int J Cancer. 2005;116:116–121.

109. Hu FB, Rimm E, Smith-Warner SA, et al. Reproducibility and validity of dietary patterns assessed with a food-frequency questionnaire. Am J Clin Nutr. 1999;69:243–249.

110. Millen BE, Quatromoni PA, Copenhafer DL, Demissie S, O’Horo CE, D’Agostino RB. Validation of a dietary pattern approach for evaluating nutritional risk: the Framingham Nutrition Studies. J Am Diet Assoc. 2001;101:187–194.

111. Tucker KL, Dallal GE, Rush D. Dietary patterns of elderly Boston-area residents defined by cluster analysis. J Am Diet Assoc. 1992;92:1487–1491.

112. Wirfalt AK, Jeffery RW. Using cluster analysis to examine dietary patterns: nutrient intakes, gender, and weight status differ across food pattern clusters. J Am Diet Assoc. 1997;97:272–279.

113. Newby PK, Tucker KL. Empirically derived eating patterns using factor or cluster analysis: a review. Nutr Rev. 2004;62:177–203.

114. Newby PK, Muller D, Hallfrisch J, Qiao N, Andres R, Tucker KL. Dietary patterns and changes in body mass index and waist circumference in adults. Am J Clin Nutr. 2003;77:1417–1425.

115. Fung TT, McCullough ML, Newby PK, et al. Diet-quality scores and plasma concentrations of markers of inflammation and endothelial dysfunction. Am J Clin Nutr. 2005;82:163–173.

116. Kant AK. Dietary patterns and health outcomes. J Am Diet Assoc. 2004;104:615–635.

117. Newby PK, Hu FB, Rimm EB, et al. Reproducibility and validity of the Diet Quality Index Revised as assessed by use of a food-frequency questionnaire. Am J Clin Nutr. 2003;78:941–949.

118. Kennedy ET, Ohls J, Carlson S, Fleming K. The Healthy Eating Index: design and applications. J Am Diet Assoc. 1995;95:1103–1108.

119. Kant AK, Schatzkin A, Graubard BI, Schairer C. A prospective study of diet quality and mortality in women. JAMA. 2000;283:2109–2115.

120. Martinez-Gonzalez MA, Fernandez-Jarne E, Serrano-Martinez M, Marti A, Martinez JA, Martin-Moreno JM. Mediterranean diet and reduction in the risk of a first acute myocardial infarction: an operational healthy dietary score. Eur J Nutr. 2002;41:153–160.

121. Trichopoulou A. Traditional Mediterranean diet and longevity in the elderly: a review. Public Health Nutr. 2004;7:943–947.

122. McCullough ML, Feskanich D, Stampfer MJ, et al. Diet quality and major chronic disease risk in men and women: moving toward improved dietary guidance. Am J Clin Nutr. 2002;76:1261–1271.

123. Kennedy E. Dietary diversity, diet quality, and body weight regulation. Nutr Rev. 2004;62(7 Pt 2):S78–S81.

124. Trichopoulos D, Lagiou P. Dietary patterns and mortality. Br J Nutr. 2001;85:133–134.

125. Hoffmann K, Schulze MB, Schienkiewitz A, Nothlings U, Boeing H. Application of a new statistical method to derive dietary patterns in nutritional epidemiology. Am J Epidemiol. 2004;159:935–944.

126. Hoffmann K, Zyriax BC, Boeing H, Windler E. A dietary pattern derived to explain biomarker variation is strongly associated with the risk of coronary artery disease. Am J Clin Nutr. 2004;80:633–640.

127. Schulze MB, Hoffmann K. Methodological approaches to study dietary patterns in relation to risk of coronary heart disease and stroke. Br J Nutr. 2006;95:860–869.

(p.118) 128. Weikert C, Hoffmann K, Dierkes J, et al. A homocysteine metabolism-related dietary pattern and the risk of coronary heart disease in two independent German study populations. J Nutr. 2005;135:1981–1988.

129. Schulze MB, Hoffmann K, Manson JE, et al. Dietary pattern, inflammation, and incidence of type 2 diabetes in women. Am J Clin Nutr. 2005;82:675–684.

130. Third National Health and Nutrition Examination Survey (NHANES III) Public-Use Data Files. U.S. Department of Health and Human Services, Centers for Disease Control and Prevention, National Center for Health Statistics, 2004.

131. Bingham SA, Luben R, Welch A, Wareham N, Khaw KT, Day N. Are imprecise methods obscuring a relation between fat and breast cancer? Lancet. 2003;362:212–214.