1 82S Am iC/in Nuir 199 l;54:182S-7S. Printed in USA. © 1991 American Society for Clinical Nutrition

Nutrition epidemiology: how do we know what they ate?�3

Elizabeth Barreti-Connor

ABSTRACT It is generally believed but difficult to prove

that diet plays a role in the risk of various diseases. This paperreviews strengths and deficiencies of select diet-assessmentmethods used in epidemiologic studies with particular reference

to their use in the study of osteoporosis. Direct observation orweighed food records are useful primarily as validation for less

intrusive methods. Complete food history by interview or fooddiary (by self report) is expensive and time consuming. A 24-hdiet recall obtained by a trained dietitian can provide accurate,quantitative information on recent intake but does not represent

usual intake. Food frequency questionnaires provide better es-

timates of usual diet but are less quantitative and subject toproblems ofrecall and seasonality. No method is universally thebest. Lack of an expected diet-disease association may reflectexposure misclassification, inadequate statistical power, or lim-ited range ofthe nutrients studied. Given the differences in diet-

assessment methods, studies of dietary calcium and osteoporotic

fracture have had surprisingly similar results. Am J Clin Nutr

199 1;54:182S-7S.

KEY WORDS Dietary assessment, reproducibility, validity,

osteoporosis

Introduction

The association of behavior with subsequent health is of in-

creasing concern to both the scientific and the lay communities.No health-related behavior has generated more universal atten-

tion than diet because everyone eats and has, therefore, a vested

interest.

Analytic studies ofdiet and current or future disease in humanbeings are the core of nutritional epidemiology. In case-control

studies, people with and without disease are compared with re-gard to what they now say was their previous or usual diet. Incohort studies, diet is determined in a population free of disease

at baseline, and is compared with diets of people who later do

or do not develop the disease. Minor variations on these twothemes exist but all are dependent on the assessment of diet in

individuals.

In the l980s a great deal was written about the merits anddemerits of specific methods for ascertaining diet for epidemi-

ologic studies. Didactic and often unsubstantiated claims weremade about the relative value of different methods. Grants were

awarded or denied on the basis of the prejudices of reviewerswith regard to the proposed diet-assessment tool. This paperreviews selected diet-assessment methods and their strengths and

deficiencies, with particular reference to their use in the study

of diet and osteoporosis.

Individual diet-assessment methods

Clues to diet-disease associations have often been derived fromcross-cultural and geographic comparisons of food-disappearance

rates vs mortality. Associations observed in such ecological

studies must be confirmed by studies of individual diet vs in-dividual disease. The five main methods of diet assessment in

individuals used for epidemiologic research are summarized in

Table 1. All have serious flaws with regard to cost, representa-

tiveness, quality of quantitive estimates, and/or study-induced

behavior change.

Direct observation and weighed-food records

Direct observation or in-home weighed-food records are theonly methods of diet assessment applicable to free-living pop-

ulations that assure the quantitative and qualitative validity of

all nutrients consumed. Both are usually too expensive (in the

context of the sample size required) for epidemiologic studiesofdiet and disease. If more than 1 d is necessary to assess usual

diet, direct observation ofdiet is less likely to be an option than

are weighed-food records. In-home weighed-food records may

work particularly well in countries where participants are ac-

customed to recipe units given by weight rather than by measure.When direct observation is used as a “gold standard” to val-

idate reported recent intake, neither the amount nor the directionofthe error is predictable. Both over- and underestimation havebeen reported. Two studies comparing direct observation withweighed-food analysis are shown in Table 2 (1, 2). It can be seen

that, contrary to popular expectation, overweight women mayoverestimate their intake (2).

Although direct observation or weighed-food records accu-

rately represent current intake, they may not reflect usual intake.

The need to weigh and record intake may lead to a reduced-

calorie or more monotonous diet. Further, when studied in thehome, subjects know their diet is being observed, directly orindirectly, by record review. Being observed is liable to induce

behavior change. Nearly everyone in Westernized cultures hassome knowledge of how they should eat and could be temptedto have a better diet to impress the observer. Few of us would

allow anyone cloaked in the aura and authority of medical sci-

ence to monitor our feeding of hotdogs and chips to our children.

From the Department of Community and Family Medicine, Uni-versity of California, San Diego, La Jolla, CA.

2SupportedbyNlH/NIA 1 R37AG07181.3Address reprint requests to E Barrett-Connor, Department of Com-

munity and Family Medicine, University of California, San Diego,M-007, La Jolla, CA 92093-0607.

by guest on October 11, 2015

ajcn.nutrition.orgD

ownloaded from

NUTRITIONAL EPIDEMIOLOGY l83S

* Reference 5.

TABLE 1

Individual diet-assessment methods for epidemiologic research

Method Expensive Behavior change Quantitative Representative *

Observation Yes, very Yes Yes ?Diet history Yes, very No Yes YesDiet diary or record Yes Yes Semiquantitative ?24-h diet recall Yes No Yes NoFood frequency questionnaire No No Semiquantitative Yes

* Usual.

Food history

The “next-best-thing” in epidemiologic diet assessment isprobably the food history, usually practiced as a refinement of

the method described by Burke (3) in the 1940s. The quality of

food-history data was considered to be one of the reasons whyan association between dietary fat and cholesterol and coronaryheart disease could be shown within a population in the ChicagoWestern Electric study, where most other within-population

studies using less extensive dietary data have failed to show anassociation (4).

This approach, which includes a 24-h diet recall, a history of

usual foods, and data on food preparation, requires a 1-2 h

interview by a specially trained nutritionist. It is very dependent

on the quality of the interviewer and is too costly for wide ap-

plication. Concordance of other methods with the diet history

TABLE 2

Mean percentage error in reported estimate of quantity vs observed

intake

Food group* Percentage error

%

Food group*Combined main dishes -29.2Dairy products -5.7Vegetables -22.5

Fruits -2.4Salads -53.0Cereals +11.9Breads -19.9Starches -17.8Soups -51.8Desserts -30.8

Foodt

Cottage cheese +23.3Roast turkey +95.0Green beans +25.0Boiled ham +85.0Cooked spaghetti +70.0Cola drink +6.0Potato chips +260.0Blueberries + 10.0Slice of bread + 120.0Orange juice +7.5

* As reported by 86 healthy women postpartum (1).

t As reported by 30 overweight women (2).

is quite variable. As shown in Table 3 in a study by Morgan (5),

caloric intake estimated from past and current food history was

different from estimates derived from a 24-h recall or a 4 d food

record, both of which resembled each other. Nevertheless, the

food history is used as the gold standard against which other

methods not validated by observation or multiple weighed diet

records are measured.

Food record or diary

A food record or diary, usually obtained for 3-7 d, can theo-

retically avoid the costs ofan interviewer. This is a false savings,in that the food record is most accurate when participants are

trained by dietitians in how to estimate quantity and record

intakes. As shown in Table 4, intake from a 7-d food record

does not always parallel the intake based on a food history (6).One problem with the food record is compliance. Writing

down everything soon gets tedious and the characteristics (dietary

and otherwise) of people who will do so are apt to differ fromthose ofpeople who will not. One could argue that only the mostcompulsive would actually complete a 7-d diary and that con-

cordance with another diet method might be much greater in

such individuals than that for a total study population. Another

problem is a training effect, a change in food intake due to par-

ticipation. Recording all food consumed for > 1 or 2 consecutive

days is a well-known behavior-modification method to reduceintake. Snacks and condiments, high in calories, fat, and sugar,

mayjust not be worth the trouble when keeping a 7-cl diet record.Nevertheless, the food record is often used as the gold standard

for validating other methods when neither a food history nor

observation is possible.

Twenty-four-hour diet recall

The 24-h diet recall was designed to assess recent nutrientintake quantitatively (5-8). When correctly performed by a

TABLE 3Average daily joule intake of 400 women based on diet-assessmentmethod

Method Energy

kJ/d

24-h recallCurrent diet historyPast diet history

4-drecord

6760908495617451

by guest on October 11, 2015

ajcn.nutrition.orgD

ownloaded from

184S BARRETT-CONNOR

* Reference 12.

TABLE 4

Comparison of percentage differences between Burke history and24-hour recall and between 24-hour recall and 74 record*

Nutrient

Difference

Massachusetts

(n = 28)

New York

(n = 51)

Rhode Island

(n = 87)

Burke history and 24-hrecall (%)

Energy +21.1 +23.3 +9.7Protein +23.8 +20.1 -7.2

Calcium +20.6 +21.5 +0.1Phosphorus +23.8 +20.9 -3.7Iron +32.2 +17.3 -10.9

7-Day record and 24-h

recall (%)Energy +2.4 +6.5 +0.9

Protein +1.1 +1.9 +4.3Calcium +13.1 +2.5 +11.9Phosphorus +3.1 + 1.3 +8.2Iron +7.7 +1.7 +3.4

* Reference 6.

trained dietitian using food models and containers to assess

quantity, the interview takes 30-60 mm. Therefore, this method

is relatively expensive, although less so, and less interviewer-

dependent, than is the traditional diet history.

The two biggest advantages of the 24-h recall are the short-

term memory required and the quantitative estimates of food

intake. As noted by Balogh et al (7) it should be almost an axiom

that most people can remember what they ate yesterday better

than they can remember what they usually eat. It is assumed,

therefore, that the information is both quantitatively and qual-

itatively more accurate than is the remembered usual or remote

diet. Because the diet recall is usually obtained only once from

an unprepared participant, there is no training effect. (Some

have argued that a diet recall without warning is more of a psy-

chological test and that memory would be improved by notifying

the participant that details of yesterday’s diet would be asked.

Under this protocol, behavior change could not be excluded.)

The major disadvantage of the 24-h recall is the inability of

a single day’s intake to describe usual diet. This lack of repre-

sentativeness has led many investigators to conclude that a single

24-h diet recall is worthless for epidemiologic research (8-12).

Twenty-four-hour recalls are certainly not recommended to de-

tect actual deficiency states in individuals, because most vitamins

and trace minerals can vary from day to day and still be adequate

overall. Similarly, a 24-h recall is quite misleading ifone wishes

to examine a particular food, such as fish, that is not eaten daily.

Estimates of reliability for major nutrients are also discour-

aging. Both Beaton et al (8) and Liu et al (9) found that multiple

24-h recalls were required to reliably place subjects in the same

quintile of intake for some nutrients; for calcium the number

was 1 7-19 recall d. VanStaveren et al (12) used tissue biopsies

analyzed for the ratio ofpolyunsaturated to saturated fatty acids

to evaluate the extent ofdietary-fat misclassification as a function

of the number of 24-h recalls per subject. They concluded that

between 3 and 7 recalls were necessary to adequately estimate

the habitual fat intake of an individual (Table 5).

The degree of reproducibility or representativeness varies with

the study populations’ monotony of diet and with the method

used for validation. Table 4, for example, shows remarkabledifferences in the dietary intake by 24-h recall and diet history

in two geographic areas and remarkable concordance in a third.

Also, the intake of some foods may be more reproducible than

others. For example, in the Rancho Bernardo study, those who

had moderate milk intake as young adults continued to drink

milk as they grew older and people who drink milk tend to doso daily (E Barrett-Connor, unpublished observations, 1988).

Foodfrequency questionnaires

A more representative picture of diet would be expected to

be a better predictor ofchronic disease than would a single day’s

diet. The food frequency questionnaire was devised many years

ago in an attempt to obtain a self-administered, inexpensive,

and rapid estimate of usual intake (13-16). Initially, food fre-quency questionnaires were very short, with a limited number

offood items that were selected to test a single hypothesis. They

have become considerably longer for use in cohort studies, where

many diet-disease associations may be sought. Currently popularfood frequency questionnaires include well over 100 food itemsand may be self- or interviewer-administered.

The major advantages of the food frequency method are cost

and representativeness. Costs are greatly reduced because the

data are usually obtained by a self-administered questionnaire,without need for trained interviewers. Costs are further reducedwhen forms are designed to be scanned directly to computer

entry, eliminating the need for manual coding and keypunching.

As noted above, the other major advantage is that food frequency,typically asked for the past year, sometimes for more remote

years, is more representative of usual intake than a short 1- or

3-d record or recall could be expected to be. This tends to reducemisclassification by increasing the chances of correctly ranking

subjects (such as placing in quintiles by usual intake), which is

more likely to reflect a diet-disease association (17, 18).The food frequency method is not without problems, however.

The order ofthe listing is arbitrary but may influence responses.

Completion of even a short nonquantitative self-administered

questionnaire requires a certain level of literacy. If the ques-

tionnaire is very short, the limited number of food items can

address only one or two specific hypotheses, which is not efficient

for a cohort study. The need to list specific foods also tends tomake the questionnaire fairly culture specific. Food frequencyquestionnaires for Japanese-American men include mochi-gashi,duri-manju, and monaka whereas the Oxford (England) ques-

tionnaire asks about spotted dog. Because there is a limit to the

TABLE 5Estimates of probabilities of misclassification for specified number of

observations used �

Number of dietarymeasurements averaged

(n = 57)

Adjacent category

(p)

Opposite category

(q)

1 0.382 0.1843 0.316 0.132

7 0.237 0.132

by guest on October 11, 2015

ajcn.nutrition.orgD

ownloaded from

NUTRITIONAL EPIDEMIOLOGY 185S

* Reference 20.

TABLE 6

Intraclass correlation coefficients, measuring within individual

agreement of daily nutrient estimates by two different methods

in 40 young women *

Comparisonnutrient

14 vs 74record

34 vs 74record

FFQ vs 74

recordt

Energy(kJ)Protein (g)Fat (g)

Carbohydrate(g)

Calcium (mg)

0.45�0.42f

0.46�

0.58�

0.63�

0.79�0.76t

0.74�

0.90f

0.89f

0.090.02

0.04

0.190.24

* Reference 19.

t FFQ, food frequency questionnaire.�P<0.005.

number offoods that can be listed, some surprising dietary habits

may be missed. Although some of the questionnaires providespace to include diet items not included in the food list, the

completeness of such responses is unknown.

Self-administered food frequency methods are at best semi-

quantitative because only fixed or subjective definitions of small,medium, or large portions are possible. The combination ofmissing foods and semiquantitative methods limits the accuracy

of the estimated caloric intake. Because calories may be an in-

dependent risk factor and are often used in the analysis to correct

for individual variation and for exercise, the lack of accuratelyassessed calories is not inconsequential.

Perhaps the most significant problem with a food frequency

questionnaire is uncertain validity. For many nutrients of in-

terest, such as calcium, which is under homeostatic control, nobiochemical assay of serum is useful. Aside from biochemicalassays for selected vitamins and antioxidants, there is no easy

way to confirm the usual food intake of most nutrients. Con-

cordance of results based on small groups of more extensivelystudied subjects raises questions about the representativeness of

such compliant individuals(15). Comparison with other methodsprovides divergent results, and does not indicate which of these

results are correct. As shown in Table 6, from a study by Stuff

et al (19), correlation coefficients with a 7-d diet record were

better for 1- or 3-d records (obtained from the same 7-d dietrecord) than for a food frequency questionnaire.

It is important to note that reproducibility, also called reli-

ability, is not the same as validity. There is no question that the

representativeness of intake by using food frequency question-naires is higher than for the 24-h recall, but a part ofthe improvedreliability is an artifact. Because reproducibility is in part a func-

tion of the precision of the data, differences between repeated

recalls increase with decreasing simplicity of the question. For

example, reported consumption of green vegetables is expectedto vary less from day to day than is consumption of broccoli.

Similarly, any instrument that affords few or no options for por-

tion size has less variability and more reproducibility than does

a more quantitative recall.

Discussion

Diet assessment is only as good as the food-composition da-

tabank and the coders. Reports of intracoder variation of up to

30% clearly have important implications for the ability to dem-

onstrate associations between diet and disease. Computerized

coding eliminates coder variation and is an advantage of itemized

food frequency methods.

Any mention of nutrient databanks should raise the question

ofwhether nutrients or foods should be studied. Nutrients come

in foods and are eaten with other foods, which may have striking

effects on absorption and metabolism. Even when a good bio-logical mechanism supports the study ofa single nutrient, as for

calcium and osteoporosis, there are putative differences in ab-

sorption and excretion that are dependent on the calcium vehicle

and the protein, phosphate, oxalate, etc. in the diet.

Other kinds of associations, concordant and discordant, areoften ignored. The correlation for major nutrients is high in all

but extreme diet patterns (Table 7) (20). Suppose that peoplewho love ice cream have better bones. Is it the calcium or the

calories (or the resulting adiposity) that prevent osteoporosis?

In the United States it is difficult to study dietary calcium separate

from vitamin D, because the major source of calcium is milk

and this is fortified with Vitamin D. Conversely, people who

drink very little milk may drink considerably more coffee, soft

drinks, or alcohol. This could lead to the mistaken impression

that one of these beverages increases the risk of osteoporosis,when, in fact, the critical variable is the low milk intake.

Multicollinearity and confounding are sophisticated terms for

these covariances. Sophisticated solutions for analysis and in-

terpretation are lacking. For example, sometimes the relativecontribution of two nutrients is assessed by putting them both

in a multivariate model to see which one makes the larger con-

tribution, or whether one makes any contribution to risk after

adjusting for the other. This technique is only as good as the

data. When one item is more accurately recalled or quantitatedthan another, it may assume an artificial priority.

Because so many nutrients can be derived (the simplest outputs

usually give at least protein, simple and complex carbohydrate,

saturated and unsaturated fatty acids, and several vitamins andminerals) and because the complexity ofbiologic processes makes

diverse associations biologically plausible, there is also a problem

with multiple testing. Unless the diet-disease association is sug-

gested by other data in animals or humans, or in other words,

based on an a priori hypothesis, there is always the risk that an

association will fall out by chance. Additional evidence for a

real association should be sought, eg, showing a dose-responserelationship between the amount of a nutrient and the risk of

disease.

The converse risk, that a clinically important association willbe missed, probably is even greater. This is because the qualitative

TABLE 7Intercorrelations of major nutrients for men aged 20-59 y,Lipid Research Clinics’ 24-h diet recall*

Nutrient pairs r

Protein X carbohydrate 0.48Protein X fat 0.72Protein X alcohol 0.05Carbohydrate X fat 0.58

Carbohydrate X alcohol 0.05Fat X alcohol 0.02

by guest on October 11, 2015

ajcn.nutrition.orgD

ownloaded from

186S BARRETT-CONNOR

TABLE 8

Odds ratio for calcium inta

used for classifying diet in

ke quintile and by number off

106 women vs true odds ratio*

ood records

True 14 record 34 record 74 record

3.0 1.93 2.17 2.312.5 1.69 1.83 2.042.0 1.61 1.60 1.571.5 1.40 1.32 1.311.0 1.00 1.00 1.00

* Reference 21.

and quantitative errors in the reported diet lead to misclassifi-

cation of the exposure. Randomly distributed misclassification

of exposure tends to bias all associations towards the null. For

this reason any observed association usually underestimates the

true relative risk (Table 8) (21).Misclassification also means that very inconsistent results can

be obtained from study to study, particularly in case-controlstudies where exposure misclassification may not be randomlydistributed and where other biases may exist. Persons with dis-eases prevented or modified by diet often assume a better diet

after diagnosis, but recall of the remote (prediagnosis) diet iscolored by current diet (22). Hospital control subjects may have

changed their diets as a result of the condition for which theywere hospitalized. There is frequent failure to obtain diet data

for the age or interval of interest, when this is known. In thecohort model, the diet is ascertained before disease onset butthe method rarely allows assessment of interim diet or otherbehavior changes that may be relevant. For studies of osteopo-

rosis, diet in young adult life may be the critical period butcollecting such remote data (for case-control studies) or con-

ducting such a long follow-up (for cohort studies) is rarely pos-

sible.

Another issue is the sample size and the power of the study.Grant reviewers like to see sample-size estimates and power cal-

culations for studies of diet and disease, yet rarely are the nec-essary data available to make these calculations in advance. Thevalidity of the data, the range or distribution of the data, andthe strength of the association are seldom known. Because amuch larger sample is required for a cohort study (to yield the

necessary number of events in a reasonable period of time),there are typically fewer resources for repeated or detailed diet

assessments. Increasing the size ofthe study sample reduces some

of the effects of misclassification but increasing the sample sizemay not increase the power sufficiently to compensate for alimited range of the relevant nutrient. Often the range of the

nutrients of interest within the study population is not known

until the diet data are collected. For example, a paper reportingno association of dietary fat with breast cancer in > 85 000women (23) was criticized because nearly all of the women in

this very large cohort were in a narrow range of high fat intake.

On the basis of ecologic studies, within this range only a smallincrement in relative risk would have been expected (24).

Consistency of results is usually an important criterion for

causality in epidemiologic studies ofassociations. In nutritionalepidemiology, lack of consistency does not exclude a causal as-

sociation ( 1 1). The reader can discern many possible explana-tions for the divergent results in diet studies ofhip fracture. Two

recent case-control studies, one in Hong Kong (25) and one in

Britain(26), are shown in Table 9. Study subjects were of different

ethnicity (known to affect bone density) and had very different

average dietary calcium intakes. In both studies, exercise was

more protective than was diet. In both studies, those in the highestquartile of calcium intake were least likely to incur a fracturebut one study concluded that calcium prevented fractures and

the other did not. Although the calcium-fracture association was

not consistent and stepwise, it should be noted that the dietary

calcium in Hong Kong was ascertained entirely from a nine-item food frequency questionnaire whereas that from Britain

used a six-item questionnaire. (A cross-cultural comparison todetermine if the Chinese had higher fracture rates in the face ofsuch limited dietary calcium cannot be obtained with the case-control design, which ascertains neither incidence nor prevalence

of disease.)Prospective studies allow an estimate of fracture-incidence

rates by calcium intake. Comparisons between populations can

be made when the diet-assessment tool and nutrition databaseare prepared and standardized for planned comparisons. The

different study design, diet-assessment methods, and range of

calcium intake could each explain the lack of a statistically sig-nificant (P > 0.05) association ofdietary calcium with hip frac-

ture in a recent population-based prospective study in the United

Kingdom (27) and the significant independent inverse association

(P = 0.008) (28) in a US study, shown in Table 10. The relativerisk (for the lowest tertile ofcalcium intake) ofO.7 in the UnitedKingdom study had confidence limits ofO.l-3.9, which include

the statistically significant lower risk ofO.4 for the highest tertile

ofcalcium intake in Rancho Bernardo. Exactly comparable rel-ative risks cannot be calculated with the data provided in theUnited Kingdom study, but these data could be interpreted as

showing rather similar results, given the study differences.In conclusion, the answer to the question, “How do we know

what they ate?” is that we don’t, exactly, and that no widely

applicable method is a priori better at making an estimate. Recent

excellent, concise reviews and entire books dealing with themethods and limitations of nutritional assessment (see refs 29-32) demonstrate that little has changed since 1960, when Young

and Trulson (33) wrote, “In general one must conclude that, on

TABLE 9

Case-control studies of calcium intake and risk of hip fracture inBritain and Hong Kong

Calcium quintile

Britain*

Men Women

Hong

Men

Kongt

Women

1 (low) 6.2f 1.2 2.1 1.92 5.8 1.4 1.4 1.93 3.3 1.1 1.7 1.14 6.2 1.2 1.5 1.25 (high) 1.0 1.0 1.0 1.0Mean calcium intake of

control subjects (mg) 843 651 177 168Interquartilerange 560-1042 467-799 75-226 75-176

* Reference 26. n = 300 cases and 600 control subjects.

t Reference 25. n = 400 cases and 800 control subjects.f Odds ratio based on setting the highest calcium quintile as the ref-

erence risk at 1.0.

by guest on October 11, 2015

ajcn.nutrition.orgD

ownloaded from

NUTRITIONAL EPIDEMIOLOGY 187S

TABLE 10

Two prospective studies of dietary calcium and the risk of hipfracture*

Rancho

Bernardo, CA United Kingdom

Diet method 24-h recall 74 recordDesign Cohort Nested case control

Number in study 957 141Number of fractures 33 42Age(y) 50-79 65+

Lowest-tertile calcium <416 women <694 women(mg/d) <544 men <588 men

Relative risk O.4t 0.7f

* References 27 and 28.

t Relative risk for highest tertile of calcium intake.� Relative risk for lowest tertile of calcium intake.

an individual basis, results to be obtained from one methodcannot be predicted by another method. With different methods

one is measuring different things. Though comparisons of onemethod with another have been made, these comparisons are

between methods whose accuracy and reliability are not known;therefore no conclusions may be reached regarding whichmethod is the more accurate or reliable.” It is fortunate thatneither complete accuracy nor reproducibility is essential to

produce useful research on any behavior, including diet and

disease. Excessive certainty about the value or nonvalue of anymethod of diet assessment, or the truth of any diet-disease as-

sociation or its absence, should be avoided. Thoughtful com-parisons ofthe results ofdifferent studies are necessary and often

demonstrate considerable consistency despite the limits of dietaryassessment. a

References

1. Linusson EEl, Sanjur D, Erickson EC. Validating the 24-hour recallmethod as a dietary survey tool. Latinoam Nutr l974;24:277-94.

2. Lansky D, Brownell KD. Estimates of food quantity and calories:errors in self-report among obese patients. Am J Clin Nuts l982;35:

727-32.3. Burke BS. The dietary history as a tool in research. J Am Diet Assoc

l947;23: 1041-6.

4. Shekelle RB, Shryock AM, Paul 0, Lepper M, et al. Diet, serum

cholesterol, and death from coronary heart disease. The WesternElectric Study. N EngI J Med 198 1;304:2:65-70.

5. Morgan RW, Jam M, Miller AB, et al. A comparison of dietarymethods in epidemiologic studies. Am J Epidemiol 1978;107:488-98.

6. Young CM, Hagan GC, Tucker RE, Foster WD. A comparison ofdietary study methods. II. Diet history vs seven-day record vs 24-hour record. J Am Diet Assoc 1952;28:218-21.

7. Balogh M, Kahn H, Medalie JH. Random repeat 24-hour dietaryrecalls. Am J Clin Nutr 197 1;24:304-lO.

8. Beaton GH, Mimer J, Corey P, et al. Sources of variance in 24-hourdietary recall data: implications for nutrition study design and in-terpretation. Am J Gin Nutr l979;32:2456-559.

9. Liu K. Measurement error and its impact on partial correlations

and multiple linear regression analyses. Am J Epidemiol l988;127:

864-74.

10. Block G. A review of validations of dietary assessment methods.

Am J Epidemiol 1982;l 15:492-505.

1 1. Willett W. Nutritional epidemiology: issues and challenges. Int J

Epidemiol 1987; 16(suppl):3 12-7.12. VanStaveren WA, Burema J, Deurenberg P, Katan MB. Weak as-

sociations in nutritional epidemiology: the importance of replicationof observations on individuals. Int J Epidemiol l988;l7:964-69.

13. Hankin JH, Stallones RA, Messinger HB. A short dietary methodfor epidemiologic studies. III. Development ofquestionnaire. Am J

Epidemiol 1968;87:285-98.

14. Samet JM, Humble CG, Skipper BE. Alternatives in the collection

and analysis of food frequency interview data. Am J of Epidemiol1984; 120:572-8 1.

15. Mullen BJ, Krantzler NJ, Grivetti LE, Schutz HG, Meiselman HL.Validity ofa food frequency questionnaire for the determination of

individual food intake. Am J Gin Nutr l984;39:136-43.

16. Sampson LS. Food frequency questionnaires as a research instru-

ment. Clin Nutr l985;4: 164-70.

17. Sempos CT, Johnson NE, Smith EL, Gilligan C. Effects of intra-

individual and interindividual variation in repeated dietary records.Am J Epidemiol 1985;12l:l20-30.

18. Rhoads CG. Reliability ofdiet measures as chronic disease risk fac-

tors. Am J Clin Nutr l987;45:1073-9.19. StuffJE, Garza C, O’Brian Smith E, Nichols BL, Montandon CM.

A comparison ofdietary methods in nutritional studies. Am J Gin

Nutr 1983;37:300-6.20. Gordon T, Fisher M, Rifkind BM. Some difficulties inherent in the

interpretation of dietary data from free-living populations. Am J

Gin Nutr l984;39:152-6.21. Freudenheim JL, Johnson NE, Wardrop RL. Nutrient misclassifi-

cation: bias in the odds ratio and loss of power in the Mantel testfor trend. Int J Epidemiol l989;18:232-8.

22. Wu ML, Whittemore AS, Jung DL. Errors in reported dietary intakes.

II.Long-term recall.Am J of Epidemiol l988;l28: 1137-45.

23. Willett WC, Stampfer Mi, Colditz GA, Rosner BA, et al. Dietaryfat and the risk ofbreast cancer. N EngI J Med 1987;316:22-8.

24. Goodwin PJ, Boyd NE Critical appraisal ofthe evidence that dietaryfat intake is related to breast cancer risk in humans. J Natl CancerInt l987;79:473-85.

25. Lau E, Donnan S, Barker DJP, Cooper C. Physical activity andcalcium intake in fracture ofthe proximal femur in Hong Kong. BrMed J 1988;297:144l-3.

26. Cooper C, Barker DJP, Wickham C. Physical activity and calcium

intake in fracture of the proximal femur in Britain. Br Med J

1988;297: 1443-6.27. Wickham CAC, Walsh K, Cooper C, et al. Dietary calcium, physical

activity, and risk of hip fracture: a prospective study. Br Med J

l989;299:889-92.

28. Holbrook TL, Barrett-Connor E, Wingard DL. Dietary calcium andrisk of hip fracture: 14-year prospective population study. Lancet

l988;2: 1046-9.29. Lee-Han H, McGuire V, Boyd NF. A review of the methods used

by studies ofdietary measurement. J Clin Epidemiol l989;42:269-79.

30. Block G, Hartman A. Issues in reproducibility and validity of dietarystudies. Am J Clin Nutr 1989;50:1 133-8.

31. Moon TE, Micozzi MS. eds. Nutrition and cancer prevention. In-vestigating the role of micronutrients. New York: Marcel Dekker,

Inc, 1989.32. Willett, Walter. Nutritional epidemiology. New York: Oxford Uni-

versity Press, 1990.33. Young CM, Trulson MF. Methodology for dietary studies in epi-

demiological surveys. Il-Strengths and weaknesses of existingmethods. Am J Public Health 1960;50:803-14.

by guest on October 11, 2015

ajcn.nutrition.orgD

ownloaded from