PERSPECTIVE
                 Understanding Nutritional Epidemiology and Its
                 Role in Policy1,2
                                     3,4                 3                       3–5                       3–5
                 Ambika Satija,          Edward Yu, Walter C Willett,                 andFrankBHu *
                 3Department of Nutrition and 4Department of Epidemiology, Harvard School of Public Health, Boston, MA; and 5Channing
                 Division of Network Medicine, Brigham and Women’s Hospital, Boston, MA
                   ABSTRACT
                 Nutritional epidemiology has recently been criticized on several fronts, including the inability to measure diet accurately, and for its reliance
                 onobservational studies to address etiologic questions. In addition, several recent meta-analyses with serious methodologic flaws have arrived
                 at erroneous or misleading conclusions, reigniting controversy over formerly settled debates. All of this has raised questions regarding the
                 ability of nutritional epidemiologic studies to inform policy. These criticisms, to a large degree, stem from a misunderstanding of the
                 methodologic issues of the field and the inappropriate use of the drug trial paradigm in nutrition research. The exposure of interest in
                 nutritional epidemiology is human diet, which is a complex system of interacting components that cumulatively affect health. Consequently,
                 nutritional epidemiology constantly faces a unique set of challenges and continually develops specific methodologies to address these.
                 Misunderstandingtheseissues canleadtothenonconstructive andsometimesnaivecriticismsweseetoday.Thisarticle aimstoclarifycommon
                 misunderstandings of nutritional epidemiology, address challenges to the field, and discuss the utility of nutritional science in guiding policy
                 by focusing on 5 broad questions commonly asked of the field. Adv Nutr 2015;6:5–18.
                 Keywords:       dietary assessment, food policy, meta-analysis, nutritional epidemiology, randomized controlled trials, prospective cohort studies
                 Introduction                                                                          also dangerous because they can be perceived as misleading
                 Epidemiologyhaslonghaditsshareofskeptics,withTaubes’                                  messages, or can lead to the communication of misleading
                 1995 article being the most well-known (1). However, more                             messagestothepublicbypopularmediaandtheconsequent
                 recent commentaries have attacked nutritional epidemiol-                              adoption of unhealthy practices by the population at large.
                 ogy on several fronts. Ioannidis (2) criticizes the observa-                          For instance, after the publication of the latter meta-analysis,
                 tional nature of epidemiologic studies and small trials,                              New York Times columnist Mark Bittman told his readers
                 stating that “definitive solutions won’t come from another                            that they “can go back to eating butter” (6).
                 million observational papers or small randomized trials.”                                 Manyauthorshavesuggestedthat large randomized con-
                                                                                                                                  6
                 He refers to an article by Archer et al. (3), which calls into                        trolled trials (RCTs) are the only solution to circumventing
                 question the validity of data from the NHANES and suggests                            the problems in observational research. In reality, RCTs are
                 that “the ability to estimate population trends in caloric                            far from being the panacea in the study of diet and chronic
                 intake and generate empirically supported public policy rel-                          disease, and the results of such trials can be misleading. A
                 evant to diet-health relations from US nutritional sur-                               key reason is that the exposure of interest in nutritional
                 veillance is extremely limited.” Furthermore, questionably                            epidemiology—dietary intake—is complex, with interac-
                 designed and executed meta-analyses have disseminated                                 tions and synergies across different dietary components,
                 conflicting messages about nutrition and health, such as                              which can be difficult to study with use of a linear drug trial
                 the conclusion that being overweight lowers the risk of all-                          approach. A complex behavioral exposure such as diet also
                 cause mortality (4) and that replacing saturated fat with                             makes other aspects important in pharmacologic RCTs,
                 polyunsaturated fats has no substantial impact on cardiovas-                          such as high compliance and blinding, difficult and infea-
                 cular risk (5). Such conclusions are not only confusing but                           sible in most dietary intervention trials. Consequently,
                 1Supported by NIH grants HL60712, DK58845, P30 DK46200, and U54CA155626.              6 Abbreviations used: CVD, cardiovascular disease; DLW, doubly labeled water; IHD, ischemic
                 2Author disclosures: A Satija, E Yu, WC Willett, and FB Hu, no conflicts of interest.  heart disease; NTD, neural tube defect; RCT, randomized controlled trial; SSB,
                 *To whom correspondence should be addressed. E-mail: nhbfh@channing.harvard.edu.       sugar-sweetened beverage; TFA, trans FA; WHI, Women’s Health Initiative.
                 ã2015 American Society for Nutrition. Adv. Nutr. 6: 5–18, 2015; doi:10.3945/an.114.007492.                                                                            5
              nutritional epidemiology has design and analysis issues            tables. The nutrient content of a food varies with season, lo-
              unique to the field, and understanding the details of nutri-       cation of production, growing conditions, storage, process-
              tional epidemiologic studies requires a deep knowledge of          ing, and cooking techniques, and many of these factors are
              nutritional science and its methodologic background.               unaccounted for in food composition tables. The degree to
                 The purpose of this article is to clarify common misun-         which this is problematic differs from nutrient to nutrient.
              derstandings of nutritional epidemiology, address the chal-        Although for some nutrients, such as dietary FAs, it is rea-
              lenges to the field, and discuss the utility of nutritional         sonable to assume that these variations do not substantially
              science in guiding policy. In particular, we address 5 broad       affect calculated intakes, for others, such as selenium, the
              questions that have been commonly raised about nutritional         variation can result in calculated intakes that are substan-
              epidemiologic studies.                                             tially different from true intakes (7). In general, however,
                                                                                 this source of error does not substantially compromise the
              Can We Reliably Measure Dietary Intakes in                         ability to rank individuals with respect to nutrient intake
              Individuals and Populations?                                       so as to evaluate associations with health outcomes (7,
              Measuring diet in free-living populations is challenging be-       13). Nevertheless, estimating nutrient composition from
              cause individual diets are complex exposures with innumer-         food intake data is a challenge, especially given the changing
              able and sometimes poorly characterized components that            food landscape, and it is crucial that we continue to improve
              are consumed in varying amounts and combinations by dif-           the accuracy of food composition databases.
              ferent individuals. Dietary variables are rarely dichotomous;         When participants provide biological specimens, re-
              often, but not always, the entire population is “exposed” to       searchers can additionally measure intake by assaying bio-
              some degree. Diet is also a time-varying exposure, with in-        markers. Examples of biomarkers include doubly labeled
              dividual dietary habits and food composition changing              water (DLW) (for total energy intake), urinary nitrogen
              over time. It is not surprising, then, that most dietary assess-   (for protein intake), 24-h urinary sodium and potassium,
              ment methods have a component of error, which could be             blood lipid profiles, serum and plasma folate, and selenium
              randomday-to-day,diurnal,andseasonalvariationinanin-               and other trace minerals in toenails. Biomarkers allow for
              dividual’s diet over time, or because of systematic mecha-         objective measurement of intake without any bias because
              nisms, such as omission of foods when collecting data.             of self-reporting. The limitations of biomarkers, however,
              Nonetheless, several techniques have been developed to as-         have prevented their wider use. In particular, many foods
              certain dietary intake from free-living populations, and           and nutrients lack sensitive or specific biomarkers, their
              these methods have shown good validity with use of multi-          assessment always includes error from multiple sources,
              ple criteria. Although each assessment method comes with           they maynotbeindicatorsofindividuallong-termintake,
              its own set of limitations, strengths unique to each method        andobtainingandtestingfor biomarkersisexpensiveand
              make it appropriate for use in specific applications (7–10).       burdensome.Thus,useofbiomarkerstoinvestigatenutrient-
                 Multiple-week diet records, which require participants to       disease relations has been mostly confined to nested case-
              record everything they eat or drink over the course of several     control studies and small trials. Biomarkers are also useful
              weeks, are regarded as the gold standard for ascertaining di-      in assessing the validity of less-expensive, self-reported as-
              etary information because, unlike other methods, they do           sessments of diet, such as FFQs.
              not rely on memory. The high participant burden and cost              AnFFQconsistsofastructuredfood list and a frequency
              of keeping diet records has limited their use in large-scale       response section on which the participant indicates his/her
              epidemiologic studies; however, their ability to accurately        usual frequency of intake of each food over a certain period
              ascertain detailed dietary information makes them useful           of time in the past, usually 1 y. This is the most common
              in validation studies of other diet assessment methods,            choice for measuring intake in large observational studies
              and in monitoring compliance in trials. Another limitation         owing to its ease of use, low participant burden, and ability
              of diet records is that the process of recording can change        to capture usual long-term dietary intake. These features
              an individual’s diet, rendering the data atypical of usual in-     makepossible repeated assessments over time, which is im-
              take, although estimated intakes from diet records have been       portant to capture longer term variation in diets. Table 1
              found to correlate reasonably well with those from multiple        presents a comparative summary of the advantages, disad-
              24-h recalls (11). Repeated 24-h recalls involve a respondent      vantages, and applications of these dietary assessment
              reporting all foods consumed in the previous 24 h or calen-        methods.
              dardaytoatrainedinterviewer inpersonorover thephone.                  Thus, a collection of diverse diet assessment methods is
              Although reliance on the participant’s memory leaves room          available; their appropriate application, alone or in combi-
              for measurement error, a skilled interviewer can produce           nation, allows for a reasonably comprehensive assessment
              highly detailed and useful nutritional data comparable to a        of the diet of free-living populations. Nevertheless, recent
              diet record (11, 12). This method has been widely employed         critiques of these dietary assessment methods have called
              in dietary intervention trials. It is also used in national sur-   into question their utility in examining diet-disease relations
              veys to monitor trends in nutritional intake.                      and informing policy. A recent example is the article by Ar-
                 Apotential source of error common to these methods is           cher et al. (3), which criticizes the use of 24-h dietary recall
              in the estimation of nutrients with use of food composition        data periodically collected in the NHANES. Archer et al.
              6 Satija et al.
                 TABLE 1 Comparison of diet assessment methods
                                      Several day/week
                                          diet records            Multiple 24-h recalls        Asingle 24-h recall             Validated FFQ                  Biomarkers
                 Advantages         Provides accurate, de-       Provides fairly accurate,    Provides detailed,           Provides time-inte-          Provides an objective
                                       tailed, open-ended           detailed, open-             open-ended data on           grated data that rep-        assessment of intake.
                                       data on dietary in-          ended data on die-          dietary intake, with-        resents usual long-          Represents bioavail-
                                       take, with no reliance       tary intake, without        out reliance on long-        term intake. Can as-         able dose, which is
                                       on memory, and di-           reliance on long-           term memory.                 sess past dietary            relevant when it is
                                       rect computation of          term memory.                                             intake.                      used in etiologic
                                       portion sizes.                                                                                                     analyses.
                                    Errors from omission,        Has lower respondent         Has lower respondent         Theleast expensive and       May be available in ret-
                                       portion size estima-         burden and is less          burden and is less           most easily adminis-         rospect (analysis of
                                       tion, and recall are         expensive than diet         expensive than diet          tered diet assess-           stored specimens).
                                       least likely.                records, and works          records and multiple         ment method, with
                                                                    well in low-literacy        recalls; works in low-       the lowest respon-
                                                                    contexts.                   literacy contexts.           dent burden.
                 Disadvantages      Needs literate, moti-        Thereis scope for short-     Thereis scope for short-     There is scope for long-     Biomarker may not be
                                       vated participants;          term recall error,          term recall error,           term recall error.           sensitive to intake,
                                       participant burden is        omissions, and errors       omissions, and errors        Omissions possible           may have low speci-
                                       very high whendone           in portion size             in portion size esti-        because of fixed-            ficity, may not be
                                       over several days.           estimation.                 mation. Has high             food list.                   time-integrated, may
                                       Could also alter usual                                   random within-per-           FFQs need to be cul-         not represent usual
                                       eating habits.                                           son error.                   ture- and population-        long-termintake,and
                                                                                                                             specific.                    is subject to labora-
                                                                                                                                                          tory errors and other
                                                                                                                                                          sources of bias.
                                    Expensive and re-            Has high interviewer         Has high interviewer         Semi-quantitative.           Expensive and more in-
                                       source-intensive diet        burden and is more          burden and is more         Potential for errors in        vasive. Biomarkers
                                       assessment method.           expensive than a            expensive than FFQs.         nutrient estimation          are not available for
                                    Potential for errors in         single recall and         Potential for errors in        from food composi-           many nutrients.
                                       nutrient estimation          FFQs.                       nutrient estimation          tion tables.
                                       from food composi-        Potential for errors in        from food composi-
                                       tion tables.                 nutrient estimation         tion tables.
                                                                    from food composi-
                                                                    tion tables.
                 Applications       Validation of other diet     Validation of other diet     National surveillance of     Association analyses in      Validation of other diet
                                       assessment methods.          assessmentmethods.          mean population              large epidemiologic          assessmentmethods.
                                                                                                intake.                      studies.
                                    Monitoring compliance        Monitoring compliance        Assessment of trends in      Assessing past dietary       Association analyses in
                                       in dietary interven-         in dietary interven-        dietary intake (earlier      intake.                      epidemiologic stud-
                                       tion trials.                 tion trials.                NHANES).                                                  ies and monitoring
                                                                    Assessment of trends                                                                  compliance in inter-
                                                                    in dietary intake                                                                     vention trials
                                                                    (current NHANES).
                 comparedreported energy intake as assessed by the 24-h re-                          individuals, a single recall, as was used by Archer et al. in
                 calls with expected basal metabolic rate and concluded that                         their analysis, will tend to capture extremes of dietary intake
                 recalled energy intake data were implausibly low and recom-                         as opposed to usual current intake, increasing the likelihood
                 mended that NHANES data be eliminated in considering                                that any individual’s single recall will be implausibly high or
                 public policy. This finding represents the danger of misun-                          low. This random variation adds noise to the data, overesti-
                 derstanding methodologic issues and making inferences                               mating the variance, and flattening the distribution, thereby
                 with use of faulty logic. A recent article by Hébert et al.                         increasing the numbers of individuals in the extremes of the
                 (13) comprehensively refutes the conclusions drawn from                             distribution. Thus, repeated 24-h recalls on nonconsecutive
                 this study. The following section discusses key points from                         days are recommended to reduce within-person error. More
                 this article while providing an overview of measurement er-                         epidemiologic studies that use 24-h recalls to assess diet now
                 ror assessment and correction in nutritional epidemiology.                          obtain multiple replicate measures on each participant, and
                     Nutritional epidemiologyhasadvancedconsiderablyover                             starting in 2002, a second 24-h recall was introduced in the
                 the last 50 y with respect to understanding types and sources                       NHANEStoaddress some of these issues (8).
                 of measurement error in dietary intake data (7, 14). An                                 However, as noted earlier, error in diet assessment need
                 insufficient appreciation of this can lead to erroneous con-                         not be completely random. Systematic sources of variation
                 clusions like those of Archer et al. (3). Because of the con-                       include omission of foods consumed by individuals, errors
                 siderable day-to-day variation in dietary intake among                              in estimating portion sizes, and over- or under-reporting
                                                                                                                                    Understanding nutritional epidemiology          7
              becauseofsocialapproval(respondincertainwaystogetso-              widened with time. These data underscore the importance
              cial praise) or social desirability (respond in certain ways      of developing nutritional policies to improve diet quality
              avoid social criticism) (15–17). All of these could have led      and reduce health disparities.
              to the under-reporting of energy intake observed by Archer           Because of its low cost and low participant burden, self-
              et al. The underestimation of energy intake from self-            administered computer-processed FFQs are the only option
              reported data has long been known to nutrition researchers,       in most large cohort studies to assess usual dietary intakes.
              andmanystrides in methodology have been made to reduce            FFQs usually have lower random within-person variation
              this source of measurement error (7, 18–22). Under-reporting      than other dietary assessment methods because they are de-
              because of omission or portion size estimation errors is un-      signed to assess average usual intake over the past year. For
              likely to be differential with respect to determinants of the     this reason, they are better equipped to assess long-term di-
              outcome of interest. In addition, there have been improve-        etary intake, the exposure of etiologic interest for most dis-
              ments in 24-h recall methodology that reduce these sources        eases (7). Because of their reliance on memory, FFQs may
              of error, such as the USDA 5-pass method (8), which is            suffer from greater measurement error relative to recalls
              structuredtominimizeomissionoffoodsandtohelppartic-               and records if these methods are used for many days to re-
              ipants report accurate portion sizes by using visual aids. This   flect longer-term intakes (for certain nutrients, just a few
              5-pass method, which was introduced into the NHANES               days of diet records or recalls might be enough, provided
              starting in 2002, was found to agree reasonably well with ac-     the days are spread out over the entire reference period of
              tual intake assessed by direct observation (r = 0.57, P < 0.05)   the FFQ). Nevertheless, FFQs have been shown to have ac-
              (23), as well as with energy intake as assessed by the DLW        ceptable validity when compared to reference measures
              technique (r = 0.32 for males; r = 0.25 for females), which       (29, 30), with typical correlation coefficients for individual
              is considered the gold standard for energy intake assessment      nutrients or foods ranging from 0.4 to 0.7 (7). Adjustment
              (provided body compartment masses such as fat mass and            for total energy intake, along with use of repeated FFQs in
              water mass are stable over time because DLW is a measure          long-term prospective cohort studies, further improves
              of energy expenditure), although the DLWtechnique has er-         these validity coefficients. Although extended dietary rec-
              rors of its own (8). Another solution is to use an isocaloric     ords are the most popular reference method, when bio-
              statistical model in analysis, i.e., adjust for total energy in-  markers are available, triangulation methods can be used
              take. In analytic epidemiology we are generally less inter-       to obtain improved estimates of correlations of FFQ intake
              ested in the association of absolute energy/macronutrient         with true intake (31). These validity coefficients can be
              intake with health outcomes, and more with how dietary            used to correct for measurement error in epidemiologic
              composition relates to risk of disease because this is what       analyses, and the application of these measurement error
              is most modifiable by individuals or populations. Hence, ad-      correction methods is increasingly being extended to more
              justing for energy intake is standard practice in nutritional     complicated analyses (18–20). These techniques have al-
              epidemiology. Adjusting for energy intake also diminishes         lowed for valid inferences to be drawn from large cohort
              extraneous sources of variation in dietary intake, and to         studies with use of FFQ data.
              some extent also reduces systematic sources of under- and            Despite these developments in reducing measurement er-
              over-reporting (7, 21, 22). Issues related to measurement         ror in dietary intake data, continued improvementsindietary
              error are not isolated to dietary assessment methods, but         assessment methodology and measurement error correction
              extend to assessment of most behavioral exposures and bio-        are needed to advance the field. Nevertheless, the consider-
              markers, including physical activity (24, 25), with which Ar-     able progress made over the past few decades, especially the
              cher and colleagues, having worked considerably in the area       use of repeated measures of diet over time, has enabled nu-
              of physical activity epidemiology, are familiar (26, 27).         tritional epidemiologists to reliably collect and use dietary in-
                 Hence, although there is no perfect method, there is am-       formation in both individuals and populations.
              ple evidence that dietary measurements in national surveys
              have reasonably good reliability and validity. The conclusion     WhatIstheRoleofNutritionalEpidemiologyin
              of Archer et al. that dietary data with use of these methods      Inferring Causality?
              cannot support public policy is misleading. National survey       Oneofthemaincriticisms leveled against nutritional epide-
              data such as the NHANES represent a small fraction of the         miology is that it relies predominantly on observational
              totality of evidence on the basis of which national guidelines    data, which is deemed to be inferior to experimental data
              and public policy are made. A main purpose of using na-           in determining causality. Figure 1 illustrates the typical hier-
              tional survey data is to assess average population intakes        archy of evidence from various study designs. While ran-
              andtrends. For example, we recently examined trends in di-        domized trials with hard endpoints occupy the highest
              etary quality from 1999 to 2010 in the US adult population        position in this hierarchy, they are usually not the most ap-
              among a nationally representative sample of 29,124 adults         propriate or feasible study design to answer nutritional epi-
              aged 20–85 y with use of the NHANES data (28). We found           demiologic questions regarding long-term effects of specific
              that better dietary quality, measured by the Alternate            foods or nutrients (unless they can be packaged in a pill).
              Healthy Eating Index, was associated with higher socioeco-           In the absence of evidence from large RCTs on hard
              nomic status, and the gap between the rich and the poor           endpoints, nutritional epidemiologists typically rely on
              8 Satija et al.