Comparative advantage of 3-day food records over 24-hour recall and 5-day food frequency validated...

5
RS.................................................................... ................................................................ RS RE8EIRC8 Comparative advantage of 3-day food records over 24-hour recall and 5-day food frequency validated by observation of 9- and 10-year-old girls PATRICIA B. CRAWFORD, DrPH, RD; EVA OBARZANEK, PhD; JOHN MORRISON, PhD; Z I SABRY, PhD ABSTRACT Objective The validity of the 24-hour recall, 3-day food record, and 5-day food frequency was assessed to decide on a dietary assessment method for the National Heart, Lung, and Blood Institute (NHLBI) Growth and Health Study. Design All subjects were assigned to one of three dietary assessment methods. Unobtrusive observers recorded types and amounts of foods eaten during lunch, and these were compared with the foods reported by the girls in the study. Setting School lunchrooms in California and Ohio. Subjects 58 girls, aged 9 and 10 years. Main outcome measures Reporting errors for dietary assessment methods. Statistical analyses performed Descriptive statistics, matched pair t tests, and Spearman correlation coefficients. Results Comparison of the intakes of energy and selected macronutrients showed different ranges of, and median percentage absolute errors for, each dietary assessment method. Percentage absolute errors ranged between 20 and 33 for the 5-day food frequency method; 19 and 39 for the 24-hour recall; and 12 and 22 for the 3-day food record. The proportion of missing foods (ie, observed food items not reported) and phantom foods (ie, reported food items not observed) by each method were 46% and 40%, respectively, for the 5-day food frequency; 30% and 33%, respectively, for the 24-hour recall; and 25% and 10%, respectively, for the 3- day food record. Applications/conclusions Errors in food reporting and quantification can vary with the type of dietary methodology. Agreement between observed and reported intakes from 3-day food records made it the best overall choice. On this basis, it was selected as the method of assessment for the NHLBI Growth and Health Study. JAm Diet Assoc. 1994; 94: 626-630. P. B. Crawford (corresponding author) is a project manager and Z. I. Sabry is a principal investigator with the School of Public Health, University of California, Berkeley. E. Obarzanek is a project officer with the Divi- sion of Epidemiology and Clinical Applications, National Heart, Lung, and Blood Institute, Bethesda, Md. J Morrisonis a principal investigator with Children's Hospital Medical Cener, Cincinnati, Ohio. Address correspondence to: P. B. Crawford, DrPH, RD, School of Public Health, 140 Warren Hall, University of California, Berkeley, CA 94720. pidemiologic studies of children have demonstrated the difficulty of measuring their dietary intake (1). In addition to the problems of food recall and recording, which are commonly encountered in measuring adults' dietary in- takes, problems with children include low literacy, lack of knowl- edge of foods and food measurement, lack of experience in food preparation, lack of familiarity with components of mixed dishes and added ingredients, general lack of interest, and short atten- tion span. Because there is a need to study the dietary intake of children in nutrition monitoring and surveillance, epidemiologic research, clinical trials, and program evaluation (2), it is impor- tant to validate dietary methods for the type of population to be examined and the intended use of the dietary data. Validation studies of dietary assessment methods have often relied on multiple food records as the reference standard. On occasion, the standard has included debriefing respondents and interviewers to assess accuracy or perception of accuracy (3). Unobtrusive observation, however, is preferable for validation of the instruments of dietary intake measurement (4). The value of this method relies on ascertainment of actual food intake without introducing bias from the subject's awareness that food intake is being observed (5). Unobtrusive observation was used to determine the most valid method of assessing dietary intake in 9- and 10-year-old girls for the Growth and Health Study (GHS) of the National Heart, Lung, and Blood Institute. In this article we present the results of that validation. We tested three methods of collecting dietary informa- tion commonly used in epidemiologic studies: 24-hour recall, 3- day food record, and food frequency. The food frequency was limited to a 5-day period because we were using a school setting. Additionally, we believed it unlikely that children of this age would be able to complete a frequency time frame of more than a week. The purpose of the validation was to determine which method would yield dietary data with the smallest error and the most accurate qualitative and quantitative description of foods eaten. MATERIALS AND METHODS The GHS is an ongoing, multicenter prospective study investigat- ing the development of risk factors for obesity and cardiovascular disease in a cohort of 2,379 black and white girls aged 9 to 10 years at baseline. A detailed description of the study's purpose, cohort, and methods has been published elsewhere (6). The validation study involved 60 black and white girls, 9 to 10 years of age, recruited from fourth and fifth grades in two schools. One was an inner-city school in the San Francisco Bay Area of California in which there were nearly equal numbers of black and white children. The other was a suburban neighborhood school in Cincinnati, Ohio; again, there were nearly equal numbers of black and white children. All girls in a fourth-grade and a fifth-grade class in each school were asked to participate. Teachers and school administrators approved the validation study in advance. The compliance rate was high (97%) because subjects participated during school hours. Two of the 60 girls did not complete the dietary assessment portion of the study because they were absent from school. 626 / JUNE 1994 VOLUME 94 NUMBER 6

Transcript of Comparative advantage of 3-day food records over 24-hour recall and 5-day food frequency validated...

RS.................................................................... ................................................................RS

RE8EIRC8

Comparative advantage of 3-day food records over24-hour recall and 5-day food frequency validated byobservation of 9- and 10-year-old girlsPATRICIA B. CRAWFORD, DrPH, RD; EVA OBARZANEK, PhD; JOHN MORRISON, PhD; Z I SABRY, PhD

ABSTRACT

Objective The validity of the 24-hour recall, 3-day food record,and 5-day food frequency was assessed to decide on a dietaryassessment method for the National Heart, Lung, and BloodInstitute (NHLBI) Growth and Health Study.Design All subjects were assigned to one of three dietaryassessment methods. Unobtrusive observers recorded typesand amounts of foods eaten during lunch, and these werecompared with the foods reported by the girls in the study.Setting School lunchrooms in California and Ohio.Subjects 58 girls, aged 9 and 10 years.Main outcome measures Reporting errors for dietaryassessment methods.Statistical analyses performed Descriptive statistics,matched pair t tests, and Spearman correlation coefficients.Results Comparison of the intakes of energy and selectedmacronutrients showed different ranges of, and medianpercentage absolute errors for, each dietary assessmentmethod. Percentage absolute errors ranged between 20 and33 for the 5-day food frequency method; 19 and 39 for the24-hour recall; and 12 and 22 for the 3-day food record. Theproportion of missing foods (ie, observed food items notreported) and phantom foods (ie, reported food items notobserved) by each method were 46% and 40%, respectively,for the 5-day food frequency; 30% and 33%, respectively, forthe 24-hour recall; and 25% and 10%, respectively, for the 3-day food record.Applications/conclusions Errors in food reporting andquantification can vary with the type of dietary methodology.Agreement between observed and reported intakes from 3-dayfood records made it the best overall choice. On this basis, itwas selected as the method of assessment for the NHLBIGrowth and Health Study. JAm Diet Assoc. 1994; 94:626-630.

P. B. Crawford (corresponding author) is a projectmanager and Z. I. Sabry is a principal investigator withthe School of Public Health, University of California,Berkeley. E. Obarzanek is a project officer with the Divi-sion of Epidemiology and Clinical Applications, NationalHeart, Lung, and Blood Institute, Bethesda, Md.J Morrison is a principal investigator with Children'sHospital Medical Cener, Cincinnati, Ohio.

Address correspondence to: P. B. Crawford, DrPH, RD,School of Public Health, 140 Warren Hall, University ofCalifornia, Berkeley, CA 94720.

pidemiologic studies of children have demonstrated thedifficulty of measuring their dietary intake (1). In additionto the problems of food recall and recording, which arecommonly encountered in measuring adults' dietary in-

takes, problems with children include low literacy, lack of knowl-edge of foods and food measurement, lack of experience in foodpreparation, lack of familiarity with components of mixed dishesand added ingredients, general lack of interest, and short atten-tion span. Because there is a need to study the dietary intake ofchildren in nutrition monitoring and surveillance, epidemiologicresearch, clinical trials, and program evaluation (2), it is impor-tant to validate dietary methods for the type of population to beexamined and the intended use of the dietary data.

Validation studies of dietary assessment methods have oftenrelied on multiple food records as the reference standard. Onoccasion, the standard has included debriefing respondents andinterviewers to assess accuracy or perception of accuracy (3).Unobtrusive observation, however, is preferable for validation ofthe instruments of dietary intake measurement (4). The value ofthis method relies on ascertainment of actual food intake withoutintroducing bias from the subject's awareness that food intake isbeing observed (5).

Unobtrusive observation was used to determine the most validmethod of assessing dietary intake in 9- and 10-year-old girls forthe Growth and Health Study (GHS) of the National Heart, Lung,and Blood Institute. In this article we present the results of thatvalidation. We tested three methods of collecting dietary informa-tion commonly used in epidemiologic studies: 24-hour recall, 3-day food record, and food frequency. The food frequency waslimited to a 5-day period because we were using a school setting.Additionally, we believed it unlikely that children of this age wouldbe able to complete a frequency time frame of more than a week.The purpose of the validation was to determine which methodwould yield dietary data with the smallest error and the mostaccurate qualitative and quantitative description of foods eaten.

MATERIALS AND METHODSThe GHS is an ongoing, multicenter prospective study investigat-ing the development of risk factors for obesity and cardiovasculardisease in a cohort of 2,379 black and white girls aged 9 to 10 yearsat baseline. A detailed description of the study's purpose, cohort,and methods has been published elsewhere (6).

The validation study involved 60 black and white girls, 9 to 10years of age, recruited from fourth and fifth grades in two schools.One was an inner-city school in the San Francisco Bay Area ofCalifornia in which there were nearly equal numbers of black andwhite children. The other was a suburban neighborhood school inCincinnati, Ohio; again, there were nearly equal numbers of blackand white children. All girls in a fourth-grade and a fifth-gradeclass in each school were asked to participate.

Teachers and school administrators approved the validationstudy in advance. The compliance rate was high (97%) becausesubjects participated during school hours. Two of the 60 girls didnot complete the dietary assessment portion of the study becausethey were absent from school.

626 / JUNE 1994 VOLUME 94 NUMBER 6

Table 1Means+± standard deviations and least significant differencesa between observed and reported energy and nutrient intakes from lunch, bymethod

Energy 24-Hour recall (n = 19) 3-Day record- 5-Day food frequency-and Observed Recall Least mean of 3 days mean of 5 daysnutrient significant (n = 20) (n = 19)

difference Observed Record Least Observed Frequency Leastsignificant significantdifference difference

Energy (kcal) 564 + 209 610 + 274 120 489 +-200 490-+ 232 55 502 +165 685 + 221 115

Carbohydrate (g) 68+30 75+31 16 66+26 67+29 8 63+23 89+30 16

Protein (g) 22+8 21+8 4 16+9 16+10 2 18+8 24-+10 5

Fat(g) 24+11 26+17 8 19+9 19+10 3 21+7 28+11 5

Saturated fat (g) 10+5 11 6 2 8+5 7+4 1 8+4 11 +4 2

Cholesterol (mg) 67+34 60-36 15 47+34 49+41 8 49+26 71 -37 17

aDifference in means that would be significant with a paired ttest at the 5% level.

Table 2Medians, means, and standard deviations (SDs) of differencesa between observed and reported energy and nutrient intakes from lunch, bymethod

Energy and 24-Hour recall (n = 19) 3-Day food record (n =20) 5-Day food frequency (n = 19)nutrient Median Mean SD Median Mean SD Median Mean SD

Energy (kcal) -42 -46 249 16 - 1 117 -136 - 184*** 238

Carbohydrate (g) - 7 - 7.5 32.6 0.5 - 0.9 16.3 -18 - 25*** 33

Protein (g) 2 1.2 7.9 0.5 -0.3 3.9 -4 -6.7 10.6

Fat (g) 2 -2.5 15.7 1 0.4 6 -5 -6.9* 10.7

Saturated fat (g) 0 -0.2 4.3 1 0.5 2.1 -2 -2.8*** 3.8

Cholesterol (mg) 11 7 31 4 -2 19 -7 -22 36

aDifferences were calculated by subtracting reported from observed intakes.*P<.05. ***P<.001.

The observation study was conducted by GHS observers inschool lunchrooms. The observers were trained to follow a com-mon protocol for unobtrusive observation of food consumption.They discreetly located girls selected to be in the study by usingclass pictures obtained earlier. If asked, the GHS observers saidthey were lunchroom monitors. Each observer watched eitherone or two girls during the lunch period, depending on thecafeteria size. No more than two girls could be observed by a singleobserver because of the difficulty involved in observing thetrading of foods that occurred.

Each of the 60 girls was randomly assigned to one of the threemethods: 24-hour recall, 3-day food record, or 5-day food fre-quency. The 20 girls assigned to the 24-hour recall method wereobserved eating lunch. On the following day, each girl was takenout of class and interviewed using a standard 24-hour recall, open-ended style. A staff person who was not involved in the observa-tion phase of the study conducted the interview using portion-sizeaids. Girls were not told that information from their recalls wouldbe compared with observation data collected on the previous day.

The 20 girls assigned to the 5-day food frequency method wereobserved during the lunch period for five consecutive days (Mon-day through Friday). On the following Monday, these girls wereasked to complete a 5-day food frequency for the lunchtimeportion of the day for Monday through Friday of the previousweek. The instrument was developed specifically for children and

was adapted for low literacy by administering it one-on-one. Thequestionnaire covered 42 types of foods; spaces were allowed foradditional foods, descriptors, and portion sizes. The key words,"lunchtime" and "Mondaythrough Friday," were repeated through-out to keep the girls' attention on the appropriate time period.

The following food frequency technique was used. Each girlwas asked about the number of times she had eaten certain foodsduring the past 5 school days (Monday through Friday). Once thenumber of times was determined, the subject was asked aboutportion sizes using a set of food-portion aids. The interviewerfilled in information on the open-ended, quantitative food fre-quency. The wording was: "I'm going to ask you some questionsabout how often you ate certain foods at lunchtime last week. Ifyou have any questions, please stop me and ask. 'Times per week'means Monday through Friday only." Questions were phrased asfollows: "How often did you drink milk at lunchtime Mondaythrough Friday last week? You may answer with the number oftimes or 'never' or 'don't know."' If the girl drank milk, she wasasked, "Do you know whether it was whole milk, low-fat or nonfatmilk, or chocolate milk?" Then she was asked, "Now can you tellme how much milk you drank?" The interviewer used aids todetermine quantity and averaged portions if more than one sizewas mentioned.

The third group of 20 girls was given the 3-day food record. Theform for this record was based on the one used by children of the

JOURNAL OF THE AMERICAN DIETETIC ASSOCIATION / 627

RESEARCH

Table 3Median percentage absolute error (PAE)' and Spearman correlations QJ) between observed and reported energy and nutrient intakes fromlunch, by method

Energy 24-Hour recall 3-Day food record 5-Day food frequencyhandle PAE r.PAE r.PAE r

Energy 1 9 .57 1 2 .85 28 .19

Protein 1 9 .61 1 4 .94 26 .40Carbohydrate 26 .46 1 6 .78 33 .11

Fat 39 .57 20 .83 23 .31

Saturated fat 20 .79 18 .90 20 50

Cholesterol 31 .74 22 .91 21 .43

'PAE= (observed-reported) 100observed

Table 4Percentage of missing foodsO and phantom foodsb from lunch, by method

Method Percentage of missing foods' Percentage of phantom foodsbMedian 25th Percentile 75th Percentile Median 25th Percentile 75th Percentile

24-hour recall (n =19) 30 9 57 33 0 563-day food record (n = 20) 25 1 6 31 1 0 0 265-day food frequency (n =19) 46 35 53 40 23 52

aMedian percentage of observed foods not reported.bMedian percentage of reported foods not observed.

sarne age in the Berkeley Longitudinal Nutrition Study (7). Thefon-ns had spaces for name, date, times that foods (or drinks) wereconsumed, portion sizes, and descriptors (including method ofpreparation, if known). GHS staff instructed the girls on themethod of keeping a 3-day food record. The 3-day food recordswere individually reviewed with the children on the Monday of thefollowing week. This review was conducted to provide additionalinformation on the foods, to add missing foods when rememberedduring the interview, and to elaborate or clarify portion sizesaccording to standard measurement aids. The review process isparticularly important with children who are less literate and needhelp with writing or spelling.

All methods were administered (or reviewed in the case of the3-day food record) one-on-one by a nutrition interviewer. Esti-mated portion sizes for all methods were determined using rulersand standard sets of measuring cups and spoons.

Data from both schools collected by the three methods werecoded centrally using the mini-list food-coding system (8). Themird-list coding system contains 202 commonly consumed foodswith provisions for substituting and grouping when necessary.This coding system provides values for 49 nutrients obtained fromUS Department of Agriculture food tables, the literature, labora-tory analyses, and information from private food companies.Analysis variables were lnited to those of particular relevance tothe goals and objectives of GHS: energy, carbohydrate, protein,total fat, saturated fat, and cholesterol.

Statistical MethodsMedians and ranges were obtained for reported and observedintakes of energy and the aforementioned five nutrients. Descrip-tive statistics for differences (observed vs reported) were alsocalculated for each method. Paired tests were performed foreach nutrient to test for systematic bias. All t tests were two tailed.

Least significant differences were calculated for energy and thefive nutrients. Least significant difference is defined as the small-

est difference that would be statistically significant with a pairedt test; it is a reflection of the sensitivity of the method based on thevariances of the measures and the number of subjects.

Percentage absolute errors (PAEs) were computed betweenruissing foods (observed food items not reported) and phantomfoods (reported food items not observed). Pairs of descriptivestatistics were obtained by calculating the proportion of missingand phantom foods for each child. A median for both of theproportions was obtained for each nutrient by method. Becausewe obtained absolute values, errors of missing foods do not cancelerrors of phantom foods when computing means. Therefore, PAEis defined as the absolute value of the following.

observed value - reported value x 100observed value

We also calculated Spearman correlations between observed andreported nutrient intakes.

Of the foods both reported and observed, the percentage oferror in quantifications of amounts eaten was calculated using thetrained observer's estimates of portion size. The distribution ofpercentage errors in the child's quantification of foods was groupedinto four categories: 10% error or less, 1 1% to 25% error, 26% to50% error, and greater than 50% error.

RESULTSTable presents the means and standard deviations of lunchtimeintakes of energy and five nutrients. Observed intakes are com-pared with those reported by the girls according to the threemethods. A consistent bias resulting in the overestimation ofenergy and nutrients was observed for the 5-day food frequencymethod of reporting. Bias was not observed for the 24-hour recallor the 3-day record.

Least significant differences between observed and reportedvalues were half as large for energy and the five nutrients using the3-day records compared with the 24-hour recall and the 5-dayfood frequency (Table ). These lunchtime intakes are further

628 / JUNE 1994 VOLUME 94 NUMBER 6

Table 5Distribution of percentage absolute errora in quantification of correctly reported foods from lunch, by method

Error 24-Hour recall 3-Day food record 5-Day food frequency% No. % No. % No.

<10% 50 47 36 78 21 40

11%-25% 8 8 12 25 9 17

26%-50% 18 17 24 52 20 37

>50% 24 23 28 60 50 93

Total 100 95 100 215 100 187

aPAE = observed (g) - reported (g)observed (g) x 100

compared in Table 2 using the medians, means, and standarddeviations of the differences between observed and reportedintakes. A paired t test consistently yielded no differences be-tween observed and reported intakes using the 24-hour recall andthe 3-day food record. However, significantly higher mean valuesfor reported than for observed intakes analyzed from the 5-dayfood frequency method were found using the paired t test. TheWilcoxon signed rank test was also performed and similar resultswere obtained. The variability of the 3-day food record was half aslarge as that derived from the other methods.

Table 3 summarizes PAE and Spearman correlations betweenthe observed and reported energy and nutrient intakes for each ofthe three methods. The PAEs were relatively high and correla-tions were low for the 5-day food frequency method. For the 24-hour recall, PAEs and correlations were intermediate. For the 3-day food record, correlations ranged from .78 for carbohydrate to.94 for protein, and median PAEs ranged from 12% for energy to22% for cholesterol.

The magnitude of underreporting (missing foods) and over-reporting (phantom foods) by the girls is presented in Table 4.The 5-day food frequency method had the highest level of missingfoods. For this method, the median percentage of observed fooditems that were not reported was 46%. The 3-day food record wasthe method with the lowest percentage of missing food items-25%. Smaller errors were found for reported food items that werenot observed, but the same pattern emerged. The 3-day foodrecord was the method with the fewest phantom food items. Themedian percentage of food items reported but not observed was10% for the 3-day food record, 33% for the 24-hour recall, and 40%for the 5-day food frequency.

The three methods were also compared to determine thedistribution of errors in the girls' quantifications of correctlyreported foods. The PAE was calculated for each food item notedby the observer and reported by the girl. Each of the correctlyreported food items was then classified by level of quantificationerror into four categories. Table 5 presents the PAE by method.Fifty percent of the correctly reported foods on the 24-hour recallhad quantification errors of less than 10%, whereas the quantifi-cation errors were 36% and 21%, respectively for the 3-day foodrecord and the 5-day food frequency. Fifty percent of the foodreported on the 5-day food frequency had quantification errorsgreater than 50%. Portion sizes were more often overestimatedwith the food frequency method.

DISCUSSIONValidations of dietary methods, though valuable, are relativelyrare, even with the current wealth of research on dietary collec-tion methods. In 1986, The American Dietetic Association corn-

piled an annotated bibliography on dietary methodology in whichclassification of the references revealed that fewer than 7% weredirect or observational validation studies (9). Some were classi-fied as indirect validation studies that compared the energy ornutrients consumed with energy expenditure or other biochemi-cal measures (10,11). Most validation studies compared therelative accuracy of one method to another. These studies may bemore accurately regarded as calibration studies. Most of theseused food records as the standard reference forjudging the othermethod (12,13). Some used repeated 24-hour recalls as thestandard (14).

Most observational studies with children have compared re-sults using the 24-hour recall method. Three studies comparedthe recall of caretakers of young children with observers' reportsof the children's intake. In a study of children aged 4 to 9 years,Klesges et al (15) found that the caretakers' reports of dietaryintake were highly correlated with observers' reports (between.64 and .83). They also reported mean differences betweenobserved and reported intake at lunch: between-34.7 and + 11 forenergy, -. 9 and +4.4 for grams of fat, and -8.9 and +12 for gramsof carbohydrate. A study by Basch and colleagues (16) found thatcaretakers of 4- to 7-year-old children recalled an average of 42kcal more per meal than were calculated based on observation. Ina study comparing observed with recalled energy intakes byparents of 4- to 9-year-old children, correlations ranged between.41 and .61 (17).

Comparing the 24-hour recall intakes of 9- to 10-year-oldchildren (n=36) with that of observers, Emmons and Hayes (18)reported correlations between observed and reported values of.77 for energy and .82 for protein. In our study, these correlationswere .57 for energy and .61 for protein. The values reported byEmmons and Hayes may have been somewhat higher because thechildren who participated were aware of the study and were notdirectly observed. Food on the tray was recorded before and aftereating. Children ate only the school lunch. Possible trading offoods was not mentioned.

Other dietary recall studies have investigated the combinedoccurrences of missing foods and phantom foods. Meredith andcoworkers (19) used a dietary recall to compare the school lunchand observation data of 9- to 18-year-olds. The recall was obtainedbetween 30 minutes and 2 hours after the lunch period. Theauthors reported a food agreement score of 35%. The combinederrors of missing foods and phantom foods were 65% for the 94children studied. Combined errors of missing foods and phantomfoods in our study group using the 24-hour recall method weresimilar--63% .When college-aged students were studied, Krantzleret al (20) found a food-item agreement score of 69%-nearlydouble that found with studies of young children. They reported

JOURNAL OF THE AMERICAN DIETETIC ASSOCIATION / 629

RESEARCH

an even higher food agreement score of 75% when the 3-day foodrecord method was used for college-aged students. In our studyof 9- to 10-year-old girls, the food agreement score was the high-est for the 3-day food record-5% agreement.

Baranowski et al (21) used a 1-day food frequency with 24third- to sixth-grade children and found an 82.9% agreementbetween the children's and the observers' food records. Parentshad assisted their children in completing the form, however, andboth parents and children were aware that they were beingobserved, which could have enhanced their ability to rememberfoods. Another possible contributor to the agreement score wasthe use of common food terms on both the observers' andchildren's forms.

The findings from our study demonstrated differential dietaryreporting results for 9- to 10-year-old girls when different dietarymethodologies were used. Girls who used a 3-day record reportedtheir observed intakes more accurately than did girls who used a24-hour recall or a 5-day food frequency.

Intakes for breakfast, dinner, or snacks could not be observed,which limited the study to intakes at lunchtime. This factor shouldnot negate the primary advantage of the 3-day record. Theadvantage is that the period of recording is soon after eating; thus,the method does not rely as heavily on the subject's memory as dothe other two methods. This advantage would most likely be truefor other meals and snacks, but the degree of similarity cannot bepredicted because lunchtime eating includes only lunch foodseither brought from home or provided by the school and includesno restaurant foods or snacktime eating.

APPLICATIONSDietitians and nutritionists need results from actual observationstudies to assess the relative value of the various dietary assess-ment tools used with children. These assessments must be con-sidered, however, in the context of the cost and time burden forthe intended task.

In the GHS, it was necessary to obtain dietary intake datarelevant to the development of risk factors for cardiovasculardisease. The method selected-the 3-dayfood record-was laborintensive and costly, but it held the promise of sufficient accuracyfor the intended study.

The food frequency, though not sufficiently accurate for the useproposed in the GHS, was the least time consuming for subjectsand staff members. For some studies, this feature may be desir-able. The food frequency tended to yield a relatively high percent-age of phantom foods and errors in food quantification. Theseresults need to be considered in relation to the objectives of theintended task for dietary assessment.

The 24-hour recall interview was more time consuming thanadministration of the food frequency but less so than the debrief-ing for the 3-day food record. The issue of most concern with the24-hour recall was the high degree of reporting of phantom foodsand missing foods. We do not know whether some girls remem-bered a different lunch meal from the one they actually ate the daybefore, or whether the girls who forgot the meal felt compelled toprovide our interviewers with some information. Girls were givenpermission to say they could not remember, but none did so in ourpresence.

The 3-day food record was the most time consuming to admin-ister: the girls had to be instructed in its use and then debriefed.Nonetheless, it offered the highest degree of accuracy of reportingwith the lowest proportion of missing foods and phantom foods.Accuracy of the quantification of foods was similar to that seenwith the 24-hour recall method. The PAEs between observed andreported intakes were lower for energy and all nutrients exam-ined, which indicates higher correlations and less bias for the 3-day food record. ·

The research was performed under contract NO1-HC-55023-26 of the National Heart, Lung, and Blood Institute,National Institute of Health, Bethesda, Md.

The authors thank Paul Canner, PhD, who assisted withanalysis; Rhea Larsen, MS, who directed the study inCincinnati, Ohio; Leona Shapiro, DrPH, who providedguidance and support throughout; Ruth Huenemann, DSc,and Ann Burroughs, DrPH, for their valuable and timelyassistance; and Kate McBurrey, DrPH, who was respon-siblefor the dietary coding.

References1. Wynder EL. The need for frequent and better dietary assessment ofchildren. Prev Med. 1990; 19:489-490.2. Healthy People 2000: National Health Promotion and DiseasePrevention Objectives. Washington, DC: US Dept of Health and HumanServices; 1990.3. Rizek RL, Pao EM. Dietary intake methodology, 1: USDA surveys andsupporting research. JNutr. 1990; 120:1525-1529.4. Simons-Morton BG, Baranowski T. Observation in assessment ofchildren's dietary practices. JSch Health. 1991; 61:204-207.5. Block G. A review of validations of dietary assessment methods. AmJEpidemiol. 1982; 115:492-505.6. NHLBI Growth and Health Study Research Group. Obesity andcardiovascular disease risk factors in 9- and 10-year-old black and whitegirls: NHLBI Growth and Health Study. JAm Public HealthAssoc. 1992;82:1613-1620.7. Crawford P, Hankin J, Huenemann R. Environmental factors associ-ated with preschool obesity: dietary intakes, eating patterns andanthropometric measurements. JAm Diet Assoc. 1978; 72:589-596.8. Pennington J. Dietary Nutrient Guidelines. Westport, Conn: AVIPublication Co; 1976.9. Research Dietetic Practice Group. Validity and Reliability inDietary Methodology-AnAnnotated Bibliography. Part 3. Chicago,

ll: American Dietetic Association; 1986.10. Willett WC, Stempfer MJ, Underwood VA, Speizer FE, Rosner B,Hennekens CH. Validation of a dietary questionnaire with plasmacarotenoid and alpha-tocopherol levels.AmJClinNutr. 1983; 38:631-639.11. Bandini LG, Schoeller DA, Cyr HN, Dietz WH. Validity of reportedenergy intake in obese and nonobese adolescents. Am J Clin Nutr.1990; 52:421-425.12. Flegal KM, Larkin FA, Metzner HL, Thompson FE, Guire KE.Counting calories: partitioning energy intake estimates from a foodfrequency questionnaire. Am JEpidemiol. 1988; 128:749-760.13. Van Staveren WA, Burema J. Dietary methodology: implications oferrors in the measurement. Recent studies on food and nutrient intakein Ireland. Proc Nutr Soc. 1990; 49:281-287.14. Jenner DA, Neylon K, Croft S, Beilin LJ, Vandongen RA. Compari-son of methods of dietary assessment in Australian children aged 11-12years. Eur J Clin Nutr. 1989; 43:663-673.15. Klesges RC, Hanson CL, Eck LH, Durff AC. Accuracy of self-reportsof food intake in obese and normal-weight individuals: effects of paren-tal obesity on report of children's dietary intake. Am J Clin Nutr. 1988;48:1252-1256.16. Basch CE, Shea S, Arliss R, Contento IR, Rips J, Gutin B, IrigoyenM, Zybert P. Validation of mothers' reports of dietary intake by 4- to 7-year-old children. Am JPublic Health. 1990; 81:1314-1317.17. Eck L, Klesges RC, Hanson C. Recall of a child's intake from onemeal: are parents accurate? JAm Diet Assoc. 1989; 89:784-789.18. Emmons L, Hayes M. Accuracy of 24-hour recalls of young children.JAm Diet Assoc. 1973; 62:409-415.19. Meredith A, Matthew A, Zickefoose M, Weagley E, Wayave M,Brown EG. How well do school children recall what they have eaten? JAm DietAssoc. 1951; 27:749-751.20. Krantzler NJ, Mullen BJ, Schutz HG, Grivetti LE, Holden CA,Mebelman HL. Validity of telephoned diet recalls and records forassessment of individual food intake. Am J Clin Nutr. 1982; 36:1234-1242.21. Baranowski T, Dworkin R, Henske JC, Clearman DR, Dunn JK,Nader PR, Hooks PC. The accuracy of children's self-reports of diet:Family Health Project. JAm Diet Assoc. 1986; 86:1381-1385.

630 / JUNE 1994 VOLUME 94 NUMBER 6