quality control of dietary data collection in the cardia study

Quality Control of Dietary Data Collection in the CARDIA Study

Joan E. Hilner, MPH, MA, RD; Arline McDonald, PhD; Linda Van Horn, PhD, RD; Charlotte Bragg, MS, RD; Bette Caan, DrPH; Martha L. Slattery, PhD, RD; Robert Birch, PhD; Carey G. Smoak, MSPH; and Janet Wittes, PhD CARDIA Coordinating Center, Division of General and Preventive Medicine, University of Alabama at Birmingham, Birmingham, Alabama (J.E.H.); Department of Community Health and Preventive Medicine, Northwestern University Medical School, Chicago, Illinois (A.M. and L.V.H.); Division of General and Preventive Medicine, Behavioral Medicine Unit, University of Alabama at Birmingham, Birmingham, Alabama (C.B.); Division of Research, Kaiser Permanente Medical Care Program (Northern California Region), Oakland, California (B.C.); Department of Family and Preventive Medicine, University of Utah, Salt Lake City, Utah (M.L.S.); Response Technologies, Memphis, Tennessee (R.B.); Department of Neurology, Loma Linda University, School of Medicine, Loma Linda, California (C.G.S.); and New England Research Institute, Inc., Watertown, Massachusetts (J. W.)

ABSTRACT: The Coronary Artery Risk Development in (Young) Adults (CARDIA) Study developed and implemented quality control (QC) measures to minimize misclassification associated with dietary data. Manual and automated data inspection were used to monitor quality. Of the 5111 participants who completed a dietary history, 717 (14%) had dietary forms reviewed and 153 (3%) had the interview audiotaped. Results show that for the 717 forms reviewed, the overall form complet ion error rate was 0.22% and the "critical" error rate (i.e., those errors impact ing on nutr ient computat ions) was 0.12%. The propor t ion of forms free of any type of error increased over time (p ~ 0.0001). The discrepancy rate in recording and interviewing methods as est imated from the 153 audio taped interviews was 0.7%. Inter-interviewer differences were small as indicated by the audio taped interviews and the propor t ion of error-free forms completed by interviewers. The results indicate that the dietary data collected in CARDIA were completely and accurately recorded for use in analysis.

Address reprint requests to: Joan Hilner, MPH, MA, RD, CARDIA Coordinating Center, 1717 1 l th Avenue South, Medical Towers Building, Room 504, Birmingham, AL 35205.

Received July 24, 1989; revised September 6, 1991. This study was supported by contracts NO1-HC-48047, NO1-HC-48048, NO1-HC-48049, NO1-HC-

48050, and NO1-HC-95095 from the National Heart, Lung and Blood Institute, National Institutes of Health.

156 0197-2456/92255.00

Controlled Clinical Trials 13:156-169 (1992) © Elsevier Science Publishing Co., lnc. 1992

655 Avenue of the Americas, New York, New York 10010

Quality Control of Dietary Data Collection 157

INTRODUCTION

Epidemiology is an exercise in measurement [1]. The degree to which we are able to measure disease and exposure accurately in epidemiological studies impacts on our ability to classify individuals correctly and subsequently determine the magnitude of the association between a given exposure and disease. Quality control (QC) efforts conducted during data collection should theoretically minimize misclassification. Although detailed descriptions of the QC procedures used in the collection of dietary data have been published in the past 5 years [2-6], usually only brief mention of dietary QC is reported [7-9].

In this report, we describe the QC measures used in the collection of dietary data in a longitudinal study of Coronary Artery Risk Development in (Young) Adults (CARDIA). In addition, we provide the results of the monitoring techniques.

METHODS

The CARDIA cohort, consisting of 5115 black and white men and women, aged 18-30 years old, with a range of formal education, was initially examined during 1985-1986 in four clinical centers: Birmingham, Alabama; Chicago, Illinois; Minneapolis, Minnesota; and Oakland, California. The recruitment procedures and the baseline examination have been described elsewhere [10- 12].

The CARDIA Dietary History [13] is an interviewer-administered instru- ment modeled after the Burke method [14]. It is similar in organization to the history used in the Western Electric Study [15] and was validated in a cohort demographically similar to CARDIA [16]. The CARDIA Dietary History obtained information on both general dietary practices and "usual" intake during the past month with a quantitative food frequency questionnaire. In the food frequency section of the interview, participants were queried about their intake of specific food groups. For specific food items, participants were asked about typical portion sizes, frequency of consumption, and preparation methods (Figure 1). Physical models of food, measuring cups, measuring spoons, and cue cards with lists of food items were used to aid recall of specific foods and portion sizes. Interviewers recorded participant responses onto paper forms and coded the history after the interview. Data entry personnel at each clinic keyed the completed histories onto computers. The diet history data were analyzed at the Coordinating Center to calculate nutrient intakes for each participant.

Quality Control Methods

In conjunction with the development and implementation of the CARDIA Dietary History, a series of measures were designed for QC purposes (Table 1). The specific QC methods were selected (1) to minimize interviewer effects, (2) to decrease missing or erroneous data, and (3) to reduce coding and transcription errors at both data collection and data entry.

158 J.E. Hilner et al.

GRAINS/CEREALS CONTINUED

24. O. D o y o u u m m l l y ~ l i l ~ o r c o M c ~ l ?

O. I'knv m ~ ok) ~ u s ~ have?

O. Whm tdml do y ~ ~mal~/Itme?

Q. How one?

Commeem

NO

CP

~ b AmmsM

s)

l te0

~es

Figure 1 Example of dietary history structure.

Interviewer Effect

Interviewers can introduce systematic and random errors both during the interview and in the completion of the data collection form [17,18]. Participant response may vary according to the manner and degree of interviewer prob- ing. The interviewer's behaviors, such as mannerisms, gestures, and non- verbal communication, may affect the participant's responses. Several steps were taken to reduce interviewer effect.

The interview structure, including the wording and order of questions, was standardized to minimize differences within and among interviewers. A detailed manual containing the protocol, manual of operations, study forms, and specific coding rules was developed as a training tool and was used as a reference throughout the study.

The 13 original CARDIA interviewers (either registered dietitians [R.D.] or R.D.-eligible nutritionists) participated in a 1-week centralized training session. The nutrition coordinator, located at the Coordinating Center, certified

Quality Control of Dietary Data Collection

Table 1 Quality Control Measures Implemented for Dietary Data Collection

159

Quality Control Measure

Intended to Minimize:

Interviewer Missing Coding Effect Data Errors

Form Development: Standardized interviewing format Standard food items on history precoded

to the Nutrition Coordinating Center's (NCC) nutrient database

Training~Certification: Development of CARDIA Nutrition

Manual containing the protocol, manual of operations, dietary forms, and coding rules

Centralized training of nutritionists and data entry personnel

Certification of nutritionists with specific criteria for passing

Monitoring During Data Collection: All histories reviewed for completeness

of data at the clinics by nutritionists 10% random sample of forms reviewed

for completeness by nutrition coordinator at Coordinating Center

153 audiotaped interviews and the corresponding history reviewed for accuracy in documentation and conduct of interview by nutrition coordinator

Data Entrt/ and Processing: Precoded-screens for data entry Automated editing at data entry Error processing of each participant's

record prior to computation of nutrient data

X X X

X X X

X

X

X X

X

X

X X

X X X X X

interviewers only after a set of criteria were fulfilled (Table 2). Three additional nutritionists were subsequently trained locally by the centrally trained chief nutritionist who used the same curriculum and performance criteria employed in the central training session. Recertification was not required because con- tinuous monitoring procedures were utilized throughout the examination cycle.

It was possible that as the initial training became more distant in time, more errors in conducting the dietary history might occur. To detect this anticipated "drift" from the study protocol, the study initiated a monitoring procedure to assess interviewing style and the accuracy with which an interviewer recorded the participant's responses. Each CARDIA nutritionist submitted one tape-recorded interview with the corresponding dietary history form to the nutrition coordinator every 4 weeks for the first 3 months of the examination and every 6 weeks thereafter. The nutrition coordinator specified the week that taping of interviews should occur; the chief nutritionist in each


Table 2 Criteria for Certification of Nutritionists

1. Attended centralized or local training session. 2. Conducted a total of 10 interviews (two during training and eight during the

certification period). 3. Chief nutritionist observed and evaluated two interviews with both interviews

rated as acceptable. 4. Edited/completed the 10 dietary history forms from the interviews. 5. Forms from the last two interviews submitted to the nutrition coordinator for

evaluation of form completion. Both forms required a form completion error rate of 3% or less. If the error rate for either form exceeded 3%, four additional interviews were conducted with the forms from the last two interviews submitted to the nutrition coordinator for evaluation.

clinical center selected the interviews to be taped as participants entered the clinic. All participants consented to tape recording of the interview. A total of 153 tapes were submitted from all 16 nutritionists over the 14 months of the first examination.

The nutrition coordinator compared the audiotape with a copy of the completed form. Discrepancies were classified into two categories: improper recording of a participant's responses and deviations from the interview protocol. The discrepancy rate for each form was calculated as the total number of discrepancies divided by the total number of fields.

The nutrition coordinator provided regular evaluations to each nutritionist to maintain level of performance and to correct discrepant interviewing behaviors quickly. The reliability of the nutrition coordinator in reviewing the audiotaped interviews was not checked. While this would be desirable, it was not done given the extra cost involved.

Missing Data

Given the length of the dietary history and the need for 3-5 pieces of information per reported food item, incomplete data collection was a primary concern. To minimize the amount of missing data, nutritionists reviewed the dietary forms (ideally before the participant left the clinic) using the CARDIA Form Completion Checklist, which is similar to the checklist used by the University of Minnesota's Nutrition Coordinating Center (NCC) nutritionists for 24-hr-recall dietary data [2]. A second nutritionist reviewed each form for completeness before data entry.

The nutrition coordinator manually reviewed a randomly selected 10% sample of all diet histories (n = 564) in addition to the 153 forms submitted in conjunction with the tape-recorded interviews. A rigid review process was adopted; an error was defined as any discrepancy in form completion, irre- spective of its importance in the ultimate computation of nutrients.

The form completion error rate was calculated using the following formula:

total number of errors Form Completion Error Rate =

total number of responses

where the denominator (number of responses) is equal to the sum of the following:


1. Number of food items reported, 2. Number of sections marked "no," 3. Number of questions regarding dietary practices, and 4. Number of questions used to end and assess the interview.

Because the first two terms in the denominator vary with each interview, the denominator of the error rate differs for each diet history. The mean number of responses was 165.9 (SD = 21.1) with a range of 126-277.

Of course, not all errors impact equally on the quality of the dietary data; some are considered "critical" and others "noncritical." Critical errors, which usually resulted from missing data (e.g., serving size, frequency of consumption), precluded the inclusion of the nutrients from that particular food item in the participant's record. Noncritical errors included those with no effect on the computation of nutrients. For example, failure to record the time that an interview ended would be an error in form completion but would have no effect on the level of nutrients calculated for that participant's record.

The critical form completion error rate was calculated using the following formula:

total number of critical errors Critical Form Completion Error Rate =

total number of food items

where the denominator is restricted to the number of food items reported. The other components of the denominator for the overall form completion error rate are excluded because missing data for these fields are noncritical errors.

Transcription and Coding Errors Transcription and coding errors occur during data collection [18]. Food

items on the CARDIA Dietary History were precoded to the NCC database codes [19,20] to decrease discrepancies in coding decisions and to reduce transcription errors. In addition, a precoded screen was developed for data entry to reduce potential transcription errors. Although most food items on the history were precoded, additional foods and brand names could be entered on the form and the data entry screen. Thus, precoding reduced, but did not eliminate, the potential for coding and transcription errors.

CARDIA's manual and computerized QC efforts did not prevent typographic errors in data entry of the amount, serving size, or frequency of consumption (unless such errors resulted in omissions or out-of-range values; see next section). Double keying of the CARDIA dietary history was not done due to the added expense of duplicate entry of this lengthy form. To estimate the magnitude of typographic errors, the data files for the records of the 153 audiotaped interviews were compared to the information recorded on the forms.

Automated Editing Automated editing was performed at the time of data entry using the Viking

Forms Manager [21] modified by the C programming language [22]. Programs were designed to validate participant identification numbers, to require entry


of serving size, f requency, etc., for any food i tem repor ted by the part icipant and to apply range checks on the legality of serving units (e.g., cups, ounces, teaspoons, tablespoons), f requency units (e.g.; day, week, month) , and food preparat ion and fat codes associated with the NCC coding system.

Automated QC procedures were assessed by compiling the f requency of different errors ("critical" and "noncritical"), the total n u m b e r of food i tems reported, and the number of i tems processed with and wi thout errors for each part icipant record as the nutr ient files were created.

RESULTS

Interviewer Effect

In reviewing 153 audio taped interviews with their cor responding dietary history, the mean number of recording errors was 1.3 per audio tape form (SD = 1.4); 36.6% of the forms had no recording errors and 45.7% had one or two errors. Deviations from protocol in the conduct of the interview oc- curred more frequently, with a mean of 2.7 per tape form (SD = 2.3); 15.7% of the interviews had no deviations from protocol and 41.2% had one or two deviations. Overall, the total number of discrepancies (recording errors and deviations from protocol) per tape form was 4.1 (SD = 2.9). The mean discrepancy rate was 0.7%. Four (3.3%) of the 153 interviews were rated as unsatisfactory, indicating a total of more than 10 errors. Since interviewer errors did not differ significantly among clinics, clinic-specific results are not presented.

Missing Data

Table 3 summarizes the results of the form complet ion review. Of the 717 forms reviewed, 563 (78.5%) had no form complet ion errors; an additional 13.3% of forms had one error (Figure 2). The mean form complet ion error rate was 0.22% (SD = 0.52%). The highest error rate found was 4.14%, cor responding to six errors in 145 responses (responses equal to the sum of

Table 3 Form Comple t ion a Results

Number of forms reviewed 717 b Number (%) of error-free forms:

All errors combined 563 (78.5%) Critical errors 663 (92.5%)

Mean error rate ( + SD). All errors combined 0.22% (± 0.52%) Critical errors 0.12% (± 0.50%)

Range: All errors combined 0-6 errors Critical errors 0-5 errors

Highest error rate 4.14%

aForm completion deals only with the completeness of recorded information and in no way reflects accuracy of recorded information. bForm completion data was obtained by combining information from bimonthly form set checks (n = 564) and the monthly nutrition quality control efforts (n = 153).


100

80

60

40

20

Percent of F o r m s

92.5

B

o 1

13.2

5.0 4 . 3 1 1 1 . 5 1 . 1 1.7 0.1 0.6 0,1 0.1 0

2 3 4 5 6

N u m b e r o f Errors

All Errors ~ "Critical" Errors

Figure 2 Percent of dietary history forms by number of form completion errors.

food i tems repor ted , sections m arked "no , " ques t ions regard ing die tary practices, and quest ions used to assess the interview). Again, there were no significant differences in error rates be tween clinics.

Of the total 264 errors found in fo rm review, 61 (23.1%) were cons idered critical because they wou ld impac t on the computa t ion of nutr ients . The m e a n critical fo rm comple t ion error rate is 0.12%; 7.5% of forms had one or more critical errors (Table 3). Figure 2 shows the p ropor t ion of fo rms wi th critical errors; f requencies of the specific critical errors are p rov ided in Table 4. The mos t f requent critical error was failure to record the serving size and a m o u n t for a food i tem repor t ed b y the participant; this accounted for 47.5% of such errors.

Ana lyses were conduc ted to de te rmine t ime t rends and differences between in terv iewers in the p ropor t ion of error-free forms. The p ropor t ion of error-free fo rms increased over four t ime per iods for all errors as well as

Tab le 4 Form Complet ion: N u m b e r of Critical a Errors Ident if ied Dur ing Form Review

Number of forms reviewed 717

Type of Error 1. Serving size and amount 29 2. Frequency in day, week, or month column 14 3. Preparation and fat codes 11 4. Codes for brand name food items 5 5. Codes for food items marked with an asterisk 2

TOTAL 61

aA critical error is one that has a direct impact on the computation of nutrients. Values reflect the number of times the items were not recorded on a form.


critical errors (Table 5). X 2 t e s t s for linear t rend were highly significant for all errors and for critical errors (p < 0.0001).

The overall propor t ion of error-free forms did not differ significantly among interviewers (Table 6). The four nutri t ionists with fewer than 20 forms dur ing the examinat ion were excluded from analyses because of the small n u m b e r of observations; three of the four nutri t ionists were locally rather than centrally trained. The propor t ion of error-free forms was general ly lower among the locally trained interviewers.

Errors Identified during Data Processing

Table 7 shows the total numbe r of food items and the percent of error-free forms for selected errors identified in compiling the nutr ient files dur ing data processing (i.e., compute r ized check of the data). Overall, nearly one-half million food i tems were processed with few errors being detected. (As pre- viously stated, this error rate does not include typographic errors made dur ing data entry.) The n u m b e r of food items in error varied with the type of error, ranging from 3 (0.001%) to 981 (0.21%); the percent of error-free records ranged from 85% to 99.9%.

The most f requent error detected dur ing processing was " a m o u n t eaten out-of-range." These errors (981, or 0.21%) were detected via the NCC database default range checks used to identify an a m o u n t as excessive. The second most f requent error was " food code missing" (385, or 0.08%).

The data ent ry error rate per form (fields in error/total fields) ranged from 0 to 6.0% with a mean data en t ry error rate of 0.68%. The mean n u m b e r of fields per form was 593 (range 187-1123) and the mean n u m b e r of fields in error was 3.8 (range 0-29). There were 55.6% of forms with no data ent ry errors.

DISCUSSION

Results of dietary research are of ten criticized because accurately est imating usual dietary intake is conceived to be more difficult than measur ing physiological variables such as blood pressure. Most epidemiological studies mus t

Table 5 Form Complet ion: Percent Error-Free Forms by Qual i ty Control Period a

Quality Control % Error-Free Forms Period n All Errors b Critical Errors b

1 65 43.1 81.5 2 181 64.6 85.6 3 195 87.2 95.9 4 276 89.9 97.5

aQuality control periods: 1 = April-June 1985; 2 = July-September 1985; 3 = November 1985- January 1986; 4 = March-May 1986. b×2 test for trend, p < 0.0001.


Table 6 Form Completion: Percent Error-Free Forms by Interviewer

% Error-Free Forms

Interviewer n All Errors Critical Errors

Center 1 1 45 2 45 3 59 4 a 15

Center 2 5 67 6 46 7 40

Center 3 8 74 9 63

10 41 11 a 17

Center 4 12 53 13 73 14 ~ 17 15 51 16 a 10

80.0 93.3 84.4 95.6 88.1 96.6 73.3 80.0

80.6 95.5 67.4 93.5 80.0 95.0

77.0 89.2 81.0 92.1 73.2 80.5 64.7 82.4

83.0 94.3 80.8 95.9 58.8 94.1 76.5 92.2 70.0 90.0

×2 = 9.74 X 2 = 15.29 n.s. n.s.

aExcluded from X z test for homogeneity because n - 20.

rely on est imated intake by recall or food records because it is nei ther logis- tically feasible nor cost-effective to record dietary intake directly am o n g free- living individuals. Potential sources of error inherent in the collection of such data include interviewer effect, missing data, and transcription and coding errors. The QC efforts for dietary data collection implemen ted dur ing the first CARDIA examinat ion were deve loped to minimize these potential sources of error. T-he CARDIA s tudy used a number of QC methods , including auditory, visual, and au tomated inspection of the data; the results repor ted indicate that the dietary data in CARDIA were sufficiently complete and accurate for the purposes of analysis of risk factor data.

While the process of evaluating discrepancies in recording data and interviewing me thods was more subjective and time consuming than that used for form completion, it was an impor tant aspect of QC. It p rovided a mech- anism for feedback from the nutri t ion coordinator to each nutrit ionist , par- ticularly with regard to conduct of interview, which is especially impor tan t whe n interviews are being conducted by m an y interviewers at several loca- tions. The total of 4.1 discrepancies per taped form was low and yielded a discrepancy rate of 0.7%. Participants were randomly selected for taped interviews; interviewers had no control over the participant 's responses or the part icipant 's reaction to the interview. Because nutrit ionists were aware that interviews were being taped, they could alter their behavior accordingly; therefore, some underes t imat ion of the error m a y be present .

To estimate the degree of bias present in the discrepancy rate, the form

166

Table 7

J.E. Hilner et al.

Selected Errors Identified in Data Processing of the Dietary History for the Entire CARDIA Cohort

Total Number (%) % of Food Items a Error-Free

in Error Records n = 474,164 n = 5111

Food code missing 385 (0.08) 94.1 Serving size missing 78 (0.02) 98.8 (amount eaten) Serving unit missing 3 (0.001) 99.9 (ounces, cup, etc.) Frequency missing 115 (0.02) 98.1 (number of times) Frequency unit missing 114 (0.02) 98.1 (day, week, or month) Preparation code missing 9 (0.002) 99.8 Fat code missing 148 (0.03) 97.3 Amount eaten out-of-range 981 (0.21) 85.0

aVitamin supplements excluded from analysis.

completion error rate for the 153 forms reviewed in conjunction with the audiotapes was compared with that from the 564 forms reviewed as a random sample of all forms. The mean form completion error rate for the 153 forms (0.31%) was significantly greater than the mean error rate for the 564 forms (0.19%, p = 0.0351), indicating that interviewers did not perform better on forms that they knew would be reviewed. Perhaps interviewers were more nervous during taped interviews. To avoid questions of bias, all interviews would have to be taped and a random selection reviewed. Because such a procedure would be costly, it may not be feasible in large studies; smaller studies may consider this option.

The mean form completion error rate of 0.22% is considerably lower than the 3% error rate used in the certification criteria (Table 2), indicating good quality data with regard to completion. The error rate for critical errors is even lower (0.12%), which means that the majority of form completion errors did not adversely affect the computation of nutrients.

Automated computer-assisted editing during data entry was effective in minimizing errors; nearly one-half million foods were processed with few detectable errors. The percent of food items in error ranged from less than 0.01% to 0.21%, depending on the type of error. The most frequent error was the "amount eaten out-of-range," which was detected via the default range values residing on the NCC database. In many cases, the exclusion criteria for these values were quite conservative. All out-of-range values were reviewed by the Nutrition Working Group and food items with feasible amounts were included in the CARDIA database. The estimated data entry (field) error rate of 0.68% was reasonable given the length and complexity of the form. Data entry personnel were not centrally trained for the diet history data entry; a lower data entry error rate could have possibly been achieved with such training. It should also be noted that the actual keystroke error rate (not calculated) would be lower than the field error rate.

In an era of fiscal frugality, QC efforts are costly. The Coordinating Center


employed one full-time equivalent (nutrition coordinator) to assist the Nu- trition Working Group in the development of the form and the training manual, to implement the selected QC procedures, and to monitor and document changes to the database for the 14 months of data collection. The development of the form by the Nutrition Working Group required approximately 6 months. The use of registered or registry-eligible nutritionists at each clinic also added expense. The development of the data entry system and the automated editing procedures required one full-time equivalent for 4 months.

Determining the benefit, given the cost of QC efforts, is difficult because we are unable to define the impact of not doing QC. Although we believe that the QC methods helped assure high quality, no direct evidence is avail- able to document that the QC methods used lead to higher quality dietary interview data than would have been obtained without these QC methods. Similarly, we cannot determine whether any of the relationships being in- vestigated in CARDIA would be weakened or reversed if we had done no QC.

The significant positive linear trend in the proportion of error-free forms over time is consistent with our conviction that the feedback from the nutrition coordinator to the nutritionists had some effect in decreasing form completion errors. While one could hypothesize that errors would increase as interviews were conducted further from the time of training, it could also be hypothesized that errors would decrease with more interviewer experience. We do not know which scenario is true; this makes some aspects of QC hard to evaluate. The absence of significant differences among clinics with regard to form completion or accuracy error rates is consistent with the belief that the centralized training, certification, and constant monitoring were effective in standardizing interviewing and documenting behaviors among clinics. The lower proportion of error-free forms among the locally trained interviewers suggests that centralized training is valuable. The homogeneity in the proportion of error-free forms among interviewers suggests that the QC methods were effective in minimizing inter-interviewer differences.

It would be useful to compare the CARDIA QC results with those of other large-scale studies in an attempt to determine the relative quality of the data. However, comparing dietary QC data is difficult due to the variety of meth- odologies used in studies; the methodology and budget influences the selection of appropriate QC efforts. It is important to note that the QC efforts selected and the results obtained in CARDIA will differ from studies using more open-ended dietary collection methods such as 24-hr recalls or food records.

We urge others involved in dietary data collection to share both their QC measures and the results of the monitoring techniques. Reporting the results of QC efforts is important for several reasons. First, results in which dietary intake is associated with physiological measures and/or risk factors for disease can be better understood if the quality of the data is known. Perhaps some of the conflicting results regarding the associations between diet and disease obtained in studies can be explained by differences in the quality of the data. Publishing the QC methods and results also demonstrates to physicians, epidemiologists, and statisticians the stringent efforts often employed in dietary data collection. This information should increase the confidence that


bo th the scientific c o m m u n i t y and the general public place in the results of we l l -pe r fo rmed dietary studies.

ACKNOWLEDGMENTS The authors acknowledge the hard work of the CARD1A nutritionists and their commitment to quality dietary data collection: Birmingham--Eva Ankrom, M.S., Karen Counts, R.D., Janet Raines, M.S., R.D.; Chicago---Lisa Royce Bullard, Niki Gernhofer, M.S., R.D., Fran Oppenhei- mer, R.D.; Minneapolis---Janice Cox, R.D., Diane Harper, R.D., Hilmar Wagner, R.D., Joyce Wenz, M.S., R.D.; Oakland--Shannon Daniel, R.D., Debra Duncan, M.P.H., Jennifer Macu- liewicz, R.D., Elaine Moquette, M.P.H., R.D., Jayne Weiss, M.P.H.

REFERENCES

1. Rothman K: Modern Epidemiology. Boston: Little, Brown, 1986 2. Tillotson JL, Gorder DD, DuChene AG, Grambsch PV, Wenz J: Quality control

in the multiple risk factor intervention trial nutrition modality. Contr Clin Trials 7 (Suppl):66S-90S, 1986

3. Van Beresteyn ECH, Van 'T Hof MA, De Waard H, Dekker PR, Neeter R, Win- keldermaat HJ, Visser RM, Schaafsma G, Van Schiak M, Duursma SA: Design and quality of a raixed longitudinal study to elucidate the role of dietary calcium and phosphorus on bone mineralization in pre-, peri-, and postmenopausal women. Am J Clin Nutr 43:538-548, 1986

4. Prudhomme GJ, Canner PL, Cutler JA: Quality assurance and monitoring in the hypertension prevention trial. Contr Clin Trials 10 (Suppl):84S-94S, 1989

5. Massie MD, Strife CF, Foreman JW, Chan JCM: Quality control of the nutritional component of the growth failure in children with renal diseases study. Pediatrics 116:$40-$45, 1990

6. Nicklas TA, Forcier JE, Webber LS, Berenson GS. School lunch assessment to improve accuracy of 24-hour dietary recall for children. J Am Diet Assoc 91:711- 712, 1991

7. Frank GC, Berenson GS, Webber LS: Dietary studies and the relationship of diet to cardiovascular disease risk factor variables in 10-year-old children--The Bo- galusa heart study. Am J Clin Nutr 31:328-340, 1978

8. Bulatti E, Palli D, Amadori D, Marubini E, Puntoni R, Avellini C, Bianchi S, Cipriani F, Cocco P, Decarli A, Vindigini C, Blot W: Methodological issues in a multicentric study of gastric cancer and diet in Italy: Study design, data sources and quality controls. Tumori 75:410-419, 1989

9. Slattery ML, Abbott TM, Overall JC Jr, Robison LM, French TK, Jolles C, Gardner JW, West DW: Dietary vitamins A, C, and E and selenium as risk factors for cervical cancer. Epidemiology 1:8-15, 1991

10. Hughes GH, Cutter G, Donahue R, Friedman GD, Hulley SB, Hunkeler E, Jacobs DR Jr, Liu K, Orden S, Pirie P, Tucker W, Wagenknecht LE: Recruitment in the coronary artery disease risk development in young adults (CARDIA) study. Contr Clin Trials 8:68S-73S, 1986

11. Friedman GD, Cutter GR, Donahue RP, Hughes GH, Hulley SB, Jacobs DR, Liu K, Savage RJ. CARDIA: Study design, recruitment, and some characteristics of the examined subjects. J Clin Epidemiol 41:1105-1116, 1988

12. Cutter GR, Burke GL, Dyer AR, Friedman GD, Hilner JE, Hughes GH, Hulley SB, Jacobs DR, Jr, Liu K, Manolio TA, Oberman A, Perkins LL, Savage PJ, Serwitz JR, Sidney S, Wagenknecht LE: Cardiovascular risk factors in young adults: The CARDIA baseline monograph. Contr Clin Trials 12 (Suppl):1S-77S, 1991

13. McDonald A, Van Horn L, Slattery M, Hilner JE, Bragg C, Caan B, Jacobs DR,


Liu K: The CARDIA dietary history: Development and implementation. J Am Diet Assoc 91:1104-1112, 1991

14. Burke BS: The dietary history as a tool in research. J Am Diet Assoc 23:1041-1046, 1947

15. Shekelle RB, Shryock AM, Paul O, Lepper M, Stamler J, Liu S, Raynor WJ Jr: Diet, serum cholesterol, and death from coronary heart disease. N Engl J Med 304:65-70, 1981

16. Liu K, Slattery ML, Jacobs D Jr, Cutter G, McDonald A, Van Horn L, Hilner JE, Caan B, Bragg C, Dyer A, Havlik R: A study of the reliability and comparative validity of the CARDIA dietary history (submitted)

17. Abramson JH: Survey Methods in Community Medicine. New York, Churchill Livingstone, 1979

18. Witschi JC: Short-term dietary recall and recording methods. In Willett W (ed). Nutritional Epidemiology. New York, Oxford University Press, 1990

19. Dennis B, Ernst N, Hjortland M, Tillotson J, Grambsch V. The NHLBI nutrition system. J Am Diet Assoc 77:641--647, 1980

20. Sievert YA, Schakel SF, Buzzard IM. Maintenance of a nutrient database for clinical trials. Controlled Clin Trials 10:416-425, 1989

21. Viking Forms Manager User Manual: Version 2.21. Tulsa, Oklahoma, Viking Soft- ware Services, 1985

22. Lattice C Manual: Revision 2. New York, Lifeboat Associates, 1982

quality control of dietary data collection in the cardia study

Documents