You are on page 1of 7

from the association OF PROFESSIONAL INTEREST

Dietary Patterns: Challenges and Opportunities in


Dietary Patterns Research
An Experimental Biology Workshop, April 1, 2006
Suzen M. Moeller, PhD; Jill Reedy, PhD, MPH, RD; Amy E. Millen, PhD; L. Beth Dixon, PhD, MPH;
P. K. Newby, ScD, MPH, MS; Katherine L. Tucker, PhD; Susan M. Krebs-Smith, PhD, RD;
Patricia M. Guenther, PhD, RD

A
workshop titled “Dietary Pat- ments made on similar topics by two dietary patterns, or combinations of
terns: Challenges and Opportu- or more speakers are summarized in foods and nutrients, in relation to
nities in Dietary Patterns Re- the text under the most appropriate chronic disease. These combinations
search” was held at Experimental subheading. of foods, nutrients, or in some cases,
Biology, April 1, 2006, to address the foods and nutrients, are often in-
current advantages and limitations of tended to represent the total diet or
the various approaches to studying HISTORICAL PERSPECTIVE AND key factors of the diet.
dietary patterns, as well as to con- EVOLUTION OF DIETARY PATTERNS Nutritional epidemiologists cite
sider future research that may be nec- RESEARCH several reasons for examining dietary
essary to determine the best uses of In studies of diet and chronic disease, patterns (1-6). People eat meals with
the different approaches. This article the traditional approach has been to complex combinations of nutrients
summarizes the five presentations at examine associations between indi- that are likely to be interactive or
this workshop, which are outlined in vidual nutrients or food groups with synergistic. Many nutrients are
Figure 1. The presentation topics are risk factors or outcomes related to highly correlated, making it difficult
generally used to structure this chronic conditions. More recently, nu- to examine their separate effects. The
manuscript, although relevant state- tritional epidemiologists have studied effect of a single nutrient may be too
small to detect, but the cumulative
effects of multiple nutrients included
S. M. Moeller is a scientist, Nutrition and Healthy Lifestyles, American in a dietary pattern may be suffi-
Medical Association, Chicago, IL; at the time of the study she was an as- ciently large to detect. Analyses of a
sistant scientist, Department of Ophthalmology and Visual Sciences and large number of nutrients or food
Department of Nutritional Sciences, University of Wisconsin-Madison. J. groups may produce statistically sig-
Reedy is a nutritionist, Risk Factor Monitoring and Methods Branch, Ap- nificant associations simply by
plied Research Program, Division of Cancer Control and Population Sci- chance, and are criticized for using
ences, National Cancer Institute, Bethesda, MD. A. E. Millen is an assis- too many tests. Analyses of single nu-
tant professor, Department of Social and Preventive Medicine, School of trients may be confounded by the ef-
Public Health and Health Professions, University of Buffalo, NY. L. B. fect of dietary patterns. Results from
Dixon is an associate professor and director, Public Health Nutrition, De- clinical trials have shown positive
partment of Nutrition, Food Studies, and Public Health, New York Uni- health outcomes with changes in sev-
versity, NY. P. K. Newby is an assistant professor, Department of Pediat- eral types of dietary behaviors (7,8).
rics, Boston University School of Medicine and Department of Many methods are available for
Epidemiology, Boston University School of Public Health, Boston, MA. characterizing dietary patterns in a
K. L. Tucker is a director, Epidemiology and Dietary Assessment Pro- population. Dietary patterns have
gram, and senior scientist, Jean Mayer USDA Human Nutrition Research been assessed a priori using score-
Center on Aging, Tufts University, Boston, MA. S. M. Krebs-Smith is based approaches, a posteriori using
chief, Risk Factor Monitoring and Methods Branch, Applied Research data-driven techniques such as factor
Program, Division of Cancer Control and Population Sciences, National analysis and cluster analysis, and
Cancer Institute, Bethesda, MD. P. M. Guenther is a nutritionist, Center more recently, using reduced rank re-
for Nutrition Policy and Promotion, US Department of Agriculture, Alex-
gression (1-5). Occasionally, research-
andria, VA.
ers have combined these approaches
Address correspondence to: Suzen M. Moeller, PhD, Nutrition and
(9). These methods often show associ-
Healthy Lifestyles, American Medical Association, 515 North State St,
ations with different disease out-
Chicago, IL 60610. E-mail: suzen.moeller@ama-assn.org
comes that sometimes are stronger
0002-8223/07/10707-0011$32.00/0
than those seen in studies of individ-
doi: 10.1016/j.jada.2007.03.014
ual nutrients (10).

© 2007 by the American Dietetic Association Journal of the AMERICAN DIETETIC ASSOCIATION 1233
OF PROFESSIONAL INTEREST

Organizers

Suzen M. Moeller Chair


Jill Reedy Co-chair
Amy E. Millen Co-chair

Presenters Title of presentation

L. Beth Dixon Historical perspective and evolution of dietary patterns research


Patricia M. Guenther Index-based approaches: USDA’sa revised Healthy Eating Index
P. K. Newby Data driven approaches: cluster and factor analysis
Katherine L. Tucker Constructing variables and the application of dietary patterns
Susan M. Krebs-Smith New ways of thinking about dietary patterns
a
USDA⫽United States Department of Agriculture.

Figure 1. Presentations at the 2006 Experimental Biology workshop on “Dietary Patterns: Challenges and Opportunities in Dietary Patterns
Research.”

No one method of dietary pattern intake as a way to evaluate overall are other food-group patterning
analysis is regarded as better than all dietary adequacy, and to counsel or scores based on the five major food
of the others; there is little consensus educate patients (12,13). Nutrient groups: fruits, vegetables, grain, dairy,
on which approach should be applied density scores were created to evalu- and meat (18,19).
for any given purpose. The reproduc- ate the dietary quality of individual Index-based summary scores have
ibility and validity of dietary patterns foods in terms of nutrient content in been created based on interpretation
have been called into question and relation to total energy, but do not of current dietary guidance. Two well-
may depend on the nature of the diet evaluate a total dietary pattern (14). known index scores are the Diet Qual-
assessment measure (food frequency Dietary variety has been encour- ity Index (DQI and DQI-Revised)
questionnaire [FFQ], 24-hour recalls, aged since the inception of the Nutri- (20,21) and the Healthy Eating Index
diet records) and the subjective deci- tion and Your Health: Dietary Guide- (HEI) (22,23). The DQI-Revised and
sions made by researchers during the lines for Americans in 1980 (15). the HEI are 100-point scores of 10
analyses or during the development Krebs-Smith and colleagues (16) cre- components each, reflecting recom-
of an index. Two recent review arti- ated scores to examine variety among mendations in the 1992 Food Guide
cles highlight the diversity of dietary food groups (defined as the mean Pyramid (24) and the Dietary Guide-
patterns research and the current number of different food groups con- lines. The HEI is currently under re-
need for greater attention to method- sumed daily) and variety within food vision (25) to reflect the key diet-re-
ological issues (2,3). groups (defined as the mean number lated recommendations found in the
of different food subgroups or individ- 2005 Dietary Guidelines for Ameri-
SCORE-BASED APPROACHES ual foods within food groups con- cans (26) and to improve the psycho-
Score-based approaches are based on sumed daily) and found that variety metric properties of the index.
dietary recommendations as well as among food groups explained as much Additional scores have been cre-
other knowledge about the subject of the variation in mean adequacy ra- ated based on dietary recommenda-
matter. These approaches generally tio scores as did variety within food tions such as the Recommended
fall into four categories: nutrient ad- groups. However, neither type of va- Foods Score (27), the Not Recom-
equacy or density scores, variety or riety was associated with the varia- mended Foods Score (28), the Medi-
diversity scores, food-group pattern- tion in intakes of energy, fat, sugar, terranean Diet Score (29,30), and the
ing scores, and index-based summary sodium, or cholesterol. Kant and col- Alternate Healthy Eating Index
scores. leagues (17) assessed dietary variety (AHEI). Harvard researchers created
Nutrient adequacy scores include among food groups and showed a pos- the AHEI based on findings from the
the nutrient adequacy ratio, defined itive association between intake of Nurses’ Health and Health Profes-
as average daily intake of a nutrient two or fewer food groups in 1 day sionals Follow-up studies (31). The
divided by the age- and sex-specific (ie, less diverse diets) and increased 90-point AHEI was more strongly as-
recommended intake of that nutrient, mortality. In contrast, increased vari- sociated with reduced risk of cardio-
and mean adequacy ratios, defined as ation within all food groups (defined vascular disease in men and women
the sum of the nutrient adequacy ra- as the number of different food types compared with the HEI. This is not
tio divided by the number of nutrients within each of 10 food groups) was surprising because the HEI was de-
having a nutrient adequacy ratio. De- positively associated with energy in- veloped to reflect general dietary rec-
veloped to evaluate the overall di- take; increased variation within cer- ommendations and the components of
etary quality of food assistance pro- tain food groups (eg, sweets, snacks, the AHEI were selected based on find-
grams (11), these scores also have condiments, entrees, and carbohy- ings from previous analyses of diet
been used to test validity of simple drates) was positively associated with and cardiovascular disease.
dietary scores that assess food-group body fatness (17). In addition, there Studies of dietary patterns using

1234 July 2007 Volume 107 Number 7


OF PROFESSIONAL INTEREST

Method Strengths Limitations Research considerations

Score-based methods Characterize total diet Those that dichotomize Dietary scores are
a priori components (eg, met vs not multidimensional in design, but
Intuitively appealing met) do not take into account if a summary score is used the
Types include: nutrient the full range of amounts of end product is one number that
adequacy or density Analytically simple to foods consumed. may provide little information
scores, variety or compute about the contributing factors.
diversity scores, food- Those that include a range of Therefore, most researchers
group patterning scores, Easily reproducible and points for each component do report and discuss results from
index-based summary comparable consider variability in intake of the individual components as
scores (eg, Diet Quality foods but not amounts at the well as the total dietary score.
Index, Healthy Eating Results can be meaningful, extremes.
Index, Alternate Healthy interpretable, and For hypothesis testing, a profile
Eating Index, associated with health Dependent on the selected score can be examined related
Recommended Foods outcomes underlying dietary guidelines, to some health outcome with a
Score, Not which are generally not specific basic regression equation that
Recommended Foods to one type of disease. includes multiple components.
Score, Mediterranean
Diet Score) Subjectivity is introduced in the
interpretation of the guidelines
and construction of the scores
(which foods are selected for
inclusion).

The summation of equally


weighted dietary component
scores implies that each
component is equally important
and additively related to health.

Data-driven methods Characterize total diet Limited data on the reproducibility Most critical is a better
a posteriori and validity of the methods. understanding of the statistical
Allow for biologic interactions methods and options used to
Types include: factor among nutrients Few rigorous statistical tests have derive patterns, and the use of
analysis, principal been used to examine the statistical testing to guide
components analysis, Results can be meaningful, validity of derived solutions. pattern selection.
cluster analysis interpretable, associated
with health outcomes, and Subjectivity is introduced at A better understanding of how to
show some reproducibility various points in the procedure: model pattern variables in
across populations ● Grouping of dietary items epidemiologic studies; gene-diet
● Treatment of input variables (eg, interactions using patterns; and
Patterns can be a starting grams, servings, percent of impact of measurement error on
point for modeling different energy, standardization) patterns.
types of interactions ● Analytic choices and options
because they describe (eg, statistical algorithms, use Priorities for future research
eating behavior of rotation) include continued clarification of
● Selecting a final pattern solution the most useful categorization
Factor analysis: describes the and treatment of input variables;
variation in food intake in an improved understanding of
the population based upon cultural variation in food
correlations between patterns and the need to
dietary items; a continuous understand food groupings in
variable these contexts; and the effects
of change in dietary patterns
Cluster analysis: separates over time.
individuals into mutually
exclusive groups based
upon dietary intakes; a
categorical variable

Figure 2. Strengths, limitations, and research considerations for score-based methods and data-driven methods in dietary patterns research.

July 2007 ● Journal of the AMERICAN DIETETIC ASSOCIATION 1235


OF PROFESSIONAL INTEREST

score-based approaches have gener- are not mutually exclusive: individu- method does not. The methods can be
ally shown more varied and healthful als receive factor scores for each de- used together to better understand
diets to be associated with higher in- rived factor. Factors are continuous the number of clusters present in the
come, higher education, higher en- variables that often are categorized data.
ergy intake, nutrient adequacy, higher into quantiles. The many challenges in performing
concentrations of serum nutrients, A challenging decision in factor (or cluster analysis include choosing a
higher bone mineral density, reduced cluster) analysis is selecting the solu- similarity measure and selecting
all-cause and cardiovascular disease tion, ie, how many factors (or dietary which clustering procedure is best.
mortality, and in some cases, reduced patterns) are actually meaningful in These choices can impact the cluster
incidence of and mortality from cer- the data? A number of criteria can be solution and should be considered
tain cancers (1-5). Strengths, limita- used, including scree plots, the Kaiser carefully. Many of the procedures are
tions, and future research consider- criterion (eigenvalues ⬎1), the mag- sensitive to outliers, and many re-
ations of score-based approaches are nitude of factor loadings, examining searchers standardize their data be-
listed in Figure 2. multiple solutions, and, most com- fore entering variables into the anal-
monly, observing the general inter- ysis, although standardization can
pretability of the solution. Additional dilute differences in the patterns (37).
DATA-DRIVEN APPROACHES: FACTOR criteria include using prior knowl- As with factor analysis, the re-
AND CLUSTER ANALYSIS edge or theory to guide solution searcher must select the optimal
Methods choices, using goodness-of-fit statis- cluster solution, usually based on
Data-driven approaches in dietary tics, or confirming an exploratory fac- the interpretability of the clusters.
pattern analyses use mathematics to tor analysis using confirmatory factor Additional criteria can be used to
empirically derive eating behavior analysis. guide cluster selection, including a
patterns using dietary data collected In contrast to exploratory factor dendrogram (Ward’s method only), a
from FFQs, 24-hour recalls, or diet analysis, confirmatory factor analysis scree plot, the variance explained by
records. In factor and cluster analy- involves specifying both the number the clusters, and various statistical
ses, a larger set of dietary variables of factors and the types of variables tests (eg, cubic cluster criterion).
are aggregated and reduced to form a that will load on each factor, allowing Cluster means also can be compared
smaller set of variables. These meth- the researcher to use prior knowledge with the full sample means or among
ods do not require a prior theory of about the subject matter. The re- random split samples. Although pat-
what patterns are expected, although searcher then builds the factor model terning methods are empirical, and
a thorough understanding of the lit- and “confirms” the factor structure should be driven by the data, prior
erature on eating patterns, in con- and loadings for each variable. It may theory also may guide the investiga-
junction with a theoretical framework be used either on its own or to vali- tor. For example, a comprehensive lit-
about eating behavior, can be used date an exploratory factor analysis erature review of articles using factor
with either method. Nonetheless, it is procedure. Few studies have used and cluster analysis has shown that
critical that the researcher “let the confirmatory factor analysis (32-36), “Healthy,” “Traditional,” “Sweets,”
data do the talking.” and solutions from confirmatory fac- and “Alcohol” patterns are fairly re-
The most common factor analysis tor analysis have been found to be producible across populations (2),
method used in nutritional epidemiol- highly correlated with solutions from thus a researcher might expect to see
ogy is principal components analysis, exploratory factor analysis (r⬎0.90) variants of these patterns in his or
a form of exploratory factor analysis (32,35). her own study. Indeed, humans have
(2). Principal components analysis Factor analysis derives patterns a biological predisposition for sweet
does not assume an underlying model based on the relationships between foods, further supporting the proba-
of the factors and uses matrix algebra the input variables (ie, the foods or bility of extracting a “Sweets” type of
to identify the principal components food groups). In contrast, cluster pattern, which often includes such
in the data based on a correlation or analysis derives patterns based on foods as baked goods, dairy desserts,
covariance matrix of the input vari- differences in intakes among individ- and chocolate. Thus, researchers can
ables. The resulting components, or uals. Several different types of clus- use their knowledge of the literature
factors, are linear combinations of the tering methods are available. Clus- as well as biology when approaching a
observed variables that explain the tering methods used in nutritional cluster or factor analysis.
variance in the data. The factors can epidemiology separate individuals Before beginning an analysis, re-
be rotated to improve interpretabil- into mutually exclusive, nonoverlap- searchers must decide whether or not
ity; orthogonal rotation is commonly ping groups. Individuals belong to one to derive patterns separately for pop-
used. Output from the principal com- cluster only, and clusters can then be ulation subgroups. Very few articles
ponents analysis includes factor load- used as categorical (nominal) vari- have examined this issue with regard
ings (or scoring coefficients) for each ables in research. Two clustering to age, race/ethnicity, culture, and/or
variable, which can be interpreted as methods are most common: (a) geography. It is prudent in mixed
correlation coefficients. Factor scores K-means, an optimization technique, samples to begin by deriving patterns
are calculated for each individual by and (b) Ward’s method, a hierarchical separately for subgroups and to sub-
summing the standardized input agglomerative clustering technique. sequently use statistical testing and a
variables, weighted by their factor The first method requires the re- priori knowledge of pattern– disease
loadings. Factor scores often are un- searcher to prespecify the number of interactions to decide whether or not
correlated with each other. Factors clusters in the analysis; the second to combine subgroups.

1236 July 2007 Volume 107 Number 7


OF PROFESSIONAL INTEREST

Construction of Variables and full description of the decision Validation studies include three
As described above, decisions must be process. Rotation of differing num- studies comparing solutions derived
made at several points, including the bers of factors, for example, can pro- using an FFQ with those using diet
form of the food-group input variables duce different results and must be records (35,38,39) and one study com-
(as grams, number of servings, per- carefully considered. Because there paring solutions between a 24-hour
cent of energy; standardized or not); are numerous points at which deci- recall and 48-hour recall; reasonable
the number of clusters or factors (al- sions must be made, clear explana- correlations were observed (2). Only
though statistics exist to assist, the tion of these decisions is important for one study examined the validity of
decision is largely driven by inter- interpretation of results. patterns measured using cluster
pretability); and the choice of analysis analysis (44).
methods (various statistical algo- Most important in establishing the
Reproducibility and Validity validity of a method is whether or not
rithms, use of rotation).
Dietary data collection methods Very few reproducibility and validity it is able to reliably predict disease.
may affect the form and utility of food studies have been performed for fac- Patterns have been related to anthro-
patterns. Random error from day-to- tor and cluster analysis. Two studies pometric indicators and obesity, type
day variability will contribute to mis- examined the reproducibility of fac- II diabetes, metabolic syndrome, mor-
classification of usual dietary pat- tors derived using an FFQ over 1 year tality, cardiovascular diseases, and
(38,39), and one study examined the many cancers, although findings have
terns with, for example, a single 24-
stability of factors over 9 years (32). not been consistent (2,3). Other
hour recall; thus, multiple days are
Correlations were reasonable but un- strengths, limitations, and future re-
needed, as they are for usual nutrient
derstandably weaker over the longer search considerations for data-driven
intake estimation. FFQs may limit
time period. No studies to date have methods are outlined in Figure 2.
food groupings, and can obscure eth-
nic or other food-choice differences if examined the reproducibility of clus-
specific foods are collapsed into a ter analysis, although many studies
have examined the internal reproduc- NEW WAYS OF THINKING ABOUT
larger group. For example, combining DIETARY PATTERNS
ibility of the solution using random
tortillas with bread may reduce the Dietary patterns are examined for
split-sample techniques.
ability to separate patterns. Individ- two reasons: (a) to characterize di-
Although patterns will never be ex-
uals consuming low-fat milk or whole- etary behavior, and (b) to explain the
actly the same across studies and
grain breads are likely to consume relationship of diet to health and
populations, because they are data
other foods different from individuals other factors. The first reason sug-
driven, both factor and cluster analy-
consuming whole milk or white gests a descriptive purpose, and the
sis show reasonable reproducibility of
bread. For this reason, care must be several major patterns. Currently, at second suggests a hypothesis test.
given to form food groups that are least 84 published studies have de- Specific questions of interest include:
meaningful for both the study popu- rived eating patterns using factor
lation and the research question. analysis. Of these, 30% of populations ● What dietary patterns are shown
The number of food groups selected observed a Healthy pattern, and 27% by the population?
as input variables also is important. observed a Traditional pattern. ● How close is the individual (or pop-
Too few groups may obscure differ- Sweets and Western patterns were ulation) to meeting recommenda-
ences in use, whereas too many foods observed in 19% of populations, tions on key aspects of diet?
may lead to odd combinations with whereas an Alcohol pattern was seen ● Who is and is not following recom-
undue influence (eg, if there are a few in 11% (2) (Newby PK, unpublished mendations?
heavy users and many nonusers of data, 2006). Notably, one of the larg- ● Do identified patterns relate to
unusual foods, or when foods that are est studies to use one standardized health?
generally eaten together are broken methodological approach (exploratory ● Which dietary patterns are most
down into detailed subgroups). In ad- factor analysis) across four cohorts in predictive of health outcomes?
dition, some type of energy adjust- four European countries found two ● What combinations of food intake
ment is important before patterning, consistent dietary patterns (40). explain the most variation in a spe-
because individuals with higher en- Similar percentages were found cific health outcome?
ergy intakes may have unduly large when examining the cluster analysis ● How does adherence to dietary rec-
effects on the patterns. This is partic- literature (Healthy, 36%; Alcohol, ommendations relate to health
ularly problematic when men and 26%; Sweets, 24%; Traditional, 20%; outcomes?
women are combined, or when the and Western, 4%) (2) (Newby PK, un-
sample includes a large age range. published data, 2006). Three studies Each of these, although stemming
The decision of whether or not to have compared solutions using both from one of the two general reasons
standardize input variables is rele- factor and cluster methods and found above, is a unique question requiring
vant to cluster analysis. Comparisons that individuals in the Healthy clus- particular methods to be answered.
have suggested that unstandardized ter had higher scores on the Healthy
variables produce more realistic clus- factor (41-43). Despite these apparent
ters because standardization can give similarities, it is important to con- Linking Questions to Methods
undue influence to minor food groups sider the research question when de- The issue of what dietary patterns
(37). The number of clusters or factors termining whether to use factor or are shown by the population is per-
selected requires careful judgment cluster analysis. haps the one most frequently consid-

July 2007 ● Journal of the AMERICAN DIETETIC ASSOCIATION 1237


OF PROFESSIONAL INTEREST

ered, and both cluster and factor those with a very good diet (high Final Thoughts
analysis have been used to address it. score) from those with a poor diet (low There is an adage that states, “When
The relative merits of these two meth- score), and each component is met or all you have is a hammer, everything
ods have been described above. How- not met to the same degree. Midrange starts to look like a nail.” So too, in
ever, when deciding between the two, summary scores only indicate that dietary patterns research, we may in-
one should understand that they ap- the overall diet is neither optimum advertently limit ourselves to those
proach a problem from different an- nor poor, but give no hint regarding questions for which we think we have
gles and thus answer different ques- the individual components—whether the tools to find an answer.
tions. Cluster analysis examines they are all in the midrange, or In expanding our thinking about di-
whether or not there are groups in the whether some are completely met etary patterns, the first priority is to
population that are distinctly differ- whereas others are not. A solution to determine which questions to answer,
ent from one another, and if so, what this would be a profile score in which and the second is to ascertain
typifies their diets? Factor analysis the average score for each component whether any of the commonly used
explores (or confirms) whether there is examined in relationship to its methods will answer those questions.
are underlying patterns that explain maximum points, to identify which If not, we may need to turn to other
variation in how people eat. aspects of the diet are closest to their disciplines. For example, theories ex-
Regardless of which method is targets and which need improvement. ist to explain chaos and complexity.
used, once patterns are identified, Mathematics and computer science
their relationship to health or other have provided ways of ascertaining
outcomes can be explored. This often patterns for applications in both the
is the ultimate purpose of identifying Most important in biological and the physical sciences.
the dietary patterns. However, one Operations research has been used to
limitation of both of these methods is establishing the solve problems in decision making
that they are not designed to derive validity of a method and optimization. Human genome re-
dietary patterns that are predictive of search offers an opportunity to study
disease. For this reason, reduced rank is whether or not it complex patterns using data mining
regression has been proposed as an is able to reliably methods such as neural networks.
alternative (45). These are promising areas for further
Whereas factor analysis deter- predict disease. inquiry. The field of dietary pattern
mines linear functions of predictors research will be unnecessarily re-
by maximizing the explained varia- For hypothesis testing, there are stricted if we only continue to use the
tion of all predictor variables (that is, similar concerns. A summary score same methods with different data
the food groups), reduced rank regres- may be helpful when comparing the sets and explore only those questions
sion identifies linear functions of pre- lowest quintile to the highest, as is we now know how to explore.
dictors that explain as much response common in epidemiologic research,
variation (that is, variation in a set of because in those cases, all of the com-
intermediary health markers) as pos- ponents are varying together. Yet References
sible. The resulting factors can then again, subtleties in the midrange are 1. Michels KB, Schulze MB. Can dietary pat-
be examined in relationship to health lost—an important consideration be- terns help us detect diet-disease associa-
outcome. cause most people do not have either tions? Nutr Res Rev. 2005;18:241-248.
2. Newby PK, Tucker KL. Empirically derived
Other questions, such as how close a very good or a very poor diet. An- eating patterns using factor or cluster anal-
is the individual (or population) to other approach would be to examine a ysis: A review. Nutr Rev. 2004;62:177-203.
meeting recommendations for impor- multidimensional profile score in re- 3. Kant AK. Dietary patterns and health out-
tant aspects of diet, and who is and is lationship to some health outcome comes. J Am Diet Assoc. 2004;104:615-635.
4. Hu FB. Dietary pattern analysis: A new di-
not following recommendations, often with a basic regression equation for rection in nutritional epidemiology. Curr
are addressed by diet quality indices multiple independent variables (Xs). Opin Lipidol. 2002;13:3-9.
that are made up of various compo- The dependent variable could be 5. Dixon LB, Cronin FJ, Krebs-Smith SM. Let
nents. If the various components of an some health outcome, such as blood the pyramid guide your food choices: Captur-
ing the total diet concept. J Nutr. 2001;131:
index are unrelated to one another, pressure, each X could be a compo- 461S-472S.
then they do not collectively repre- nent of the profile score (or food 6. Jacobs DR Jr, Steffen LM. Nutrients, foods,
sent a single underlying construct. In group), and the beta would be the and dietary patterns as exposures in re-
other words, diet is multidimen- weight determined for each compo- search: A framework for food synergy. Am J
Clin Nutr. 2003;78(suppl):508S-513S.
sional, and the various aspects of diet nent. From this, one could determine 7. de Lorgeril M, Salen P, Martin JL, Monjaud
captured by the different components the effect and relative importance of I, Delaye J, Mamelle N. Mediterranean diet,
may have separate, important effects each component, and the combined ef- traditional risk factors, and the rate of car-
on health outcomes. This multidi- fect of them all, on explaining varia- diovascular complications after myocardial
infarction: Final report of the Lyon Diet
mensionality suggests the need for tion in the health outcome. However, Heart Study. Circulation. 1999;99:779-785.
caution in interpreting a single sum- without interaction terms, such a 8. Appel LJ, Moore TJ, Obarzanek E, Vollmer
mary score for any purpose. model assumes that these factors act WM, Svetkey LP, Sacks FM, Bray GA, Vogt
For descriptive purposes, a sum- independently rather than synergisti- TM, Cutler JA, Windhauser MM, Lin PH,
Karanja N. A clinical trial of the effects of
mary score is instructive only when cally and would not provide insight dietary patterns on blood pressure. DASH
scores are very high or very low. In into how they act together to influ- Collaborative Research Group. N Engl
those cases, it accurately separates ence health. J Med. 1997;336:1117-1124.

1238 July 2007 Volume 107 Number 7


OF PROFESSIONAL INTEREST

9. Wright ME, Mayne ST, Stolzenberg-So- riculture and Health and Human Services; tional cohort: The National Health Inter-
lomon RZ, Li Z, Pietinen P, Taylor PR, Vir- 2005. Home and Garden Bulletin No. 232. view Surveys, 1987 and 1992. J Nutr. 2004;
tamo J, Albanes D. Development of a com- 27. Kant AK, Schatzkin A, Graubard BI, 134:1793-1799.
prehensive dietary antioxidant index and Schairer C. A prospective study of diet qual- 43. Costacou T, Bamia C, Ferrari P, Riboli E,
application to lung cancer risk in a cohort of ity and mortality in women. JAMA. 2000; Trichopoulos D, Trichopoulou A. Tracing the
male smokers. Am J Epidemiol. 2004;160: 283:2109-2115. Mediterranean diet through principal compo-
68-76. 28. Michels KB, Wolk A. A prospective study of nents and cluster analyses in the Greek pop-
10. Kant AK. Indexes of overall diet quality: A variety of healthy foods and mortality in ulation. Eur J Clin Nutr. 2003;57:1378-1385.
review. J Am Diet Assoc. 1996;96:785-791. women. Int J Epidemiol. 2002;31:847-854. 44. Quatromoni PA, Copenhafer DL, Demissie
11. Madden JP, Yoder MD. Program Evalua- 29. Bach A, Serra-Majem L, Carrasco JL, Ro- S, D’Agostino RB, O’Horo CE, Nam BH,
tion: Food Stamps and Commodity Distribu- man B, Ngo J, Bertomeu I, Obrador B. The Millen BE. The internal validity of a dietary
tion in Rural Areas of Central PA. State use of indexes evaluating the adherence to pattern analysis. The Framingham Nutri-
College, PA: Pennsylvania State University; the Mediterranean diet in epidemiological tion Studies. J Epidemiol Community
June 1972. College of Agriculture. Bulletin studies: A review. Public Health Nutr. 2006; Health. 2002;56:381-388.
780. 9:132-146. 45. Hoffmann K, Schulze MB, Schienkiewitz A,
12. Guthrie HA, Scheer JC. Validity of a dietary 30. Trichopoulou A, Costacou T, Bamia C, Nothlings U, Boeing H. Application of a new
score for assessing nutrient adequacy. J Am Trichopoulos D. Adherence to a Mediterra- statistical method to derive dietary patterns
Diet Assoc. 1981;78:240-245. nean diet and survival in a Greek popula- in nutritional epidemiology. Am J Epide-
13. Krebs-Smith SM, Clark LD. Validation of a tion. N Engl J Med. 2003;348:2599-2608. miol. 2004;159:935-944.
nutrient adequacy score for use with women 31. McCullough ML, Willett WC. Evaluating ad-
and children. J Am Diet Assoc. 1989;89:775- herence to recommended diets in adults: The
783. Alternate Healthy Eating Index. Public
14. Drewnowski A. Concept of a nutritious food: Health Nutr. 2006;9:152-157.
Toward a nutrient density score. Am J Clin 32. Newby PK, Weismayer C, Akesson A,
Nutr. 2005;82:721-732. Tucker KL, Wolk A. Long-term stability of
15. US Department of Agriculture, US Depart- food patterns identified by use of factor anal-
ment of Health and Human Services. Nutri- ysis among Swedish women. J Nutr. 2006;
tion and your health: Dietary guidelines for 136:626-633.
Americans. Washington, DC: US Govern- 33. Park SY, Murphy SP, Wilkens LR,
ment Printing Office, 1980. Yamamoto JF, Sharma S, Hankin JH, Hen-
16. Krebs-Smith SM, Smiciklas-Wright H, derson BE, Kolonel LN. Dietary patterns us-
Guthrie HA, Krebs-Smith J. The effects of ing the Food Guide Pyramid groups are as-
variety in food choices on dietary quality. sociated with sociodemographic and lifestyle
factors: The multiethnic cohort study. J
J Am Diet Assoc. 1987;87:897-903.
Nutr. 2005;135:843-849.
17. Kant AK, Schatzkin A, Harris TB, Ziegler
34. Schulze MB, Hoffmann K, Kroke A, Boeing
RG, Block G. Dietary diversity and subse-
H. An approach to construct simplified mea-
quent mortality in the First National Health
sures of dietary patterns from exploratory
and Nutrition Examination Survey Epide-
factor analysis. Br J Nutr. 2003;89:409-419.
miologic Follow-up Study. Am J Clin Nutr.
35. Togo P, Heitmann BL, Sorensen TI, Osler M.
1993;57:434-440.
Consistency of food intake factors by differ-
18. Kant AK, Schatzkin A, Block G, Ziegler RG,
ent dietary assessment methods and popu-
Nestle M. Food group intake patterns and
lation groups. Br J Nutr. 2003;90:667-678.
associated nutrient profiles of the US popu-
36. Maskarinec G, Novotny R, Tasaki K. Di-
lation. J Am Diet Assoc. 1991;91:1532-1537. etary patterns are associated with body
19. Krebs-Smith SM, Cleveland LE, Ballard- mass index in multiethnic women. J Nutr.
Barbash R, Cook DA, Kahle LL. Character- 2000;130:3068-3072.
izing food intake patterns of American 37. Wirfalt E, Mattisson I, Gullberg B, Berglund
adults. Am J Clin Nutr. 1997;65:1264S- G. Food patterns defined by cluster analysis
1268S. and their utility as dietary exposure vari-
20. Patterson RE, Haines PS, Popkin BM. Diet ables: A report from the Malmo Diet and
quality index: Capturing a multidimen- Cancer Study. Public Health Nutr. 2000;3:
sional behavior. J Am Diet Assoc. 1994;94: 159-173.
57-64. 38. Hu FB, Rimm E, Smith-Warner SA, Fes-
21. Haines PS, Siega-Riz AM, Popkin BM. The kanich D, Stampfer MJ, Ascherio A, Samp-
Diet Quality Index revised: A measurement son L, Willett WC. Reproducibility and va-
instrument for populations. J Am Diet As- lidity of dietary patterns assessed with a
soc. 1999;99:697-704. food-frequency questionnaire. Am J Clin
22. Kennedy ET, Ohls J, Carlson S, Fleming K. Nutr. 1999;69:243-249.
The Healthy Eating Index: Design and ap- 39. Khani BR, Ye W, Terry P, Wolk A. Repro-
plications. J Am Diet Assoc. 1995;95:1103- ducibility and validity of major dietary pat-
1108. terns among Swedish women assessed with
23. Bowman SA, Lino M, Gerrior SA, Basiotis a food-frequency questionnaire. J Nutr.
PP. The Healthy Eating Index: 1994-96. 2004;134:1541-1545.
Washington, DC: US Department of Agricul- 40. Balder HF, Virtanen M, Brants HAM,
ture, Center for Nutrition Policy and Promo- Krogh V, Dixon LB, Tan F, Mannisto S, Bel-
tion; 1998. CNPP-5. locco R, Pietinen P, Wolk A, Berrino F, Van
24. Cleveland LE, Cook DA, Krebs-Smith SM, den Brandt PA, Hartman AM, Goldbohm
Friday J. Method for assessing food intakes RA. Common and country-specific dietary
in terms of servings based on food guidance. patterns in four European cohort studies. J
Am J Clin Nutr. 1997;65:1254S-1263S. Nutr. 2003;133:4246-4251.
25. Healthy Eating Index. Center for Nutrition 41. Newby PK, Muller D, Tucker KL. Associa-
Policy and Promotion, United States De- tions of empirically derived eating patterns
partment of Agriculture. Available at: http:// with plasma lipid biomarkers: A comparison
www.cnpp.usda.gov/HealthyEatingIndex.htm, of factor and cluster analysis methods. Am J
2006. Accessed May 21, 2007. Clin Nutr. 2004;80:759-767.
26. Dietary Guidelines for Americans, 2005. 6th 42. Kant AK, Graubard BI, Schatzkin A. Di-
ed. Washington, DC: US Departments of Ag- etary patterns predict mortality in a na-

July 2007 ● Journal of the AMERICAN DIETETIC ASSOCIATION 1239

You might also like