9
. J Clh Epibmbl Vol. 43, No. 12, pp. 1327-1335, 1990 0895-4356/90 s3.00 + 0.00 Printcd in GrcatBritain Pergamon Prcw pk VALIDATION OF A SELF-ADMINISTERED DIE? HISTORY QUESTIONNAIRE USING MULTIPLE DIET RECORDS GLADYS BLOCK,‘** MARGO WOODS,~ ARNOLD POTOSKY’ and CAROLYN CLIFRXD’ ‘Division of Camxr Prevention and Control, National Canccr Institute, Bethesda, MD 20892 and Wommunity Health/Nutrition Unit, Tufts University School of Medicine, Boston, Massachusctts, U.S.A. (Recehed in reuisedform 31 Muy 1990) Abstract-The validity of a self-administered diet history questionnaire has been estimated using as the reference data the mean of three 4-day diet records collected over the year prior to the administration of the questionnaire, in 19851986. Subjects were women ages 45-70 years, participants in the Women’s Health Trial Feasibility Study, a multi-center clinical trial in which some women were randomized to be taught to adopt and maintain a low-fat diet, while ethers maintained their usual diet. The questionnaire produced group mean nutrient estimates closely approzimating the values obtained by three 4-day records, e.g. in the usualdiet group, 37.7% of calories from fat by both food records and by questionnaire, and in the low-fat, group, 21.3% of calories from fat by food records and 23.7% of calories from fat by questionnaire. Correlations between questionnaire and diet records for per cent of calories from fat were 0.67 and 0.65 respectively in the two groups; most correlations were in the 050.6 range, and were similar to those achievable by a single 4-day food record. Epidemiologic methods Diet Nutrient intake Assessment Fat Vitamins INTRODUCMON To study the role of diet in disease prevention, investigators need nutríent estimates that accu- rately reflect an individual’s long-term usual intake. This can be achieved through multiple non-consecutive diet records or recalls, but the number of records required resúlts in great expense even for the macronutrients. For many of the micronutrients, the number of records required makes this approach completely infeas- ible [ll. Consequently, numerous investigators have developed diet history or frèquency ques- tionnaires for use in epidemiologic research. We report here the performance of the self- administered questionnaire developed by Black *Reprint requests should be addressed to Dr Gladys Black, NCI, Executivc Plaza North; Rm 313, 9000 Rockville Pike, Bethesda, MD 20892, U.S.A. et al. [2], among the two groups of women, one group maintaining their usual North American diet and the other group consuming a low-fat diet. The reference or “gold standard” data consist of three Cday diet records collected over the prior 1 year period. The development of the questionnaire and its nutrient database’ have been presented earlier [2], and an interview and analysis program for personal computers is available [3]. Previous published validations of this questionnaire at the individual leve1 have included its ability to assess calcium intake in elderly women, by interview 141,and its ability to assess dietary intake 10-15 years in the past, by interview and mail [5]. METHODS Subjects were participants in the Women’s Health Trial (WHT) Feasibility Study [CS], 1327

Block 1990

Embed Size (px)

DESCRIPTION

Block 1990

Citation preview

Page 1: Block 1990

.

J Clh Epibmbl Vol. 43, No. 12, pp. 1327-1335, 1990 0895-4356/90 s3.00 + 0.00 Printcd in Grcat Britain Pergamon Prcw pk

VALIDATION OF A SELF-ADMINISTERED DIE? HISTORY QUESTIONNAIRE USING MULTIPLE DIET RECORDS

GLADYS BLOCK,‘** MARGO WOODS,~ ARNOLD POTOSKY’ and CAROLYN CLIFRXD’

‘Division of Camxr Prevention and Control, National Canccr Institute, Bethesda, MD 20892 and Wommunity Health/Nutrition Unit, Tufts University School of Medicine, Boston, Massachusctts,

U.S.A.

(Recehed in reuisedform 31 Muy 1990)

Abstract-The validity of a self-administered diet history questionnaire has been estimated using as the reference data the mean of three 4-day diet records collected over the year prior to the administration of the questionnaire, in 19851986. Subjects were women ages 45-70 years, participants in the Women’s Health Trial Feasibility Study, a multi-center clinical trial in which some women were randomized to be taught to adopt and maintain a low-fat diet, while ethers maintained their usual diet. The questionnaire produced group mean nutrient estimates closely approzimating the values obtained by three 4-day records, e.g. in the usualdiet group, 37.7% of calories from fat by both food records and by questionnaire, and in the low-fat, group, 21.3% of calories from fat by food records and 23.7% of calories from fat by questionnaire. Correlations between questionnaire and diet records for per cent of calories from fat were 0.67 and 0.65 respectively in the two groups; most correlations were in the 050.6 range, and were similar to those achievable by a single 4-day food record.

Epidemiologic methods Diet Nutrient intake Assessment Fat Vitamins

INTRODUCMON

To study the role of diet in disease prevention, investigators need nutríent estimates that accu- rately reflect an individual’s long-term usual intake. This can be achieved through multiple non-consecutive diet records or recalls, but the number of records required resúlts in great expense even for the macronutrients. For many of the micronutrients, the number of records required makes this approach completely infeas- ible [ll. Consequently, numerous investigators have developed diet history or frèquency ques- tionnaires for use in epidemiologic research.

We report here the performance of the self- administered questionnaire developed by Black

*Reprint requests should be addressed to Dr Gladys Black, NCI, Executivc Plaza North; Rm 313, 9000 Rockville Pike, Bethesda, MD 20892, U.S.A.

et al. [2], among the two groups of women, one group maintaining their usual North American diet and the other group consuming a low-fat diet. The reference or “gold standard” data consist of three Cday diet records collected over the prior 1 year period. The development of the questionnaire and its nutrient database’ have been presented earlier [2], and an interview and analysis program for personal computers is available [3]. Previous published validations of this questionnaire at the individual leve1 have included its ability to assess calcium intake in elderly women, by interview 141, and its ability to assess dietary intake 10-15 years in the past, by interview and mail [5].

METHODS

Subjects were participants in the Women’s Health Trial (WHT) Feasibility Study [CS],

1327

Page 2: Block 1990

1328 GLADYS BLOCK et al.

which was designed to teach women to maintain a low-fat diet. For that study, 303 women aged 45-70 years were enrolled from three geographic sites, Seattle, Wash., Houston, Tex., and Cincinnati, Ohio. Participants were ran- domly assigned to a low-fat or a usual-diet group. The usual-diet group maintained their usual diets, which contained approximately 38% of calories from fat; the low-fat group received intensive education to bring their diets to approximately a 20%-fat level. For the pre- sent validation study, 277 women provided both the reference dietary data described further below, and also completed an extensive self-administered diet history questionnaire (“Black” questionnaire) 1 year after enrollment into the Feasibility Study; 260 met edit criteria described below, and comprise the validation study population. The age, race and education characteristics of these study participants are shown in Table 1.

The reference data

The reference data, the measure of “truth” against which the questionnaire estimates are compared, consist of multiple 4-day diet records. Four-day records were obtained from al1 subjects at the time of their entry into the study, and again 6 months and 1 year later. In addition 4-day records were obtained from low- fat group subjects at 3 months.

For analyses involving correlations, the aver- age of three 4-day diet records is used as the reference data. This is necessary since the correlation coefficient is sensitive to the intra-

Table 1. Characteristics of 260 female participants in a randomized low-fat diet trial, whose data were used to

validate the diet history questionnaire*

Usual-diet Low-fat group (%) group (%)

Characteristic n = 102 n = 158

Age 4549 yr 17.7 20.9 50-59 yr 51.0 53.2 60-70 yr 31.4 26.0

Race White 99.0 96.8 Nonwhite 1.0 3.2

Educa tion High school or less 28.4 39.9 Some college 26.5 26.0 College graduate 45.1 34.2

*By design, the Women’s Health Trial Feasibility Study enrolled more participants into the low-fat group. There are no statistically significant (p c 0.05) differences be- tween usual-diet and low-fat groups in any of these characteristics.

individual variability in the reference data, and three 4-day diet records provide a better and more stable estimate of usual diet than do two. For analyses involving group means, the ques- tionnaire estimates are compared with food record estimates from the same ó-month period covered by the questionnaire.

Subjects were instructed by trained nutrition- ists in careful recording of their food intake for the 4-day records, with emphasis on accurate measurement of food portions and documen- tation of brand names, cuts of meat, and food preparation. For mixed dishes, detailed recipes were to be recorded. The 4-day records were checked by study nutritionists, and questionable responses clarified. Coding and nutrient analysis was performed by nutritionists who had com- pleted a comprehensive training course, and who used a computerized coding system devel- oped by the Nutrition Coordinating Unit at Tufts University. The Tufts Nutrient Data Bank consists of approximately 500 food items, which allows for specificity in coding [9], and the primary nutrient data source is the USDA Nutrient Data Base for Standard Reference, Release 5, 1984.

A few errors were detected in nutrient esti- mates from the 4-day records. For three sub- jects, the 4-day record containing the error was ignored, and the estimate was based on the remaining two 4-day records.

The diet history questionnaire

The questionnaire, a modification of the “Black” questionnaire [2], was self-administered by the respondents 1 year after enrollment into the Feasibility Study. Respondents report how often they usually consume each food, as number of times per day, week, month or year, and whether their usual portion size of that food is small, medium or large with respect to a stated medium portion. The questionnaire also contains the following ancillary questions: frequency of consumption of foods from six types of restaurants; whether skin is usually eaten on poultry, or fat on meat; frequency of use of fat or oil in cooking, and types used; two genera1 questions on the overall frequency of consumption of fruits and of vegetables. Age-sex-specific portion size assumptions are used in the calculations. The impact of using age-sex-specific and individually variable as opposed to standard medium portion sizes is examined below. Vitamin supplement use is also collected, and results below report nutrient

Page 3: Block 1990

Validation of a Diet Qu&onnaire 1329

estimates both excluding and including vitamin supplements as estimated by the questionnaire.

For the WHT Feasibility Study, the question- naire was modified as follows. Nine food items were omitted from the standard “Black” list, and 5 were added as relevant to low-fat group, resulting in a food list of 94 food items. The open-ended section where a respondent may volunteer additional foods was omitted. Five ancillary questions were added, regarding lean- ness of beef or hamburger, dark or light meat poultry, water-pack or oil-pack tuna, brand name of yogurt usually eaten, and type of lunch meats or hot dogs usually eaten. These ques- tions were added in the expectation of improv- ing the precision of the fat intake estimates, particularly for the low-fat group. Finally, the respondent was asked to report on her diet over the past 6 months rather than the past year. The added and omitted items and the analysis pro- gram options in effect for this analysis are listed in the Appendix. The foods comprising the standard food list may be found in Ref. [2].

Coding and keying of the questionnaires was conducted by a centra1 Statistical Coordinating Unit at Fred Hutchinson Cancer Research Center. Complete rechecking of the coding as wel1 as double-keying was performed for 100% of the questionnaires. In addition to standard range-edits, computerized edits identified ques- tionnaires with errors suggesting careless re- sponses, such as 16 or more foods skipped. Of the 277 questionnaires, 17 or 6% were identified by a standard computer algorithm as having serious response errors, and were omitted from further analyses, leaving a study group of 260 individuals. Errors identified by the program are specified in the Appendix.

Nutrient analyses were calculated using the computer program and nutrient database devel- oped for the questionnaire [2,3]. NO attempt was made to adjust the standard “Black” data- base to make it correspond to the nutrient database or default portion sizes used for analy- sis of the Cday records. Correlation between estimates based on the two different databases must therefore be an underestimate of the corre- lations which would be observed if the two databases were identical.

Statistical methods

Means, standard deviations and Pearson cor- relation coefficients are reported for the nutri- ents estimated by the questionnaire and by multiple 4-day records. Except for percent of

calories Erom fat, al1 nutrient estimates were skewed, and were therefore transformed to achieve normality, using the transformation which most reduced skewness and improved normality.

RFSULTS

In Table 2, al1 nutrient estimates by question- naire are within f 20% of the estimates pro- duced by the mean of the four-day records. Indeed, in the usual-diet group 14 of 18 dietary intake estimates by questionnaire differ by 10% or less from the diet record estimates, and the same was true for 13 of 18 in the low-fat group. This is’particularly notable in view of the fact that the low-fat and usual-diet groups differed dramatically in the absolute value of some of their macronutrient intakes. For example, fat intake in the low-fat group was half that in usual-diet group, and this was reflected very wel1 by the questionnaire. Similarly, the major differ- ence in percent of calories from fat between low-fat and usual-diet are reflected wel1 by the questionnaire. For vitamins A and C, both with and without supplements, the questionnaire estimates are not significantly different from the mean of the food records, in both groups. For several other micronutrients, however, the ques- tionnaire estimates are statistically significantly different from the food record estimates, although not as much as 20% lower or higher, as indicated above.

Table 3 shows the mean percent of calories from fat, as estimated by the food records, within quintiles of the questionnaire estimate. In the usual-diet group the mean intake among those in the lowest quintile of the questionnaire is 3 1.5% of calories from fat. By contrast, in the lowest quintile of the low-fat group the mean intake is only 16.6%, and the highest quintile in that group consumed only 26.3% of their calo- ries from fat.

Correlations betwecn questionnaire and the mean of three 4-day records are shown in Table 4. Correlations for nutrients excluding supplements range from r = 0.47 (for vitamin A) to r = 0.67 (for percent of calories from fat) in the usual-diet group, and from r = 0.37 to r = 0.66 in the low-fat group, with an average correlation of about r = 0.55 in both groups. Correlations were algo examined among al1 responders, including those who were excluded because they had skipped too many foods, etc. (Data not shown.) In this population, their

Page 4: Block 1990

1330 GLADYS BLKICK et al.

Table 2. Group mean (and SD) as estimated by diet history questionnaire and by mean of two 4-day diet records collected at 6 months and 1 year, among 260 women ages 45-70 years

Usual-diet group Low-fat group n = 102 n = 158

Questionnaire Mean

Foo&znords Questionnaire Food records Mean Mean

(SD) (SD) (SD) (SD)

% Calories from fat

Calories (kcal)

Fat (g)

Saturated fat (g)

Monounsaturated fat (g)

Polyunsaturated fat (g)

Cholesterol (mg)

Protein (g)

Carbohydrate (g)

Vitamin A (KJ)

with supplements

Vitamin C (mg)

with supplements

Calcium (mg)

with supplements

Iron (mg)

with supplements

Thiamin (mg)

with supplementst

Riboflavin (mg)

with supplementst

Phosphorus (mg)

Potassium (mg)

Sodium (mg)

37.7 (8.9)

1509.7 (544.2)

63.8 (33.5) 21.7

(11.6) 22.9

(12.8) ll.4 (8.0)

239.6 (144.3)

63.4 (21.6) 162.4 (58.0)

7458.6 (3523.0) 10827.7 (4921.8)

123.5 (58.6) 431.5

(480.4) 819.0

(382.9) 1191.6

(455.0) 10.6 (3.5) 18.7

(10.3)

(Z)

({.$

(2

(4:3) 1141.9 (423.5) 2518.3 (875.9) 2554.6 (895.8)

37.7 (6.8)

1559.7 (326.9

(Z3 23.1 (8.5) 23.5 (8.2) 12.1 (4.6)

267.2 W;)

(13.3) 175.9 (40.8)

6931.5 (3232.4) 10495.8 (7704.3)

128.5 (56.4) 379.9

(378.0) 704.9*

(254.0) 1063.4 (438.8)

12.2* (2.9) 21.5

(12;)

(0.3;

(2:::;

(0::;

(1:::) 1051.1 (247.2) 2407.9 (567.9) 2430.6 (613.8)

23.7 (7.1)

1304.2 (351.2)

33.5 (15.0) 11.3 (5.8) ll.7 (5.8)

(3::) 145.8 (z;.;)

(18:4) 179.2 (55.3)

7316.4 (3025.4) 10769.5 (7120.6

142.8 (69.1). 486.2

(674.5) 889.2

(411.7) 1136.1 (413.3)

11.1 (3.2) 18.7 (9.8)

(0::)

(E)

(i$

(5.4) 1183.6 (408.2) 2761.4 (788.6) 2375.6 (722.2)

21.1* (5.6)

1318.8 (287.7)

31.1 (10.9) 10.1* (4.1) 10.4* (4.1)

,:::; 155.0 (66.0) 64.0

(14.7) 195.6* (50.6)

7273.4 (3566.1) 9891.9

(6200.1) 142.1 (59.5) 372.1

(485.5) 756.4*

(314.1) 1117.5 (509.5)

12.5’ (3.6) 28.1’

(52.9) 1.3*

(0.4)

(1%;

,:::; 6.6

(15.1) 1064.5* (313.4) 2627.4 (690.5) 2179.8* (536.4)

*p c 0.05 for differente between the food record and questionnaire estimates in the group shown. tFive individuals reported on their diet records that they took over 75 mg of those B vitamins on the day of the record, resulting in very high group means, but did not indicate on the questionnaire that they took them every day over the 6-month period. The median values estimated for those B vitamins with supplements are quite similar by questionnaire and food record (1.9 VS 2.1 mg for thiamin with supplements, and 2.7 VS 2.4 for riboflavin with supplements.)

exclusion resulted in only slight improvements the low-fat group. Calorie-adjusted data are not in the correlations. presented in Table 4 nor in subsequent tables.

Correlations using calorie-adjusted data Agreement between questionnaire and refer- [10] were also examined. The average of the ence data as reflected in cross-classification by correlation coefficients was increased by only quintiles corresponds extremely closely, empiri- 0.02 (i.e. from 0.55, excluding calories, to 0.57) cally, to what would be predicted from their in the usual-diet group, and was unchanged in correlation coefficients [ 111, for most nutrients

Page 5: Block 1990

Validation of a Diet Questionnaire 1331

Table 3. Group mean (and SD) of percent of calories from fat as estimated by the mean of two 4day diet records collected at 6 months and 1 year, .within quintiles of the diet history questionnaire estimate of percent of caloriw from

fat; 260 women ages 65-70 years

Quintile of Food records percent of calories from fat diet history Mean (SD)

questionnaire estimate of Usual-diet group Low-fat group percent fat n = 102 n = 158

1 31.5 16.6 (7.1) (3.9)

2 35.7 19.5 (3.8) (3.8)

3 37.1 20.1 (5.6) (4.5)

4 40.6 23.1 (5.9) (4.8)

5 43.4 26.3 (4.8) (5.4)

(data not shown). However, when the contribu- tion of supplements is included in the vitamin- mineral estimates, the agreement of the questionnaire with 4-day records is considerably better than would be expected from their corre- lation coefficients. For vitamin C, 57% are in the exactly correct quintile, and 87% are either in the exact or & one quintile. When only the top and bottom quintiles are considered, exact agree- ment ranged from 40% (for iron including supplements) to 65% (for vitamin C including supplements); and from 68 to 84% for exact or adjacent agreement. It should be noted, how- ever, that vitamin supplement use varies in differ- ent demographic subgroups, and therefore not al1 groups wil1 have this degree of agreement.

Table 5 addresses the question, “How good are the correlations seen in Table 4, in compari- son with what could be achieved if a single Cday record were used as the estimator of dietary intake?’ For this analysis, the first two 4-day records (baseline and 6 months in usual-diet, and 3 months and 6 months in the low-fat group) are used as the reference data. Against this slightly different measure of truth, the per- formance of the questionnaire administered at 1 year is compared with the performance of the 4-day record administered at 1 year. In the usual-diet group, use of a 4day record to measure dietary intake would have resulted in correlations in exactly the same range as those achieved by the questionnaire. Nine of the 18 correlations in the usual-diet group are higher by questionnaire, 9 are higher by 4-day record, and the median correlation is the same by both methods.

Table 4. Correlations between nutrient eatimates by diet history questionnaire and by mean of three 4-day diet

records, among 260 women ages 45-70 years*

Usual-diet Low-fat group group

n = 102 n =158

Percent calories from fat 0.67 0.65 Calories (kcal) 0.51 0.51 Fat (B) 0.57 Saturated fat (9) z.53 0.59 Monounsaturated fat (g) 0:59 0.58 Polyunsaturated fat (9) 0.48 0.46 Cholesterol (mg) 0.55 0.56 Protein (g) 0.48 0.56 Carbohydrate (g) 0.51 0.55 Vitamin A (IU) 0.47 0.37

including supplements 0.55 0.58 Vitamin C (mg) 0.56 0.48

including supplements 0.71 0.74 Calcium (mg) 0.56 0.62

including supplements 0.55 0.64 Iron (mg) 0.47 0.44

including supplements 0.54 0.55 Thiamin (mg) 0.57 0.47

including supplements 0.47 0.62 Riboflavin (mg) 0.63 0.62

including supplements 0.57 0.63 Phosphorus (mg) 0.59 0.66 Potassium (mg) 0.55 0.60 Sodium (mg) 0.47 0.43

Mean 0.545 0.540 Median 0.555 0.560

*Data were transformed using log or square root to reduce skewness and improve normality, as required by the stat- istical assumption of tests related to the cormlation co- efficient. Al1 correlations shown are significantly greater than zero at p < 0.01.

The same is not true in the low-fat group, where the single Cday food record produced consistently higher correlations than the questionnaire. This was not primarily because the questionnaire produced lower correlations in the low-fat group, but because the 4-day record in that group achieved higher corre- lations with the earlier food records (median r = 0.605 compared with median r = 0.505 in the usualdiet group). This is not due to greater consistency of dietary intake in the low-fat group, since standard deviations were similar in the two groups. It is likely that it is a result of the low-fat group’s intense daily training and practice, over the previous year, in the exact foods, brand names and portion sizes they consumed.

Thus, it would appear that for a population consuming a “usual” American diet, this ques- tionnaire is approximately as good as a 4-day record at ranking people with respect to their usual diet. On the other hand, for a population such as the low-fat group here, intensely intervened on and taught to watch and record

Page 6: Block 1990

1332 GLADYS BLOCK et al.

Table 5. Comparison of correlations achieved by a diet history questionnaire and those achieved by a single 4-day record, when both are measured against the same reference data,

the mean of two prior diet records; women ages 45-70 years

Usual-diet group Low-fat group n = 97t n = 145t

Food record Questionnaire1 Food record Questionnaire1

Percent calories from fat 0.53 0.56 0.66 0.56 Calories (kcal) 0.50 0.47 0.65 0.43 Fat (g) 0.44 0.50 0.63 0.48 Saturated fat (g) 0.48 0.51 0.59 0.51 Monounsaturated fat (g) 0.42 0.49 0.57 0.47 Polyunsaturated fat (g) 0.41 0.45 0.62 0.37 Cholesterol (g) 0.49 0.40 0.52 0.49 Protein (g) 0.58 0.40 0.57 0.50 Carbohydrate (g) 0.57 0.52 0.69 0.48 Vitamin A (KJ) 0.52 0.41 0.45 0.36 Vitamin C (mg) 0.48 0.57 0.58 0.40 Calcium (mg) 0.63 0.54 0.73 0.57 Iron (mg) 0.32 0.42 0.51 0.34 Thiamin (mg) 0.43 0.57 0.48 0.37 Riboflavin (mg) 0.51 0.58 0.66 0.53 Phosphorus (mg) 0.61 0.53 0.69 0.61 Potassium (mg) 0.56 0.54 0.70 0.56 Sodium (mg) 0.56 0.42 0.41 0.31

Mean 0.50 0.49 0.60 0.46 Median 0.505 0.505 0.605 0.48

tOnly those women who provided al1 three diet records are included. SCorrelations are lower than those seen in Table 4 because reference data comprised of two 4-day records constitutes a less adequate estimate of usual intake than does three 4-day records; and because the baseline record was weighted more heavily in a two-record average than in a three-record average.

their intake, a 4-day record appears to produce somewhat better correlations than the question- naire.

Methodologie issues

with reference data in the usual-diet group. In the low-fat group, for whom such questions were relevant to their low fat diet, the added questions improved the correlations only to a trivial extent.

The effect of omitting the questions added for The standard questionnaire contains two the WHT version (e.g. leanness of meat usually questions on the overall frequency of consump- eaten) is shown in Table 6, for macronutrients tion of fruit and of vegetables. These permit and fats only. (The correlations for the remain- the adjustment of total fruit and vegetable ing nutrients were unchanged.) The added ques- consumption downward to reduce the over- tions had essentially no effect on correlations estimates resulting from the use of a long list of

Table 6. Correlations between nutrient estimates by diet history questionnaire and by mean of three 4-day diet records: effect of including supplementary fat-relevant questions*; 260 women ages 45-70, participants in a randomized low-fat diet trial

Usual-diet group Low-fat groups n = 102 n = 158 ~~- - _~

Without With Without With extra extra extra extra

questions questions questions questions

Percent calories from fat 0.68 0.67 0.63 0.65 Calories (kcal) 0.50 0.51 0.50 0.51 Fat (g) 0.59 0.60 0.55 0.57 Saturated fat (g) 0.62 0.63 0.56 0.59 Monounsaturated fat (g) 0.59 0.59 0.56 0.58 Polyunsaturated fat (g) 0.48 0.48 0.45 0.46 Cholesterol (mg) 0.55 0.55 0.55 0.56 Protein (g) 0.46 0.48 0.54 0.56 Carbohydrate (g) 0.52 0.51 0.55 0.55

*Sec Appendix for supplementary questions.

Page 7: Block 1990

Validation of a Diet Questionnaire 1333

foods, while retaining the reported distribution of these foods. These adjustments reduced the vitamin A overestimate by about 600 IUs (e.g. from 7970.9 IUs to 7458.6 IUs in usual-diet), and improved the correlation coefficients slightly [e.g. from 0.47 to 0.49 for vitamin A in usual-diet, and from 0.37 to 0.40 in low-fat (data not shown)].

The effect of portion-size assumptions was investigated by rerunning the analysis, setting the program to use standard rather than age-sex specific portion sizes, and to ignore the respon- dent’s “smal1-medium-large” and use only medium. In the low-fat group, 16 of 18 nutrients had higher correlations when individually vari- able and age-sex-specific portion sizes were used, and the average correlation was improved from 0.49 to 0.54. In several cases the improve- ments resulting from use of variable portion size were non-trivial; e.g. the correlation for calories improved from r = 0.40 to 0.51, and that for sodium improved from r = 0.29 to 0.43. In the usual-diet group, 12 of 18 nutrients had higher correlations using variable, age-sex-specific portion sizes, although the improvement in the average correlation was minor. (Data not show@.

Standard medium produces serious over- estimates of nutrient intake for vitamins A, C and fiber: for example, for vitamin A, 11,172 IUs VS 6840 by food records, in the usual-diet group. In the low-fat group, as well, the means produced by the use of variable portion size were considerably closer to the diet record values, while use of a standard medium produced substantial overestimates of virtually al1 nutrients. In an interview context, variable portion size as formatted here constitutes only 20% of interview time; in a self-administered format it is less.

Finally, a program option may be enabled, which recalculates the dietary estimates for respondents who are outliers with respect to calorie intake. In the usual-diet group, 16 of 18 correlations were improved, increasing the aver- age of the correlations from 0.55 to 0.58. In the low-fat group, however, 15 of 18 correlations were decreased, and the average of the corre- lations decreased from 0.54 to 0.53. (Data not shown.) Thus this option should be used with caution in study groups in which the true dietary intake may be shifted towards the low or high end of the intake distribution, but appears to improve nutrient estimates slightly, in genera1 population groups.

DISCU!3!3iON

This study has examined the relationship between nutrient estimates produced by a par- ticular diet history questionnaire and its associ- ated nutrient analysis program, and those produced by the mean of three 4-day diet records collected over a 1-year period. Group means suggested a good ability of the question- naire to distinguish between population groups at substantially different levels of true intake. The correlations observed, generally in the range of r = 0.5-0.6, suggest a moderate to good ability of the questionnaire to place individuals along the distribution of intake. [Corrections for variability in the reference data generally produce correlations which are considerably higher than observed correlations [5, 121, and did so in this study (data not shown.)] The questionnaire results also appear to be in the same range as the correlations achievable by a single 4day record, at least in the usual-diet grot@.

These correlations are also in the same range as those seen for various physiologic measures. For example, repeated measures of the sume variable have produced correlations such as the following: serum cholesterol, r = 0.59 v31; adult blood pressure, r = 0.67-0.70 [14]; uric acid, r =0.53 [14]; and even, for 24hour urine sodium, r = 0.24 [15]. Further- more, two well-accepted measures of obesity, body mass index and skinfold measurements, correlate only in the range of r = 0.49 to r = 0.65 [16]. Thus, while attention should cer- tainly be given to increasing sample sizes in etiologic studies, so as to improve statistical power in the face of the real misclassification inherent in al1 diet data, it also would appear that dietary measures of the sort shown here should prove at least as useful in health research as have the physiologic measures mentioned above.

It is worth emphasizing that the observed correlations are with the reference data, not with “true intake’?; Because of day-to-day and week- to-week variability, even three 4-day records are an imperfect estimate of true intake: most of the nutrients had a ratio of intra-to-interindividual variante approaching or exceeding 1 .O, even on transformed data (Black, unpublished data.) It has been shown elsewhere [ 171 that correlations between a questionnaire and reference data increase as the adequacy of the reference data (i.e. the number of diet records) increases.

Page 8: Block 1990

1334 GLADYS BLQCK et al.

With that caveat, the reference data constitute one of the major strengths of the study. Among others who have conducted validation studies of diet questionnaires, few [18, 191 have collected more days of diet records. In the present study, three 4-day records were collected over a l-year period, in two different seasons, with a represen- tative inclusion of weekdays and weekends. On the other hand, although members of the usual- diet group did not receive special diet instruc- tion, al1 respondents were highly motivated participants in a diet study. Unquestionably, these results can only be generalized to respon- dents who are motivated to complete the ques- tionnaire carefully, and to investigations which have engaged in careful data quality control, double-keying, edit-checking, and querying of questionable responses.

The means shown in Table 2 indicate a surprising ability of this questionnaire to esti- mate absolute values of usual nutrient intake with considerable accuracy. This was true even in two study groups with widely differing macronutrient intake.

Other investigators have conducted vali- dations in which diet records or recalls were used as the reference method [18-211. Willett et al. [lg] used as reference data the mean of four 7-day records collected over a l-year period, in validating a self-administered frequency ques- tionnaire in a population of nurses; correlations averaged approximately 0.45 without adjust- ment for calories. In a different population, including men as wel1 as women, Willett er al. [22] validated a longer version of the question- naire against 365 days of diet records, and obtained correlations averaging 0.60. Pietinen et al. [ 181 had 12 2-day records collected over a ti-month period; an average correlation co- efficient of 0.59 was observed for comparable nutrients, in their population of Finnish men. Balogh et al. [21] validated a brief frequency interview against eight or more nonconsecutive 24-hour recalls collected at random over 1 year, in 71 male subjects. An average correlation coefficient of 0.73 was obtained for nine com- parable macronutrients; micronutrients were not reported.

Jain et al. [20] conducted a validation study of a frequency-based diet history interview, in 16 male subjects. The interview was focused on diet in the past month, and the reference data con- sisted of 30-day records maintained by their spouses. The average correlation coefficient, based on nine comparable nutrients, was 0.45.

Stuff et al. [23] compared a quantified frequency questionnaire with a 7-day record in a group of lactating women, but only obtained correlations ranging from 0 to 0.24.

These studies have differed in many respects, and the observed correlations are not directly comparable. In general, self-administered ques- tionnaires would be expected to do more poorly than interviewer-administered questionnaires, because of respondent errors; study groups in- cluding both sexes would tend to have a wider range of nutrient intake and therefore higher correlations; and questionnaires targeted to and validated against a brief recent period such as the past month would be expected to produce higher correlations than those which ask about a longer and more variable period such as the past year. Finally, the more days of diet records comprising the reference data, the better the correlations are likely to be [17].

The questionnaire examined here appears to have a reasonable ability to assess current diet, over a wide range of nutrients and levels of fat intake. In addition to producing correlations in the same range as others have achieved using comparable reference data, it yields average nutrient estimates very similar to those obtained from multiple diet records. Thus, it may be useful both in dietary research settings and in clinical or counselling situations where nutrient estimates for a particular individual’s usual diet are needed. The questionnaire and accompany- ing nutrient analysis software are available from the authors on request.

Acknowledgemenrs-The authors wish to acknowledge the thouehtful advice of Susan Goldman in the develonment and Lsting of the questionnaire.

1.

2.

3.

4.

5.

REFERENCES

Black G, Hartman AM. Issues in reproducibility and validity of dietary studies. Am J Clin Nutr 1989; 50: 1133-1138. Black G, Hartman AM, Dresser CM, Carroll MD, Gannon J, Gardner L. A data-based approach to diet questionnaire design and testing. Am J Epidemiol 1986; 124: 453-469. Smucker R, Black G, Coyle L, Harvin A, Kessler L. A dietary and risk factor questionnaire and analysis system for personal computers. Am J Epidemiol 1989; 129: 445449. Cummings SR, Black G, McHenry K, Baron RB. Evaluation of two food frequency methods of measur- ing dietary calcium intake. Am J Epidemiol 1987; 126: 796-802. Sobell J, Black G, Koslowe P, Tobin J, Andres R. Validation of a retrospective questionnaire assessing diet 10-15 years ago. Am J Epidemiol 1989; 130: 173-187.

Page 9: Block 1990

Validation of a Diet Questionnaire 1335

6.

1.

8.

9.

10.

ll.

12.

13.

14.

15.

16.

17.

18.

19.

20.

Clifford CK, Butrum RR, Greenwald P, Yates JW. Clinical trials of low fat dists and breast cancer pmvention. In: Ip C, Birt DF, Rogers AE, Metthn C, Eds. DIatary Fat aud Caneer. New York: Alan R. Lim, Inc.; 1986: 93-115. Prentice RL, Kakar F, Hunting S, Sheppard L, Klein R, Kushi LH. Aspects of the rationale for the Women’s Health Trial. J NatI Caneer Inat 1988; 80: 802-814. Insull W, Henderson MM, Prent& RL, Thompson DJ, Clifford CK, Goldman S, Gorbach S, Moskowitx M, Thompson R, Woods M. Results of a randomixed feasibility study of a low-fat diet. Ar& Intern Med 1990; 150: 421427. Catsos PD, Woods M, Selles W, Gorbach S. The Tufts nutrient calculation system. 1988; Unpublished. Willett WC, Stampfer MJ, Underwood BA, Speixer FE, Rosner B, Hennekens CH. Validation of a dietary questionnaire with plasma carotenoid and alpha-toco- pherol levels. Am J CBa Nutr 1983; 38: 631639. Walker AM, Blettner M. Comparing imperfect measures of exnosure. Am J Esddemiol 1985: 121: 183-190. - Liu K, Stamler J, Dyer A, McKeever J, McKeever P. Statistical methods to assess and minimize the role of intra-individual variability in obscuring the relation- ship between dietary lipids and serum cholesterol. J Chrou Dis 1978; 31: 399-418. Morris JN, Marr JW, Heady JA, Mills GL, Pilkington TRE. Diet and plasma cholesterol in 99 bank men. Br Med J 1963; March: 571-576. Rhoads GG. Reliability of diet measures as chronic disease risk factors. Am J CBn Nutr 1987; 45: 1073-1079. Liu K, Dyer AR, Cooper RS, Stamler R, StamIer J. Can ovemight urine replace 24hour collection to assess salt intake? Hypertersrdou 1979; 1: 529-536. Stokes 111 J, Garrison RJ, Kannel WB. The indepen- dent contributions of various indices of obeaity to the 22-year incidence of coronary heart disease: The Framingham heart study. In: Vague J, Bjomtorp P, Guy-Grand B, Rebuffe-Scrive M, Vague P, Eds. MetahoI& CumpBcatIous of Hmnan Ohe&&sr Proc 6th Iut Meeting of FadoerlwLogy, Marseille, 30 May-l June 19851 Amsterdam: Ëxcerpta Medica/Elsevier Science Publishers: 1985: 49-57. Potosky A, Black G, Hartman AM. The effect of the referenc-e data in diet questionnaire validations. J Am DIet Asaoe 1990: J Am DIet Amue 1990: 90: 810813. Pietinen P, Hartman AM, Haapa E; Rasanen L, Haapakoski J, Palmgren J, Albanes D, Virtamo J, Huttunen JK. Reproducibility and validity of dietary assessment instruments: 1. A self-administered food use questionnaire with a portion size picture booklet. Am J Epidemiol 1988; 128: 655-666. Willett WC, Sampson L, Stampfer MJ, Rosner B, Bain C, Witschi J, Hennekens CH, Speizer FE. Reproduci- bility and validity of a semiquantitative food frequency questionnaire. Am J EpIdemIoI 1985; 122: 51-65. Jain M, Howe GR, Johnson KC, Miller AB. Evahr- ation of a diet history questionnaire for epidemiologic studes. Am J EpIdemIol 1980; 111: 212-219.

21.

22.

23.

Balogh M, Kahn HA, Medalie JH. Random repeat 24hour dietary recalls. Am J CBn Nutr 1971; 24~ 304-310. Wiiett WC, Reynolds RD, Cottrell-Hoehner S, Sampson L, Browne ML. Validation of a semi- quantitative food frequency questionnaire: comparison with a 1-year diet record. J Am Dkt Anme 1987; 87; 43-47. Stuff JE, Garxa C, O’Brian Smith E, Nichols BL, Montandon CM. A compatison of dietary methods in nutritional studies. Am J CIin Nutr 1983; 37: 300-306.

APPENDIX

The following infonnation is provided so that other investigators could replicate the questionnaire nutrient analysis exactly as it was performed for this investigation.

(1) Questions and food items omitted from this question- naire version: open-ended section; brand of dry cereal used; cantaloupe out of season; Tang; mixed vegetables; winter squash, chilli peppers; livenvurst; pmnpkin pie; decaffeinated toffee (combined with toffee); tea (combined with toffee).

(2) Questions and food items added to this questionnaire version: “When you eat poultry, is it usually light meat, dark meat, both?“; “When you eat hamburger or beef, is it usually ngular, lean, extra lean?” “When you eat tuna, it is usually oil pack, water pack, either one, don’t know?” “If you eat yogurt, what brand do you usually eat?“, coded as regular, low-fat or non-fat; “If you eat lunch meats or hot dogs, what kind do you usually eat?“, coded as regular or low-fat; diet margarine and whipped butter were added as possible responses to the kind of fat usually added to vegetables; popcorn; diet salad dressing sherbet or jello; low-fat cottage cheese or low-fat cheeses; plain yogurt.

(3) Errors which caused respondents to be dropped: skipped more than 15 food items; total foods per day adds to fewer than 4; total foods from lunch dishes, soups, breads, breakfast foods, dairy products and desserts/sweets adds to fewer than two per day; “medium” or omitted portion sim for 100% of foods; “once per” unit time responses for more than 75% of foods; more than two foods had “unreasonable” frequenties (e.g. carrots six times a day).

(4) The following represent options in the nutrient analysis software which affect the calculations. Users of this software who wish to duplicate the conditions of the present analysis would use the following option set- tings: PGRTIONS = AGESEX, MEDGNLY = NO, RESTADJ = YES, ADDFATS = YES, BIGFAT = YES, LEANQUES = YES, TUNAQUES = YES, VEGADJ = YES, FRTADJ = YES, YEARCOL = 0.0384599. For the DietEdit program, FIX = YES. These settings create a nutrient analysis with the characteristics described under “Methods”, including use of age- and sex-specitïc portion sizes, use of the respondent’s portion sim information (“small”, “medium”, large”), etc. For further information about calculation options and other characteristics of the nutrient analysis system, consult the manual which accom- panies the software, available from tbe authors.