Nutrition in the Prevention and Treatment of Disease, 3rd

Edition

Visit the link below to download the full version of this book:

https://medidownload.com/product/nutrition-in-the-prevention-and-treatment-of-di
sease-3rd-edition/

Click Download Now

8 1. DIETARY ASSESSMENT METHODOLOGY

occasions not originally reported (e.g., snacks and bev- relatively little burden on the respondents, those who
erage breaks). However, interviewers should be pro- agree to give 24-hour dietary recalls are more likely to
vided with standardized neutral probing questions so be representative of the population than are those who
as to avoid leading the respondent to specific answers agree to keep food records. Thus, the 24-hour recall
when the respondent really does not know or method is useful across a wide range of populations. In
remember. addition, interviewers can be trained to capture the
The current state-of-the-art 24-hour dietary recall detail necessary so that new foods reported can be
instrument is the U.S. Department of Agriculture’s researched later by the coding staff and coded appro-
(USDA) Automated Multiple-Pass Method (AMPM) priately. Finally, in contrast to record methods, dietary
[78,79], which is used in the U.S. National Health and recalls occur after the food has been consumed, so
Nutrition Examination Survey (NHANES), this coun- there is less potential for the assessment method to
try’s only nationally representative dietary survey. In interfere with dietary behavior.
the AMPM, intake is reviewed more than once in an Computerized data collection software systems are
effort to retrieve forgotten eating occasions and foods. currently available in most developed countries, allow-
It consists of (1) an initial “quick list,” in which the ing direct coding of most foods reported during the
respondent reports all the foods and beverages con- interview. This is highly efficient with respect to pro-
sumed, without interruption from the interviewer; (2) a cessing dietary data, minimizing missing data, and
forgotten foods list of nine food categories commonly standardizing interviews [82,83]. If direct coding of the
omitted in 24-hour recall reporting; (3) time and occa- interview is done, methods for the interviewer to easily
sion, in which the time each eating occasion began and enter those foods not found in the system should be
what the respondent would call it are reported; (4) a available, and these methods should be reinforced by
detail pass, in which probing questions ask for more interviewer training and quality control procedures.
detailed information about the food and the portion Another technological advance in 24-hour
size, in addition to review of the eating occasions and dietary recall methodology is the development of auto-
times between the eating occasions; and (5) final mated self-administered data collection systems
review, in which any other item not already reported is [72,74 76,84 88]. These systems vary in the number of
asked [78,79]. In addition, research at USDA allowed foods in their databases, the approach to asking about
development of the Food Model Booklet [80], a portion portion size, and their inclusion of probes regarding
size booklet used in the NHANES in order to facilitate details of foods consumed and possible additions. The
more accurate portion size estimation. A 24-hour recall web-based Automated Self-Administered 24-hour die-
interview using the multiple-pass approach typically tary recall (ASA) developed at the National Cancer
requires between 30 and 45 minutes. Institute (NCI) [72,87,88] allows respondents to com-
A quality control system to minimize error and plete a dietary recall with the aid of multimedia visual
increase reliability of interviewing and coding 24-hour cues, prompts, and an animated character versus stan-
recalls is essential. Such a system should include a dard methods that require a trained interviewer. The
detailed protocol for administration, training, and system uses the most current USDA survey database
retraining sessions for interviewers; duplicate collec- [89] and includes many elements of the AMPM 24-hour
tion and coding of some of the recalls throughout the interview developed by USDA [78] and currently used
study period; and the use of a computerized database in the NHANES. Portion sizes are asked using digital
system for nutrient analysis. One study evaluated the photographs depicting up to eight sizes [88]. The instru-
marginal gains in accuracy of the estimates of mean ment is freely available for use by researchers, clini-
and variance with increasing levels of quality control cians, and educators. Such web-based tools allow
[81], and the authors recommended that the extent of researchers to economically collect high-quality dietary
quality control procedures adopted for a particular data in large-scale nutrition research. One study
study should be carefully considered in light of that indicates that differences between interviewer- and
study’s desired accuracy and precision and its resource self-administered recalls are minimal among adoles-
constraints. cents [76]. Other studies are underway to evaluate
There are many advantages to the 24-hour recall. differences between interviewer- and web-based
When an interviewer administers the tool and records self-administered recalls.
the responses, literacy of the respondent is not The main weakness of the 24-hour recall approach is
required. However, for self-administered versions, lit- that individuals may not report their food consumption
eracy can be a constraint. Because of the immediacy of accurately for various reasons related to knowledge,
the recall period, respondents are generally able to memory, and the interview situation. These cognitive
recall most of their dietary intake. Because there is influences are discussed in more detail in Section V.A.

A. ASSESSMENT METHODS FOR RESEARCH AND PRACTICE

 II DIETARY ASSESSMENT METHODS 9
A potential limitation is that multiple days of recalls portion sizes across a wide range of food groups and
may be needed. Whereas a single 24-hour recall can be tended to report more frequent intakes of low-fat/diet
used to describe the average dietary intake of a popula- foods and less frequent intakes of fat added to foods
tion, multiple days of recalls are needed to model esti- [106]. As was found for records, factors such
mates of the population’s usual intake distributions and as obesity, gender, social desirability, restrained eat-
their relationships with other factors (see Section V.G). ing, education, literacy, perceived health status, and
The validity of the 24-hour dietary recall has been race/ethnicity have been shown in various studies
studied by comparing respondents’ reports of intake to be related to underreporting in recalls
either with intakes unobtrusively recorded/weighed [46,55,59,61,79,93,101,106 110].
by trained observers or with biological markers.
Numerous observational studies of the effectiveness of
the 24-hour recall have been conducted with children
(see Section IV.B). In some studies with adults, group
C Food Frequency
mean nutrient estimates from 24-hour recalls have The food frequency approach [111,112] asks respon-
been found to be similar to observed intakes [4,90], dents to report their usual frequency of consumption
although respondents with lower observed intakes of each food from a list of foods for a specific period.
have tended to overreport and those with higher Information is collected on frequency, but little detail is
observed intakes have tended to underreport their collected on other characteristics of the foods as eaten,
intakes [90]. One observational study found energy such as the methods of cooking, or the combinations of
underreporting during a self-selected eating period in foods in meals. Many FFQs also incorporate portion
both men and women, similar underreporting during a size questions or specify portion sizes as part of each
controlled diet period in men, and accurate reporting question. Overall nutrient intake estimates are derived
during a controlled diet period in women; underesti- by summing, over all foods, the products of the
mates of portion sizes accounted for much of the reported frequency of each food by the amount of
underreporting [91]. Studies with biological markers nutrient in a specified (or assumed) serving of that
such as doubly labeled water and urinary nitrogen food to produce an estimated daily intake of nutrients,
generally have found underreporting using 24-hour dietary constituents, and food groups. In most cases,
dietary recalls for energy in the range of 3 34% the purpose of an FFQ is to obtain a crude estimate of
[25,40,75,79,92 98], with the largest two studies in total intakes over a designated time period.
adults using a multiple-pass method showing average There are many FFQ instruments, and many continue
underreporting to be between 12 and 23% [79,95]. For to be adapted and developed for different populations
protein, underreporting tends to be in the range of and purposes. Among those evaluated and commonly
11 28% [92,95,96,98 102]. However, underreporting is used are the Block Questionnaires [113 123], the Fred
not always found. Some have found overreporting of Hutchinson Cancer Research Center Food Frequency
energy from 24-hour dietary recalls compared to dou- Questionnaire [124,125], the Harvard University Food
bly labeled water in proxy reports for young children Frequency Questionnaires or Willett Questionnaires
and adolescents [103,104]. In addition, it is likely that [111,121 123,126 133], and the NCI’s Diet History
the commonly reported phenomenon of underreport- Questionnaire [95,123,134,135], which was designed
ing in Western countries may not occur in all cultures; with an emphasis on cognitive ease for respondents
for example, Harrison et al. [105] reported that 24-hour [136 138]. Throughout the years, population-specific
recalls collected from Egyptian women were well FFQs have been developed. Examples include FFQs
within expected amounts. Finally, in many studies, designed to capture diets of Latinos, Native Americans
energy adjustment has been found to reduce error. For [139 143], African Americans [144], Hispanics [145,146],
example, for protein density (i.e., percentage energy native Hawaiians, and Asian ethnic groups living in
from protein), 24-hour dietary recalls conducted in the Hawaii [147,148]. FFQs have been developed throughout
large biomarker studies were in close agreement or the world, such as those for countries participating in
somewhat higher compared to a biomarker-based mea- the European Prospective Investigation into Cancer and
sure [55,95,96]. Nutrition (EPIC) [45,149 154] and for Australia
In past national dietary surveys using multiple-pass [155,156], Belgium [157], France [158], Germany [159],
methods, data suggested that underreporting may Norway [160], Japan [161,162], Korea [163], and the
have affected up to 15% of all 24-hour recalls United Kingdom [164]. An FFQ-like instrument, called
[106,107]. Underreporters compared to non-underre- the Oxford WebQ, has been developed for large-scale
porters tended to report fewer numbers of foods, epidemiologic research [165]. Like an FFQ, this instru-
fewer mentions of foods consumed, and smaller ment includes a comprehensive list of foods and

A. ASSESSMENT METHODS FOR RESEARCH AND PRACTICE

10 1. DIETARY ASSESSMENT METHODOLOGY

portions, but the participant is asked to report whether are several approaches to constructing such a database
the foods listed were consumed the previous day. Such [111]. One approach uses quantitative dietary intake
instruments, like 24-hour dietary recalls, are meant to be information from the target population to define the
administered at multiple time points in a study. typical nutrient density of a particular food group cate-
Evaluation of this tool showed moderate correlations gory. For example, for the food group macaroni and
(average r 5 0.6) with interviewer-administered recalls cheese, all reports of the individual food codes reported
for the same day [165]—values slightly higher than those in a population survey can be collected, and a mean
generally obtained when full-length FFQs that query or median nutrient composition (by portion size if
intake during the past year are evaluated against inter- necessary) can be estimated. Values can also be calcu-
viewer-administered recalls. “Brief” FFQs that assess a lated by gender and age. Dietary analyses software,
limited number of dietary exposures are discussed in specific to each FFQ, is then used to compute nutrient
the next section. Because of the number of FFQs avail- intakes for individual respondents. These analyses
able, investigators need to carefully consider which best are available commercially for the Block, Willett, and
suits their research needs. Hutchinson FFQs, and they are publicly available for
The appropriateness of the food list is crucial in the the NCI FFQ.
food frequency method [114]. The entire breadth of an In pursuit of improving the validity of the FFQ,
individual’s diet, which includes many different foods, investigators have addressed a variety of frequency
brands, and preparation practices, cannot be fully cap- questionnaire design issues, such as length, closed-
tured with a finite food list. Obtaining accurate reports versus open-ended response categories, portion size,
for foods eaten both as single items and in mixtures is seasonality, and time frame. Frequency instruments
particularly problematic. FFQs can ask the respondent designed to assess total diet generally list more than
either to report a combined frequency for a particular 100 individual line items, many with additional portion
food eaten both alone and in mixtures or to report sep- size questions, requiring 30 60 minutes to complete.
arate frequencies for each food use. (For example, one This raises concern about length and its effect on
could ask about beans eaten alone and in mixtures, or response rates. Although respondent burden is a factor
one could ask separate questions about refried beans, in obtaining reasonable response rates for studies in
bean soups, beans in burritos, and so on.) The first general, a few studies have shown this not to be a deci-
approach is cognitively complex for the respondent, sive factor for FFQs [137,167 171]. This tension
but the second approach may lead to double counting between length and specificity highlights the difficult
(e.g., burritos with beans may be reported as both issue of how to define a closed-ended list of foods for a
beans and as a Mexican mixture). Often, FFQs will food frequency instrument. The increasing use of
include similar foods in a single question (e.g., beef, optically scanned or web-based instruments has neces-
pork, or lamb). However, such grouping can create a sitated the use of closed-ended response categories,
cognitively complex question (e.g., for someone who forcing a loss in specificity [172].
often eats beef and occasionally eats pork and lamb). Although the amounts consumed by individuals
Differences in definitions of the food items asked may are considered an important component in estimating
also be problematic; for example, rice is judged to be a dietary intakes, it is controversial as to whether or not
vegetable by many nonacculturated Hispanics, a judg- portion size questions should be included on FFQs.
ment not shared in other race/ethnic groups [166]. Frequency has been found to be a greater contributor
Finally, when a group of foods is asked as a single than serving size to the variance in intake of most
question, assumptions about the relative frequencies of foods [173]; therefore, some prefer to use FFQs with-
intake of the foods constituting the group are made in out the additional respondent burden of reporting
the assignment of values in the nutrient database. serving sizes [111]. Others cite small improvements in
These assumptions are generally based on information the performance of FFQs that ask the respondents to
from an external study population (such as from a report a usual serving size for each food [116,118].
national survey sample) even though true eating pat- Some incorporate portion size and frequency into one
terns may differ considerably across population sub- question, asking how often a particular portion of the
groups and over time. food is consumed [111]. Although some research has
Each quantitative FFQ must be associated with a been conducted to determine the best ways to ask
database to allow estimation of nutrient intakes for an about portion size on FFQs [136,174,175], the marginal
assumed or reported portion size of each food queried. benefit of such information in a particular study may
For example, the FFQ item of macaroni and cheese depend on the study objective and population
encompasses a wide variety of different recipes with characteristics.
different nutrient composition, yet the FFQ database Another design issue is the time frame about which
must have a single nutrient composition profile. There intake is queried. Many instruments inquire about

A. ASSESSMENT METHODS FOR RESEARCH AND PRACTICE

 II DIETARY ASSESSMENT METHODS 11
usual intakes during the past year [114,127], but it is average intake for the group. Research suggests that
possible to ask about the past week or month [176], longer food frequency lists may overestimate, whereas
depending on specific research situations. Even when shorter lists may underestimate, intake of fruits and
intake during the past year is asked, some studies vegetables [187], but it is unclear whether or how this
have indicated that the season in which the question- applies to nutrients and other food groups.
naire is administered has an influence on reporting Portion size of foods consumed is difficult for
during the entire year [177,178]. respondents to evaluate and is thus problematic for all
Finally, analytical decisions are required in how assessment instruments (see Section V.D). However,
food frequency data are processed. In research applica- the inaccuracies involved in respondents attempting to
tions in which there are no automated quality checks estimate usual portion size in FFQs may be even
to ensure that all questions are asked, decisions about greater because a respondent is asked to estimate an
how to handle missing data are needed. In particular, average for foods that may have highly variable
in self-administered situations, there are usually many portion sizes across eating occasions [188].
initial frequency questions that are not answered. One Because of the error inherent in the food frequency
approach is to assign null values because some approach, it is generally considered inappropriate to
research indicates that respondents selectively omit use FFQ data to estimate quantitative parameters, such
answering questions about foods they seldom or never as the mean and variance, of a population’s usual die-
eat [179,180]. Another approach is the imputation of tary intake [127,189 193]. Although some FFQs seem
frequency values for those not providing valid to produce estimates of population average intakes
answers. Only a few studies have addressed this issue that are reasonable [155,159,189], different FFQs will
[181,182], and it is currently unclear whether imputa- perform in often unpredictable ways in different popu-
tion is an advance in FFQ analyses. lations, so the levels of nutrient intakes estimated by
Strengths of the FFQ approach are that it is inexpen- FFQs should best be regarded as only approximations
sive to administer and process and it asks about the [190]. FFQ data are usually energy adjusted and then
respondent’s usual intake of foods over an extended used for ranking subjects according to food or nutrient
period of time. Unlike other methods, the FFQ can be intake rather than for estimating absolute levels of
used to circumvent recent changes in diet (e.g., changes intake, and they are used widely in case control or
due to disease) by obtaining information about indivi- cohort studies to assess the association between dietary
duals’ diets as recalled about a prior time period. intake and disease risk [194 196]. For estimating rela-
Retrospective reports about diet nearly always use a tive risks, the degree of misclassification of subjects is
food frequency approach. Food frequency responses more important than is the quantitative scale on which
are used to rank individuals according to their usual the ranking is made [197].
consumption of nutrients, foods, or groups of foods. The definitive validity study for a food frequency-
Nearly all food frequency instruments are designed to based estimate of long-term usual diet would require
be self-administered, require 30 60 minutes to com- nonintrusive observation of the respondent’s total diet
plete depending on the instrument and the respondent, over a long time. Such studies are not possible in free-
and most are either optically scanned paper versions living populations. One early feeding study, with three
or automated to be administered electronically defined 6-week feeding cycles (in which all intakes were
[113,124,139,183 185]. Because the costs of data collec- known), showed some significant differences in known
tion and processing and the respondent burden have absolute nutrient intakes compared to the Willett FFQ
traditionally been much lower for FFQs than for multi- for several fat components, mostly in the direction of
ple diet records or recalls, FFQs have been a common underestimation by the FFQ [198]. The most practical
way to estimate usual dietary intake in large epidemio- approach to examining the concordance of food fre-
logical studies. quency responses and usual diet is to use multiple food
The major limitation of the food frequency method recalls or records over a period as an indicator of usual
is that it contains a substantial amount of measurement diet. This approach has been used in many studies
error [55,95 98,135]. Many details of dietary intake are examining various FFQs (see [199] for register of such
not measured, and the quantification of intake is not as studies). In these studies, the correlations between the
accurate as with recalls or records. Inaccuracies result methods for most foods and nutrients are in the range
from an incomplete listing of all possible foods and of 0.4 0.7. However, recalls and records cannot be con-
from errors in frequency and usual serving size estima- sidered as accurate reference instruments because they
tions. The estimation tasks required for a FFQ are com- suffer from mistakes that may be correlated with errors
plex and difficult [186]. As a result, the scale for in the FFQs, and they may not represent the time period
nutrient intake estimates from a FFQ may be shifted of interest. Biomarkers that do represent usual intake
considerably, yielding inaccurate estimates of the without bias are available for energy (doubly labeled

A. ASSESSMENT METHODS FOR RESEARCH AND PRACTICE

12 1. DIETARY ASSESSMENT METHODOLOGY

water) [200] and protein (urinary nitrogen) [201]. health education is the goal. In the intervention setting,
Validation studies of various FFQs using these biomar- brief instruments focused on specific aspects of a die-
kers have found large discrepancies with self-reported tary intervention also have been used to track changes
absolute energy intake [40,46,49,55,75,92,94 98] and in diet. However, because of concern that responses to
protein intake [44,45,55,92,95,96,98,154,202 205], usually questions of intake that directly evolve from interven-
in the direction of underreporting. Correlations of FFQs tion messages may be biased [225] and that these
and the biomarkers have ranged from 0.1 to 0.5 for instruments lack sensitivity to detect change [226],
energy [40,75,92,95,96,98] and from 0.2 to 0.5 for protein this use is not recommended. Brief instruments of
[44,45,92,95,96,98,154,202 205]. A few studies show that specific dietary components such as fruits and
correlations between a biomarker for protein density vegetables are used for population surveillance at the
constructed from both urinary nitrogen and doubly state or local level, for example, in the Centers for
labeled water and self-reported protein density on an Disease Control and Prevention’s (CDC) Behavioral
FFQ (kcal of protein as a percentage of total kcal) are Risk Factor Surveillance System (BRFSS) [227] and the
higher than correlations between urinary nitrogen and California Health Interview Survey (CHIS) [228] (see
FFQ-reported absolute protein intake [96,98,135], indi- Section III.A). Brief instruments have also been used to
cating that energy adjustment may alleviate some of the examine relationships between some specific aspects of
error inherent in food frequency instruments. Various diet and other exposures, such as in the National
statistical methods employing measurement error mod- Health Interview Survey (NHIS) [229]. Finally, some
els and energy adjustment are used not only to assess groups suggest the use of short screeners to evaluate
the validity of FFQs but also to adjust estimates of rela- the effectiveness of policy initiatives [228,230].
tive risks for disease outcomes [55,206 216]. However, Brief instruments can be simplified or targeted
analyses indicate that correlations between an FFQ and FFQs, questionnaires that focus on specific eating
a reference instrument, such as the 24-hour recall, may behaviors other than the frequency of intake of spe-
be overestimated because of correlated errors cific foods, or daily checklists. Complete FFQs typi-
[55,96,135]. Furthermore, a few analyses comparing rela- cally contain 100 or more food items to capture the
tive risk estimation from FFQs to dietary records range of foods contributing to the many different
[217,218] in prospective cohort studies indicate that nutrients in the diet. If an investigator is interested
observed relationships are severely attenuated, thereby only in estimating the intake of a single nutrient or
obscuring associations that might exist, but such find- food group, however, then far fewer foods need to be
ings are not consistent [219]. Accordingly, some epide- assessed. Often, only 15 30 foods might be required
miologists have suggested that the error in FFQs is a to account for most of the intake of a particular nutri-
serious enough problem that alternative means (e.g., ent [231,232].
food records or 24-hour recalls) of collecting dietary Numerous short questionnaires using a food fre-
data in large-scale prospective studies should be consid- quency approach have been developed and compared
ered [220 222]. It has also been suggested that FFQ with multiple days of dietary records, 24-hour recalls,
data might be combined with recall or record data to complete FFQs, and/or biological indicators of diet.
improve estimated intakes [222 224]. Single-exposure abbreviated FFQs have been developed
and tested for a wide range of nutrients and other die-
tary components. The NCI has developed a Register of
Validated Short Dietary Assessment Instruments [233],
D Brief Dietary Assessment Instruments which contains descriptive information about short
Many brief dietary assessment instruments, also instruments and their validation studies and publica-
known as “screeners,” have been developed. These tions, as well as copies of the instruments when avail-
instruments can be useful in situations that do not able. To be included, publications were required to be in
require either assessment of the total diet or quantita- English language peer-reviewed journals and published
tive accuracy in dietary estimates. For example, a brief in January 1998 or later. Currently, the register includes
diet assessment of some specific dietary components 103 instruments assessing more than 25 dietary factors.
may be used to triage large numbers of individuals Instruments from 29 different countries have been regis-
into groups to allow more focused attention on those at tered. Instruments in the register may be searched by
greatest need for intervention or education. dietary factors, questionnaire format, and number of
Measurement of dietary intake, even if imprecise, can questions. Descriptive information about the validation
also serve to activate interest in the respondent, which study includes the reference tool, the study population
in turn can facilitate nutrition education. These brief (age, sex, and race/ethnicity), and the geographical loca-
instruments may therefore have utility in clinical set- tion. Much of the focus in brief instrument development
tings or in situations in which health promotion and has been on fruits and vegetables and on fats.

A. ASSESSMENT METHODS FOR RESEARCH AND PRACTICE

 II DIETARY ASSESSMENT METHODS 13
1 Brief Instruments Assessing Fruit and table, and snacks) questionnaire, initially developed to
Vegetable Intake assess adherence to low total fat (#30% energy from
Food frequency-type instruments to measure fruit fat) and saturated fat diets [255], asks about frequency
and vegetable consumption range from including a sin- of intake and portion size of 20 individual foods that
gle overall question to 45 or more individual questions are major food sources of fat and saturated fat in the
[234 238]. An early 7-item tool developed by the NCI U.S. diet. Its initial evaluation showed high correlations
and private grantees for the NCI’s 5 A Day for Better with dietary records [255]. In additional cross-sectional
Health Program effort was used widely in the United studies, the MEDFICTS underestimated percentage
States [239 241]. This tool was similar to one used in calories from fat; it was effective in identifying very
CDC’s BRFSS [227,242,243]. Validation studies of the high-fat intakes but was not effective in identifying
BRFSS and 5 A Day brief instruments, or “screeners,” moderately high-fat diets [256] or correctly identifying
to assess fruit and vegetable intake suggested that low-fat diets [257]. The number of mixtures reported
without portion size adjustments, they often underesti- on an FFQ (e.g., pizza and macaroni and cheese),
mated actual intake [234,239,243 245]. Using cognitive which were not specifically included in the MEDFICTS
interviewing findings (see Section V.A), the NCI tool, was negatively related to its predictive ability
revised the tool, including adding portion size ques- [257]. In a longitudinal setting, positive changes in the
tions; some studies indicate improved performance MEDFICTS score have been correlated with improve-
[246] and utility in surveillance studies. However, its ments in serum lipids and waist circumference among
performance in community interventions was mixed. cardiac rehabilitation patients [258]. In a large ethni-
In six of eight site/sex comparisons, fruit and cally diverse population, MEDFICTS and the Block
vegetable consumption was significantly overestimated Health Habits and History Questionnaire, a food fre-
relative to results from multiple 24-hour recalls [247]. quency questionnaire, were poorly correlated.
More important, the screener indicated a change in Although the accuracy of MEDFICTS differed little
consumption in both men and women when none was among age or race/ethnicity groups, its specificity to
seen with the 24-hour recalls [248]. Using cognitive correctly identify nonadherence to the prescribed diet
testing and expert guidance, the CDC has developed a was significantly worse for women than for men [259].
new fruit and vegetable screener [249] that assesses Other fat screeners have been developed to preserve
intake of all forms of fruit and subgroups of the between-person variability of intake [260 262]—
vegetables that are particularly relevant to 2010 Dietary that is, to focus on the fat sources that most distinguish
Guidelines for Americans [250]. Portion size, although differences in fat intake among individuals or groups.
not asked, will be estimated from external data about A 20-item screener developed and tested at the
portion sizes reported by sex/age groups. German site of EPIC correlated with 7-day dietary
records (r 5 0.84) and a complete FFQ (r 5 0.82)
[260,261]. A 16-item percentage energy from fat
2 Brief Instruments Assessing Fat Intake screener had a correlation of 0.6 with 24-hour recalls in
A fat screener, originally developed by Block [251] an older U.S. population [262]. However, its perfor-
and currently composed of 17 items [113], was designed mance in intervention studies of adults was variable
to account for most of the intake of fat using information [263].
about sources of fat intake in the U.S. population. The fat Often, dietary fat reduction interventions are
screener was used as an initial screen for high fat intake designed to target specific food preparation or con-
in the Women’s Health Trial [251], and in the BRFSS for sumption behaviors rather than frequency of consum-
nutritional surveillance [252]. However, the screener did ing specific foods. Such behaviors might include
not perform well in Hispanic women [252]. A similar fat trimming the fat from red meats, removing the skin
screener substantially overestimated percentage energy from chicken, or choosing low-fat dairy products.
from fat and was only modestly correlated (r 5 0.36) with Many questionnaires have been developed in various
multiple 24-hour recalls in a sample of medical students populations to measure these types of dietary beha-
[253]. In samples of men participating in intervention viors [252,264 271], and many have been found to cor-
trials, the screener was not as precise [226] or as sensitive relate with fat intake estimated from other more
[254] as complete FFQs. In addition, the screener did not detailed dietary instruments [272,273] or with blood
reflect differences observed from 24-hour recalls among lipids [268,274,275]. In addition, some studies have
different demographic groups, possibly because the found that changes in dietary behavior scores have cor-
screener did not include all of the high-fat foods that con- related with changes in blood lipids [269,274,276]. The
tribute to differences in fat intake [226]. Kristal Food Habits Questionnaire, also called the
The MEDFICTS (meats, eggs, dairy, fried foods, fat Eating Behaviors Questionnaire, was originally devel-
in baked goods, convenience foods, fats added at the oped in 1990 [277]. It measures five dimensions of

A. ASSESSMENT METHODS FOR RESEARCH AND PRACTICE

14 1. DIETARY ASSESSMENT METHODOLOGY

fat-related behavior: avoid fat as a spread or flavoring, questionnaire were questions to reflect emotional eat-
substitute low-fat foods, modify meats, replace high-fat ing and impulsive snacking.
foods with fruits and vegetables, and replace high-fat Some instruments combine aspects of food fre-
foods with lower fat alternatives. The instrument has quency and behavioral questions to assess multiple
been updated and modified for use in different settings dietary patterns. For example, the Rapid Eating and
and populations [275,278,279]. A modification tested in Activity Assessment for Patients is composed of 27
African American adolescent girls had a relatively low items assessing consumption of whole grains, calcium-
correlation (r 5 0.31) with multiple 24-hour recalls rich foods, fruits and vegetables, fats, sugary beverages
[280]. In another modification developed for African and foods, sodium, and alcohol. When compared to
American women [281], a subset of 30 items from the dietary records, correlations were 0.49 with the
SisterTalk Food Habits Questionnaire correlated with Healthy Eating Index (HEI) [294], a measure of overall
change in BMI (r 5 20.35) as strongly as did the origi- diet quality, and moderately high (range of
nal 91 items (r 5 20.36) [282]. r 5 0.33 0.55) for HEI subscores of fat, saturated fat,
cholesterol, fruit, and meats. Correlations for other HEI
subscores for sodium, grains, vegetables, and dairy
3 Brief Multifactor Instruments products were low (range of r 5 0.03 0.27) [295].
Recognizing the utility of assessing a few dimen- Because the cognitive processes for answering food
sions of diet simultaneously, several multifactor short frequency-type questions can be complex, some
instruments have been developed and evaluated, often attempts have been made to reduce the respondent
combining fruits and vegetables with dietary fiber burden by creating brief instruments with questions
and/or fat components [16,283 287]. Others assess that require only “yes no” answers. Kristal et al. [296]
additional components of the diet. For example, Prime- developed another questionnaire to assess total fat, sat-
Screen is composed of 18 FFQ items asking about con- urated fat, fiber, and percentage energy from fat that is
sumption of fruits and vegetables, whole and low-fat composed of 44 food items for which respondents are
dairy products, whole grains, fish and red meat, and asked whether they eat the items at a specified fre-
sources of saturated and trans fatty acids; 7 items ask quency. A simple index based on the number of “yes”
about supplement intake. The average correlation with responses was found to correlate well with diet as mea-
estimates from a full FFQ over 18 food groups was 0.6 sured by 4-day dietary records and with FFQs asses-
and over 13 nutrients was also 0.6 [288]. The 5-Factor sing total diet [296]. This same “yes no” approach to
Screener used in the 2005 NHIS Cancer Control questioning for a food list has also been used as a mod-
Supplement assessed fruits and vegetables, fiber, ification of the 24-hour recall [297]. These “targeted”
added sugar, calcium, and dairy servings [289], and 24-hour recall instruments aim to assess particular
the dietary screener used in the 2005 CHIS assessed foods, not the whole diet [67,298 300]. They present a
fruits and vegetables and added sugars [290]. The die- precoded close-ended food list and ask whether the
tary screener administered in the 2009 2010 NHANES respondent ate each food on the previous day; portion
included 28 items addressing consumption of fruits size questions may also be asked. For example, a web-
and vegetables, whole grains, added sugars, dairy, administered checklist has been developed to measure
fiber, calcium, red meats, and processed meats [291]. the Dietary Approaches to a Stop Hypertension diet. It
This screener was also used in the 2010 NHIS Cancer includes a listing of foods grouped into 11 categories,
Control Supplement. and it includes serving size information [301].
Some multicomponent behavioral questionnaires
have also been developed. The Kristal Food Habits 4 Limitations of Brief Instruments
Questionnaire was expanded not only to measure the The brevity of these instruments and their corre-
five fat factors described previously but also to mea- spondence with dietary intake as estimated by more
sure three factors related to fiber: consumption of cer- extensive methods create a seductive option for investi-
eals and grains, consumption of fruits and vegetables, gators who would like to measure dietary intake at a
and substitution of high-fiber for low-fiber foods [292]. low cost. Although brief instruments have many appli-
This fat- and fiber-related eating behavior question- cations, they have several limitations. First, they do not
naire correlated with food frequency measures of per- capture information about the entire diet. Most mea-
centage energy from fat (0.53) and fiber (0.50) among sures are not quantitatively meaningful and, therefore,
participants from a health maintenance organization in estimates of dietary intake for the population usually
Seattle, Washington [292]. Schlundt et al. [293] devel- cannot be made. Even when measures aim to provide
oped a 51-item Eating Behavior Patterns Questionnaire estimates of total intake, the estimates are not precise
targeted at assessing fat and fiber consumption among and have large measurement error. Finally, the specific
African American women. Newly incorporated in this dietary behaviors found to correlate with dietary intake

A. ASSESSMENT METHODS FOR RESEARCH AND PRACTICE

 II DIETARY ASSESSMENT METHODS 15
in a particular population may not correlate similarly coding the information. However, many variations of
in another population or even in the same population the Burke method have been developed and used in a
at another time period. For example, behavioral ques- variety of settings [309 312,314 318]. These variations
tionnaires developed and tested in middle-class, mid- attempt to ascertain the usual eating patterns for an
dle-aged U.S. women [277] were found to perform very extended period of time, including type, frequency,
differently when applied to Canadian male manual and amount of foods consumed; many include a cross-
laborers [302], to a low-income, low-education adult check feature [319,320].
Canadian population [303], and to participants in a Some diet history instruments have been automated
worksite intervention program in Nevada [304]. and adapted for self-administration, sometimes with
Similarly, a screener developed to assess fast-food and audio, thus eliminating the need for an interviewer to
beverage consumption in a primarily white, adolescent ask the questions [31,311,321]. Other diet histories have
population [305] was not useful in an overweight been automated but still continue to be administered
Latina adolescent population [306]. Investigators by an interviewer [322,323]. Short-term recalls or
should carefully consider the needs of their study and records are often used for validation or calibration
their own population’s dietary patterns before choos- rather than as a part of the tool.
ing an “off-the-shelf” instrument designed to briefly The major strength of the diet history method is
measure either food frequency or specific dietary its assessment of meal patterns and details of food
behaviors. intake rather than intakes for a short period of time
(as in records or recalls) or only frequency of food
consumption. Details of the means of preparation of
E Diet History foods can be helpful in better characterizing nutrient
The term diet history is used in many ways. In the intake (e.g., frying vs. baking), as well as exposure
most general sense, a dietary history is any dietary to other factors in foods (e.g., charcoal broiling).
assessment that asks the respondent to report about When the information is collected separately for each
past diet. Originally, as coined by Burke, the term die- meal, analyses of the joint effects of foods eaten
tary history referred to the collection of information not together are possible (e.g., effects on iron absorption
only about the frequency of intake of various foods but of concurrent intake of tea or foods containing vita-
also about the typical makeup of meals [307,308]. Many min C). Although a meal-based approach often
now imprecisely use the term dietary history to refer to requires more time from the respondent than does a
the food frequency method of dietary assessment. food-based approach, it may provide more cognitive
However, several investigators have developed diet support for the recall process. For example, the
history instruments that provide information about respondent may be better able to report total bread
usual food intake patterns beyond simply food fre- consumption by reporting bread as consumed at
quency data [309 312]. Some of these instruments each meal.
characterize foods in much more detail than is allowed A weakness of the approach is that respondents are
in food frequency lists (e.g., preparation methods and asked to make many judgments about both the usual
foods eaten in combination), and some of these instru- foods consumed and the amounts of those foods eaten.
ments ask about foods consumed at every meal These subjective tasks may be difficult for many respon-
[311,313]. The term diet history is therefore probably dents. Burke cautioned that nutrient intakes estimated
best reserved for dietary assessment methods that are from these data should be interpreted as relative rather
designed to ascertain a person’s usual food intake in than absolute. All of these limitations are also shared
which many details about characteristics of foods as with the food frequency method. The meal-based
usually consumed are assessed in addition to the fre- approach is not useful for individuals who have no par-
quency and amount of food intake. ticular eating pattern and may be of limited use for indi-
The Burke diet history included three elements: a viduals who “graze” (i.e., eat throughout the day rather
detailed interview about usual pattern of eating, a food than at defined mealtimes). The approach, when con-
list asking for amount and frequency usually eaten, ducted by interviewers, requires trained nutrition pro-
and a 3-day dietary record [307,308]. The detailed inter- fessionals and is thus costly. Finally, the diet history as a
view (which sometimes includes a 24-hour recall) is method is not well standardized, and thus methods dif-
the central feature of the Burke dietary history, with fer from each other and are difficult to reproduce, mak-
the food frequency checklist and the 3-day diet record ing comparisons across studies difficult.
used as cross-checks of the history. The original Burke Relative to other assessment approaches, few valida-
diet history, which requires administration by an inter- tion studies of diet history questionnaires using biolog-
viewer, has not often been exactly reproduced because ical markers as a basis of comparison have been
of the effort and expertise involved in capturing and conducted. The studies found that reported mean

A. ASSESSMENT METHODS FOR RESEARCH AND PRACTICE

16 1. DIETARY ASSESSMENT METHODOLOGY

energy intakes using the diet history approach in race/ethnicity), as covariates [342]. Frequency infor-
selected small samples of adults were underestimated mation contributes to the model by providing addi-
in the range of 2 23% compared to energy expenditure tional information about an individual’s propensity to
as measured by doubly labeled water [324 327]. consume a food—information not available from only
Generally, underreporting of protein, compared to uri- a few recalls. The recalls, however, provide informa-
nary nitrogen, was less than that for energy and only tion about the nature and amount of the food con-
sometimes significantly different [325,327 329]. These sumed. A similar approach has been used in EPIC,
results have also been seen in children [330], adoles- which combined information from two non-consecu-
cents [331,332], and the elderly [310]. Because of small tive 24-hour recalls with a food propensity question-
sample sizes in these studies, few were able to examine naire to identify those who do not consume each food
characteristics related to underreporting, and their [100]. Such methods are used to better measure usual
results were mixed, with some finding more underre- intakes (see Section V.G).
porting with higher BMI [329,330] and others finding Another statistical advance is the demonstration of
no relationship [310,326,333]. Although the diet history enhanced accuracy and statistical power of combining
approach was extensively used as the main study 24-hour recall reports and biomarkers to estimate asso-
instrument in European cohorts initiated in the 1990s, ciations between diet and disease [343]. Carroll et al.
the approach is seldom used now in new cohort stud- [222] explored the number of days of 24-hour recall
ies as other approaches have evolved. The approach is required to estimate associations between diet and dis-
sometimes used as a reference instrument [334 336]. ease in a cohort study and whether an FFQ, in addi-
tion, is beneficial. They concluded that for most
nutrients and foods, 4 non-consecutive days of 24-hour
recall report is optimal. The combination of FFQ and
F Blended Instruments multiple 24-hour recalls was superior in estimating
Better understanding of various instruments’ some nutrients and foods, especially for episodically
strengths and weaknesses has led to creative blending consumed foods.
of approaches with the goal of maximizing the Developing hybrid instruments as well as develop-
strengths of each instrument. For example, a record- ing new analytical techniques that combine informa-
assisted 24-hour recall has been used in several studies tion from different assessment methods may hold great
with children [337,338]. The child keeps notes of what promise for furthering our ability to accurately assess
he or she has eaten and then uses these notes as mem- diets.
ory prompts in a later 24-hour recall. Several research- Table 1.1 summarizes the important characteristics
ers have combined elements of a 24-hour recall and of the main self-report dietary assessment methods.
FFQ, often to assess specific dietary components. For
example, in one assessment of fruits and vegetables, a
limited set of questions is asked about the previous III DIETARY ASSESSMENT IN DIFFERENT
day’s intake and the information is combined with STUDY DESIGNS
usual frequency of consumption of common fruits and
vegetables [17,339]. Similarly, the Nutritionist Five The choice of the most appropriate dietary assess-
Collection Form combines a 2-day dietary recall with ment method for a specific research question requires
food frequency questions [340]. Thompson et al. [341] careful consideration. The primary research question
combined information from a series of daily checklists must be clearly formed, and questions of secondary
(i.e., precoded record) with frequency reports from an interest should be recognized as such. Projects can fail
FFQ to form checklist-adjusted estimates of intake. In a to achieve their primary goal because of too much
validation study of this approach, validity improved attention to secondary goals. The choice of the most
for energy and protein but was unchanged for protein appropriate dietary assessment tool depends on many
density [341]. factors. Questions that must be answered in evaluating
A recent advance is the development of statistical which dietary assessment tool is most appropriate for a
methods that seek to better estimate usual intake of particular research need include the following [195]: (1)
episodically consumed foods. A two-part statistical Is information needed about foods, nutrients, other
model developed by NCI uses information from two food components, or specific dietary behaviors? (2) Is
or more 24-hour recalls, allowing for the inclusion of the focus of the research question on describing intakes
daily frequency estimates derived from a food pro- using estimates of average intake, and does it also
pensity questionnaire (a frequency questionnaire that require distributional information? (3) Is the focus of
does not ask about portion size), as well as other the research question on describing relationships
potentially contributing characteristics (e.g., age and between diet and health outcomes? (4) Is absolute or

A. ASSESSMENT METHODS FOR RESEARCH AND PRACTICE

 III DIETARY ASSESSMENT IN DIFFERENT STUDY DESIGNS 17
TABLE 1.1 Comparison of Self-Report Dietary Assessment Methods by Important Characteristics
Dietary record 24-hour recall FFQ Diet history Screener

Type of information attainable

Detailed information about foods consumed X X X

General information about food groups consumed X X

Meal-specific details X X X
Scope of information sought
Total diet X X X X
Specific components X
Time frame asked
Short term (e.g., yesterday, today) X X X
Long term (e.g., last month, last year) X X X
Adaptable for diet in distant past
Yes X X X

No X X
Cognitive requirements
Measurement or estimated recording of foods and drinks as they X
are consumed
Memory of recent consumption X X
Ability to make judgments of long-term diet X X X
Potential for reactivity
High X
Low X X X X
Time required to complete
,15 minutes X
.20 minutes X X X X
Suitable for cross-cultural comparisons without instrument adaptation
Yes X X X
No X X X

relative intake needed? (5) What level of accuracy and related to past diet. Any of the dietary instruments dis-
precision is needed? (6) What time period is of interest? cussed in this chapter can be used in cross-sectional
(7) What are the research constraints in terms of studies. Some of the instruments, such as the 24-hour
money, interview time, staff, and respondent recall, are appropriate when the study purpose
characteristics? requires quantitative estimates of intake. Others, such
as FFQs or behavioral indicators, are appropriate
when qualitative estimates are sufficient—for exam-
A Cross-Sectional Surveys ple, frequency of consuming soda and frequency of
One of the most common types of population-level eating from fast-food restaurants.
studies is the cross-sectional survey, a set of measure- When measurements are collected on a sample at
ments of a population at a particular point in time. two or more times, the data can be used for pur-
Such data can be collected solely to describe a particu- poses of monitoring dietary trends. To assess trends
lar population’s intake. Alternatively, data can be used in intakes over time, it would be ideal for the die-
for surveillance at the national, state, and local levels as tary surveillance data collection methods, sampling
the basis for assessing risk of deficiency, toxicity, and procedures, and food composition databases to be
overconsumption; to evaluate adherence to dietary similar from survey to survey. As a practical matter,
guidelines and public health programs; and to develop however, this is difficult, and the benefits of trend
food and nutrition policy. Cross-sectional data also analysis may not outweigh the benefits of
may be used for examining associations between cur- improving the methods over time. The dietary
rent diet and other factors including health. However, assessment method used consistently throughout the
caution must be applied in examining many chronic years in U.S. national dietary surveillance is the
diseases believed to be associated with past diet interviewer-administered 24-hour recall. However,
because the currently measured diet is not necessarily recall methodology has improved over time based

A. ASSESSMENT METHODS FOR RESEARCH AND PRACTICE

18 1. DIETARY ASSESSMENT METHODOLOGY

on cognitive research, the addition of multiple inter- either the recent past (e.g., the year before diagnosis) or
viewing passes, standardization of probes, automa- the distant past (e.g., 10 years ago or in childhood).
tion of the interview, and automation of the coding. Because of the need for information about diet before
Another issue that affects the assessment of trends the onset of disease, dietary assessment methods that
over time is changes in the nutrient or food grouping focus on current behavior, such as the 24-hour recall,
databases and specification of default foods. Changes are not useful in retrospective studies. The food fre-
in the food supply are reflected in additions or subtrac- quency and diet history methods are well suited for
tions to food composition databases, whereas changes assessing past diet and are therefore the only viable
in consumption trends may lead to subsequent reas- choices for case control (retrospective) studies.
signment of default codes for foods not fully specified In any food frequency or diet history interview, the
in 24-hour recalls or records (e.g., when type of milk is respondent is not asked to recall specific memories of
not specified, the default code is now 2% milk as each eating occasion but, rather, to respond on the
opposed to whole milk in the past). Food composition basis of general perceptions of how frequently he or
databases, too, are modified over time because of true she ate a food. In case control studies, the relevant
changes in food composition, improved analytic meth- period is often the year before the diagnosis of a dis-
ods for particular nutrients, or inclusion of information ease or onset of symptoms or at particular life stages,
for new dietary components. Reflecting true changes such as adolescence and childhood. Thus, in assessing
over time is especially beneficial in trend analysis. past diet, an additional requirement is to orient the
Since 1999, the major cross-sectional surveillance sur- respondent to the appropriate period.
vey in the United States has been the NHANES [344]. The validity of recalled diet from the distant past is
This survey is conducted by the National Center for difficult to assess because definitive recovery biomarker
Health Statistics. The dietary component of the survey, information (doubly labeled water or urinary nitrogen)
called “What We Eat in America” [71], consists of 24- is not available for large samples from long ago.
hour recalls collected using the USDA’s AMPM (see Instead, relative validity and long-term reproducibility
Section II.B). The USDA also processes and analyzes the of various FFQs have been assessed in various popula-
data. The 24-hour recalls in NHANES query the intake tions by asking participants from past dietary studies to
of dietary supplements as well as foods and beverages. recall their diet from that earlier time [347,348]. These
In NHANES 2003 2004, 2005 2006, 2007 2008, and studies have found that correlations between past and
2009 2010, two 24-hour dietary recalls were conducted, current reports about the past vary by nutrient and by
allowing for estimation not only of average usual intake food group [111,349], with higher correspondence for
but also of the distributions of usual intake of the die- very frequently consumed and rarely consumed foods
tary components (see Section V.G). compared to that for foods consumed moderately often
NHANES provides high-quality dietary intake data [349,350]. Evidence suggests that correspondence
at the national level, but these data are of limited use between past and recalled past decreases with the
for state and local researchers planning and evaluating length of time between reports [347]. In particular, ret-
their programs and policies [345]. Collection of state rospective reports of diet in adolescence after long
and local data is often constrained by lack of resources recall periods (i.e., .30 years) have shown little corre-
or interview time, leading to the frequent use of less spondence with the original reports [351]. Maternal
expensive brief instruments. For example, the CDC has reports about diets of their children in early childhood
used telephone-administered screeners to periodically or adolescence have also shown low correspondence
assess fruit and vegetable intake within the BRFSS with the original reports [352,353].
[249]. The California Department of Public Health, in Correspondence of retrospective diet reports with the
its California Dietary Practices Survey, has assessed diet as measured in the original study has usually been
dietary practices among adults biennially since 1989 greater than the correspondence of current diet with past
[346]. The California Health Interview Survey used diet. This observation implies that if diet from years in the
telephone-administered screeners to assess fruit and past is of interest, it is usually preferable to ask respon-
vegetable intake in 2001, 2005, and 2009 [228]. dents to recall it than to simply consider current diet as a
proxy for past diet. Nonetheless, the current diets of
respondents may affect their retrospective reports about
past diets. In particular, retrospective diet reports from
B Case Control (Retrospective) Studies seriously ill individuals may be biased by recent dietary
A case control study design classifies individuals changes [347,354]. Studies of groups in whom diet was
with regard to current disease status (as cases or con- previously measured indicate no consistent differences in
trols) and relates this to past (retrospective) exposures. the accuracy of retrospective reporting between those
For dietary exposure, the period of interest could be who recently became ill and others [355,356].

A. ASSESSMENT METHODS FOR RESEARCH AND PRACTICE

 III DIETARY ASSESSMENT IN DIFFERENT STUDY DESIGNS 19

C Cohort (Prospective) Studies size categories. Because the diets of individuals

change over time, it is desirable to measure diet
In a cohort study design, exposures of interest are throughout the follow-up period rather than just at
assessed at baseline in a group (cohort) of people and baseline. If diet is measured repeatedly over years,
disease outcomes occurring over time (prospectively) repeated calibration is also desirable. Information
are then related to the baseline exposure levels. In pro- from calibration studies can be used for three
spective dietary studies, dietary status at baseline is purposes: to give design information, such as the
measured and related to later incidence of disease. For sample size needed [197]; to relate values from the
many chronic diseases, large numbers of individuals food frequency tool (or a brief food list thus derived)
need to be followed for years before enough new cases to values from the recalls/records [214]; and to
with that disease accrue for statistical analyses. A determine the degree of attenuation/measurement
broad assessment of diet is usually desirable in pro- error in the estimates of association observed in
spective studies because many dietary exposures and the study (e.g., between diet and disease)
many disease end points will ultimately be investi- [209,212,214,216,363 367] (see Section V.B).
gated and areas of interest may not even be recognized
at the beginning of a cohort study.
In order to relate diet at baseline to the eventual
occurrence of disease, a measure of the usual intake of D Intervention Studies
foods (see Section V.G) by study subjects is needed. Intervention studies range from relatively small,
Although a single 24-hour recall or a dietary record highly controlled, clinical studies of targeted partici-
for a single day would not adequately characterize the pants to large trials of population groups. Intervention
usual diet of individual study subjects in a cohort studies may use dietary assessment for two purposes:
study, such information could be later analyzed at the (1) initial screening for inclusion (or exclusion) into
group level for contrasting the average dietary intakes the study and (2) measurement of dietary changes
of subsequent cases with those who did not acquire resulting from the intervention. Not all intervention
the disease. Typically, researchers are interested in trials require initial screening. For those that do,
estimating additional parameters, and therefore, 24- screening can be performed using very detailed instru-
hour recalls or dietary records, if used, may require ments or less burdensome instruments. For example,
multiple administrations. Multiple dietary recalls, food frequency instruments were used in the
multiple records, diet histories, and food frequency Women’s Health Trial [251] and in the Women’s
methods have all been used effectively in prospective Health Initiative Dietary Modification Trial [368] to
studies. Cost and logistic issues tend to favor food identify groups with high fat intake and thus deter-
frequency methods because many prospective studies mine eligibility.
require thousands of respondents. However, The need for careful planning and formative
because of concern about significant measurement research in designing useful community intervention
error and attenuation attributed to the FFQ trials has been described [369]. A critical element is the
[217,220,221,357 360], other approaches are being con- existence of evidence that a particular intervention
sidered. One approach is the use of automated self- would create a measureable change in a particular
administered 24-hour recall instruments (see group and setting. Measurement of the effects of a die-
Section II.B). Another approach is collecting multiple tary intervention requires a valid measure of change
days of dietary records at baseline, with later coding from baseline to the conclusion of the intervention
and analysis of records for those respondents selected period. Often, post-intervention diet is also measured
for analysis, using a nested case control design to assess the durability of any dietary change. Some
[361,362]. Incorporating emerging technological work has been done to examine the validity of methods
advances in administering dietary records, such as to measure dietary change in individuals or in popula-
using mobile phones, increases the feasibility of such tions [279,370]. Researchers have found that dietary
approaches in prospective studies. records and scheduled 24-hour recalls were associated
If using an FFQ in the cohort, it is desirable to with changed eating behavior during the recording
include multiple recalls or records in subsamples of days and had less correspondence with biological mea-
the population (preferably before beginning the sures [371], expected weight change [372], and
study) to construct or modify the food frequency increased underreporting [373]. One study using die-
instrument and to calibrate it (see Section V.B). tary screeners and a reference measure of multiple
Information on the foods consumed could be used to non-consecutive unannounced 24-hour recalls found
ensure that the questionnaire includes the major food that the change in fruit and vegetable intake was over-
sources of key nutrients, with appropriate portion estimated relative to the control group [248]; however,

A. ASSESSMENT METHODS FOR RESEARCH AND PRACTICE

20 1. DIETARY ASSESSMENT METHODOLOGY

in the same study, the screener and the 24-hour recalls pricing fluctuations should be carefully considered.
were consistent in finding no change in percentage Another method to consider is measuring changes
energy from fat in the two groups [374]. Because of in biomarkers of diet, such as serum carotenoids
resource constraints, large intervention studies have [379,382] or serum cholesterol [383], in the population.
often relied on less precise measures of diet, including Consistency of changes in self-reported diet and appro-
FFQs and brief instruments. However, resource con- priate biomarkers provides further evidence for real
straints may be less relevant with the availability of changes in the diet. See Chapters 10 and 11 for more
automated self-administered 24-hour recall instru- in-depth discussions of the evaluation of diet in nutri-
ments and less burdensome dietary records. tion interventions and use of biomarkers in interven-
Intentional behavior change is a complex and tion studies respectively.
sequential phenomenon, as has been shown for tobacco Table 1.2 summarizes the dietary methods com-
cessation [375], and this is also true for dietary change monly used in different study designs.
[376]. Measurement of specific dietary behaviors in
addition to, or even in place of, dietary intake could be
considered in intervention evaluations when the nature IV DIETARY ASSESSMENT IN SPECIAL
of the intervention involves education about specific POPULATIONS
behaviors. If, for instance, a community-wide cam-
paign to choose low-fat dairy products were to be eval-
A Respondents Unable to Self-Report
uated, food selection and shopping behaviors specific
to choosing those items could be measured. The effects In many situations, respondents are unavailable or
of educational interventions might also be assessed by unable to report about their diets. For example, in
measuring knowledge, attitudes, beliefs, barriers, and case control studies, surrogate reports may be
perceptions of readiness for dietary change, although obtained for cases who have died or who are too ill to
the reliability of these types of questions has not been interview. Although the accuracy of surrogate reports
well assessed. has not been examined using the reference biomarkers
Whether an intervention is targeting individuals or of doubly labeled water or urinary nitrogen, the com-
the entire population, repeated measures of diet among parability of reports by surrogates and subjects has
study subjects can reflect reporting bias in the direction been studied with the goal that surrogate information
of the change being promoted [370]. Although not might be used interchangeably with information pro-
intending to be deceptive, some respondents may tend vided by subjects [384]. Common sense indicates that
to report what they think investigators want to hear. individuals who know most about a subject’s lifestyle
Social desirability [377] and social approval [378] biases would make the best surrogate reporters [385]. Adult
can be measured and the resulting scales incorporated siblings provide the best information about a subject’s
into intervention analyses. Because of their greater sub- early life, and spouses or children provide the best
jectivity, behavioral questions, screeners, and the food information about a subject’s adult life. When food fre-
frequency method may be more susceptible to social quency instruments are used, the level of agreement
desirability biases than the 24-hour recall method between subject and surrogate reports of diet varies
[69,225]. On the other hand, greater awareness of with the food and possibly with other variables, such
diet and enhanced reporting skills because of the inter- as number of shared meals, interview situation, case
vention may enhance the accuracy of reports [379]. status, and sex of the surrogate reporter. Mean fre-
Dietary records and scheduled 24-hour recalls are vul- quencies of use computed for individual foods and
nerable to reactivity. If assessment is by 24-hour recalls, food groups between surrogate reporters and subject
unannounced administration would avoid reactivity
but possibly at the expense of participation. Because
self-reports of diet are subject to bias in the context of TABLE 1.2 Dietary Assessment Methods Commonly Used in
an intervention study [370], an independent assess- Different Study Designs
ment of dietary change should be considered. For Study design Methods
example, food availability and/or sales in worksite
cafeterias, school cafeterias, or vending machines could Cross-sectional 24-Hour recall, FFQ, brief instruments
be monitored. One such method useful in community- Case control FFQ, diet history
wide interventions is monitoring food sales [380]. (retrospective)
Often, cooperation can be obtained from food retailers Cohort (prospective) FFQ, diet history, 24-hour recall, dietary
[381]. However, because the number of food items may record
be large, it may be possible to monitor only a small Intervention FFQ, brief instruments, 24-hour recall
number, and the large effects on sales of day-to-day

A. ASSESSMENT METHODS FOR RESEARCH AND PRACTICE

 IV DIETARY ASSESSMENT IN SPECIAL POPULATIONS 21
reporters tend to be similar [386 388], but agreement To examine the suitability of the initial database,
is much lower when detailed categories of frequency preliminary information about typical diets should be
are compared. Several studies have shown that agree- collected from individuals in the ethnic groups. This
ment is better for alcoholic beverages, coffee, and tea information could come from recalls or records with
than for foods. accompanying interviews or from focus group inter-
When subjects themselves report intakes in the views. These interviews should focus on the foods
extremes of a distribution, their surrogates seldom eaten and the ways in which foods are prepared in that
report intakes in the opposite extreme, although the culture. Recipes and alternative names of the same
surrogates tend to report intakes in the middle of the food should be collected, and field interviewers should
distribution [389]. This may limit the usefulness of sur- be familiarized with the results of these focus groups.
rogate information for individual-level analyses that Recipes and food names that are relatively uniform
rely on proper ranking. Furthermore, the quality of should be included in the nutrient composition data-
surrogate reports between spouses of deceased subjects base. Even with these modifications, it may be prefera-
and spouses of surviving subjects may differ substan- ble for the field interviewers to collect detailed
tially [390]. Thus far, however, little evidence suggests descriptions of ethnic foods reported rather than to
that dietary intakes are systematically overreported or directly code these foods using preselected lists most
underreported depending on the case status of the sub- common in computer-assisted methods. This would
ject [391 393]. Nonetheless, use of surrogate respon- prevent the detail of food choice and preparation from
dents should be minimized for obtaining dietary being lost by a priori coding.
information in analytical studies. When used, analyses USDA continues to incorporate new foods into the
excluding the surrogate reports should be done to National Nutrient Database for Standard Reference
examine the sensitivity of the reported associations to (SR) (see Section V.E). For example, approximately 200
possible errors or biases in the surrogate reports. If foods identified as Native American or Alaskan Native
planning a study using surrogate reports, the sample have been incorporated into the SR24 and also are
size should be inflated to account for a higher inci- available in the University of Minnesota Nutrient
dence of missing data, the inability to recruit surro- Database System. If a newly reported food is not avail-
gates for some number of cases, and the reduced able in the food composition database being used, a
precision of dietary estimates. default code that is thought to closely mirror the nutri-
ent composition of the new food can be used.
Use of FFQs developed for the majority population
B Ethnic Populations may be suboptimal for many individuals with ethnic
The widespread use of many ethnic foods in the eating patterns. Many members of ethnic groups con-
United States throughout the population and the sume both foods common in the mainstream culture
increasing diversity of the population have broadened and foods that are specific to their own ethnic group.
the food composition databases and food lists used for Modification of the existing food list can be accom-
the general population. Nonetheless, special modifica- plished through expert judgment, qualitative interviews
tions may be needed in the content of dietary assess- with the target population, and/or examination of the
ment methods when the study population is composed frequency of reported foods in the population from a set
of individuals whose cuisine or cooking practices are of dietary records or recalls. FFQs for Navajos [398],
not mainstream [394]. If the method requires an inter- Alaska Natives [399], Chinese Americans [400], indivi-
view, interviewers of the same ethnic or cultural back- duals in northern India [401], Hispanics [146,402],
ground are preferable so that dietary information can Israelis [403], and African Americans in the southern
be more effectively communicated. If dietary informa- United States [404] have been developed using these
tion is to be quantified into nutrient estimates, exami- approaches.
nation of the nutrient composition database is In addition to the food list, however, there are other
necessary to ascertain whether ethnic foods are important issues to consider when adapting existing
included and whether those foods and their various FFQs for use in other populations. The relative intake
preparation methods represent those consumed by the of different foods within a food group line item may
target population [395]. It is also necessary to examine differ, thus requiring a change in the nutrient database
the recipes and assumptions underlying the nutrient associated with each line item. For example, Latino
composition of certain ethnic foods. Some very differ- populations may consume more tropical fruit nectars
ent foods may be called the same name, or identical and less apple and grape juice than the general U.S. pop-
foods may be called by different names [396,397]. For ulation and therefore would require a different nutrient
these reasons, it may be necessary to obtain detailed composition standard for juices. In addition, the portion
recipe information for all ethnic mixtures reported. sizes generally used may differ [405]. For example, rice

A. ASSESSMENT METHODS FOR RESEARCH AND PRACTICE

22 1. DIETARY ASSESSMENT METHODOLOGY

may be consumed in larger quantities in Latino and method to compare with self-reported instruments
Asian populations; the amount attributed to a large por- [426,427]. As predicted from Baranowski and Domel’s
tion for the general population may be substantially model, it has been found that children’s estimates of por-
lower than the amount typically consumed by Latino tion size have large errors [428], and they are less able
and Asian populations. Adaptation of an existing FFQ than adults to estimate portion sizes [429] (see
considering all of these factors has been done for an Section V.D). Overall, the consensus seems to be that the
elderly Puerto Rican population [406], for white and characteristics of different age groups call for the use of
African American adults in the Lower Mississippi Delta different assessment approaches.
[407], and for the Hawaii Los Angeles Multiethnic For preschool-aged children, information is obtained
Cohort Study [408]. The Southern Community Cohort from surrogates, usually the primary caretaker(s), who
Study incorporated both race/ethnicity and geographic may typically be a parent or an external caregiver. If
region into its FFQ database [409]. information is obtained only from one surrogate
With some ethnic populations, it may be preferable reporter, the reports are likely to be less complete. Even
to administer an FFQ using an interviewer rather than for periods when the caregiver and child are together,
self-administration because literacy and language bar- foods tend to be underestimated [430]. A “consensus”
riers may limit participation in the study as well as recall method, in which the child and parents report as
quality of response. In addition, portion size models, a group on a 24-hour recall, has been shown to give
which interviewers can bring to a home interview, may more accurate information than a recall from either par-
be preferable to portion size pictures available in a self- ent or child alone [431]. Sobo and Rock [432] describe
administered instrument [399]. such interviews and suggest tips for interviewers to
The NCI Dietary Calibration/Validation Studies maximize data accuracy.
Register [199] can be used to search for studies using For older children, extensive research has been con-
FFQs in specific race/ethnicity groups. Performance of ducted on the 24-hour recall approach [433]. Baxter
FFQs varies across ethnic groups [410]. Questionnaires et al. [434] found that among fourth graders, 24-hour
aimed at allowing comparison of intakes across multi- recall improves as the time between reporting and eat-
ple cultures have been developed. Although some ing decreases, and meal-specific intrusions (i.e., reports
studies have found no appreciable validity differences of foods not consumed) are fewer in an open format
across various race/ethnicity groups [259], most have interview than in a time-forward format interview (i.e.,
found validity differences [145,406,408,411 413]. beginning at the earliest meal in the time period and
Understanding these validity differences is crucial to working forward to the next meal). These intrusions are
the appropriate interpretation of study results. often associated with additional intrusions at the same
meal [434]. Because accuracy of recall is greater when
the time between eating and reporting is shorter, there
will be differential error by meal; meals further away
C Children (e.g., at the beginning of the 24-hour recall period) will
Assessing the diets of children is considered to be have substantially more error [435,436].
even more challenging than assessing the diets of To make 24-hour recalls more feasible, self-adminis-
adults. Children tend to have diets that are highly vari- tered automated 24-hour recall tools have been devel-
able from day to day, and their food habits can change oped and tested for children [85]. An interviewer-
rapidly. Younger children are less able to recall, esti- administered 24-hour recall and a self-administered 24-
mate, and cooperate in usual dietary assessment proce- hour recall using the Food Intake Recording Software
dures; so much information by necessity has to be System (FIRSSt) were compared to unobtrusive obser-
obtained by surrogate reporters. Although they are vations in fourth graders. Compared to observed
more able to report, adolescents may be less interested intake, the interviewer-administered 24-hour recall was
in giving accurate reports. Baranowski and Domel associated with a 59% match, 17% intrusion, and 24%
[414] have posited a cognitive model of how children omission rates, whereas the automated recall was asso-
report dietary information. ciated with a 46% match, 24% intrusion, and 30% omis-
Dietary assessment in children and adolescents has sion rates [85]. Baranowski et al. are developing a
been discussed and reviewed [415 422]. The 24-hour second-generation version of the FIRSSt with tailored
recall, dietary records (including precoded checklists food lists and prompts [437,438].
[10]), dietary histories, FFQs, brief instruments [423 425], Other self-administered web-based tools have been
and blended instruments such as a dietary record-assisted developed for school-age children and adolescents. The
24-hour recall [337] have all been used to assess children’s Web-Span, developed in Canada, includes a dietary
intakes. The use of direct observation of children’s diets component [439,440] and has been used in school-
has also been used extensively, most often as a reference age children and adolescents in school. When two

A. ASSESSMENT METHODS FOR RESEARCH AND PRACTICE

 IV DIETARY ASSESSMENT IN SPECIAL POPULATIONS 23
non-consecutive days of Web-SPAN were compared accurate. The choice of which instrument to use may
with 3-day records, reported energy intake was signifi- depend on the study objectives and study design fac-
cantly higher on the Web-SPAN than on the records; tors, all of which will influence the appropriateness
correlations between the two methods ranged from and feasibility of different approaches [449].
0.24 to 0.40 [440]. The Synchronized Nutrition and
Activity Program (SNAP), a web-based program,
directs children to report the previous day’s food
intake by ticking the number of times they consumed
D Elderly
each of 40 foods and nine drinks. However, compared Measuring diets among the elderly can, but does not
to a 24-hour recall for the same period, SNAP generally necessarily, present special challenges [450 452]. Both
underestimated counts of the foods assessed, and the recall and food frequency techniques are inappropriate
accuracy decreased with increasing counts [441]. if memory or cognitive functioning is impaired.
New technology has also been incorporated into Similarly, self-administered tools may be inappropriate
other dietary assessment approaches. Williamson et al. if physical disabilities such as poor vision are present.
[442] developed and tested digital photography in Interviewer administration is difficult when hearing
school cafeterias. This observation method consists of problems are present [452]. Direct observation in insti-
standardized photography of the food selected before tutional care facilities [450] or shelf inventories for
the meal and the plate waste following the meal. Using elders who live at home can be useful. Even when cog-
reference portions of measured quantities of the foods, nitive integrity is not impaired, several factors can
expert judgment is used to estimate the amount of each affect the assessment of diet among the elderly.
food consumed [442]. Because of the frequency of chronic illness in this age
Another approach that has been taken with school- group, it is more probable that special diets (e.g., low
age children is a blended instrument, the record sodium, low fat, and high fiber) would have been
assisted 24-hour recall, in which the children record recommended. Such recommendations could not only
only the names of foods and beverages consumed affect actual dietary intake but also bias reporting
throughout a 24-hour period. This information serves because individuals may report what they should eat
as a cue for the later 24-hour recall interview. The rather than what they do eat. Alternatively, respon-
European Food Consumption Validation Project, a con- dents on special diets may be more aware of their diets
sortium of 13 institutes from 11 European countries, and may more accurately report them. When dentition
provisionally recommended a similar approach—a is poor, the interviewer should probe regarding foods
food recording booklet for foods eaten away from that are prepared or consumed in different ways.
home—for schoolchildren 7 14 years old. Studies Relative to other age groups, the elderly are more apt
examining the validity of this approach have had to take multiple types of nutritional supplements [453],
mixed results [337,338,443]. which present special problems in dietary assessment
Adaptation of food frequency instruments originally (see Chapter 2). Because of the concern of malnutrition
developed for adults requires consideration of the among the elderly, specific instruments to detect risk
instrument itself (food list, question wording and for- of malnutrition, such as the Nutrition Screening
mat, and portion size categories) and the database for Initiative in the United States [454,455], the Australian
converting responses to nutrient intakes. Food fre- Nutrition Screening Initiative [456], the Mini
quency instruments, some web administered, have Nutritional Assessment [457], the Geriatric Nutritional
been developed and tested for use in child and adoles- Risk Index [458], and the Simplified Nutritional
cent populations [7,184,444 446]. A web-based food Appetite Questionnaire [459], have been developed.
behavioral questionnaire underestimated the intake of Some researchers have suggested that the short-term
middle-school children compared to a multiple-pass memory required for the 24-hour recall may be more
24-hour recall [447]. Generally, correlations between difficult for the elderly, who are more adept at long-
food frequency type instruments and more precise ref- term memory [450]. However, interviewers conducting
erence instruments have been lower in child and ado- a FFQ among elderly respondents noted difficulty in
lescent populations than in adult populations. New maintaining interest and concentration, whereas these
technology-based methods, such as disposable cam- issues were not found during the more engaging
eras, mobile phones with cameras [28], and smart 24-hour recall interview [451].
phones, are being developed for collecting records and Validation studies using doubly labeled water and/
may be particularly useful among adolescents, who or urinary biomarkers among the elderly are limited
prefer these methods to traditional methods [448]. [40,460 462]. Generally, energy underreporting has
For school-age children and adolescents, there is no been found to be positively related to elevated BMI
consensus of which dietary assessment method is most and lower education, similar to younger populations.

A. ASSESSMENT METHODS FOR RESEARCH AND PRACTICE