This article was downloaded by: [New York University] On: 10 June 2015, At: 04:04 Publisher: Routledge Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Nutrition and Cancer Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/hnuc20

The reproducibility of a food frequency questionnaire among controls participating in a case‐control study on cancer a

a

b

H. B. Bueno de Mesquita , F. W. M. Smeets , S. Runia & K. F. A. M. Hulshof

c

a

Department of Epidemiology , National Institute of Public Health and Environmental Protection , PO Box 1, BA Bilthoven, 3720, The Netherlands b

Department of Dietetics , The Utrecht University Hospital , GA Utrecht, 3508, The Netherlands c

Department of Human Nutrition , TNO Toxicology and Nutrition Institute , AJ Zeist, 3700, The Netherlands Published online: 04 Aug 2009.

To cite this article: H. B. Bueno de Mesquita , F. W. M. Smeets , S. Runia & K. F. A. M. Hulshof (1992) The reproducibility of a food frequency questionnaire among controls participating in a case‐control study on cancer, Nutrition and Cancer, 18:2, 143-156, DOI: 10.1080/01635589209514214 To link to this article: http://dx.doi.org/10.1080/01635589209514214

PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,

Downloaded by [New York University] at 04:04 10 June 2015

systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://www.tandfonline.com/page/ terms-and-conditions

Downloaded by [New York University] at 04:04 10 June 2015

The Reproducibility of a Food Frequency Questionnaire Among Controls Participating in a Case-Control Study on Cancer H. B. Bueno de Mesquita, F. W. M. Smeets, S. Runia, and K. F. A. M. Hulshof

Abstract This study was designed to test the reproducibility of a food frequency questionnaire used in a population-based case-control study on diet and pancreatic cancer. Repeat questionnaires covering the same time period were obtained using 63 male and female population controls, 35-79 years of age. For selectedfood items included in the case-control study, the attenuation of the odds ratios due to random error was estimated. Using 54 male and female population controls, 35-79 years of age, we conducted a second study to examine the agreement between original and repeat interviews when the time interval between interview and the period of interest was constant. In the first study, the median correlation coefficient was 0.72 for foods (ranging from 0.36 and 0.59 for subgroups of vegetables to 0.96 for alcoholic beverages) and 0.77 for nutrients (ranging from 0.62 for β-carotene to 0.85 for energy and 0.91 for ethanol). In the second study, the median correlation coefficient was 0.68 for foods (ranging from 0.28 for eggs to 0.87 for alcoholic beverages) and 0.75 for nutrients (ranging from 0.48 for β-carotene to 0.76 for energy). We conclude that for most items the agreement between original and repeat estimates was moderate (r> 0.50) to high (r > 0.70). Moderate agreement was found for 28 of 33 food items (85%) and for all 21 nutrient items (100%) and high agreement for 19 of 33 of food items (56%) and 15 of 21 nutrient items (71 %). In the second study, agreement was somewhat lower but closely paralleled the results of the first study. On average, random error presumably attenuated most of the observed diet-cancer relationships only moderately; i.e., an observed odds ratio of 1.5 and a correlation coefficient of 0. 70 yield an unattenuated odds ratio of 2.1. (Nutr Cancer 18, 143-156, 1992)

H. B. Bueno de Mesquita and F. W. M. Smeets are affiliated with the Department of Epidemiology, National Institute of Public Health and Environmental Protection, 3720 BA Bilthoven, The Netherlands. S. Runia is affiliated with the Department of Dietetics, The Utrecht University Hospital, 3508 GA Utrecht, The Netherlands. K. F. A. M. Hulshof is affiliated with the Department of Human Nutrition, TNO Toxicology and Nutrition Institute, 3700 AJ Zeist, The Netherlands. Copyright © 1992, Lawrence Erlbaum Associates, Inc.

Downloaded by [New York University] at 04:04 10 June 2015

Introduction

From 1984 to 1988 a population-based case-control study was carried out in The Netherlands to explore further the possible relationship between habitual diet and carcinoma of the exocrine pancreas and biliary tract (1,2). A semiquantitative food frequency questionnaire (SQFFQ) was designed to assess comprehensively the normal diet about one year before diagnosis of cases or interview of controls. The SQFFQ is a recently developed dietary assessment method (3). The relative validity and reproducibility of such questionnaires used in other epidemiological studies on diet and cancer have been reported as adequate for the purpose of ranking subjects with respect to their usual dietary intake (4-12). Because studies vary according to purpose, study population, questionnaire, time intervals, and other factors, it is recommended that the relative validity and repeatability be evaluated each time a new questionnaire is designed (3,13,14). Testing the relative validity among patients with highly lethal cancers of the pancreas or biliary tract was not considered feasible. Because of limited resources, relative validation of our questionnaire for controls was also not carried out. A similar highly structured dietary history questionnaire used in a case-control study of breast cancer in The Netherlands showed an adequate coverage of individual fat intake (15). In general, except for higher estimates of consumption of vegetables and fruits in the case-control study, the actual intakes of food and nutrients reported by the population controls participating in the case-control study on pancreatic cancer were similar to those observed for the population at large on the basis of two-day dietary records (16). Dietary assessment methods usually contain a high degree of random misclassification or random measurement error. It is well known that random measurement error strongly attenuates risk estimates of the diet-disease relationships (17,18). The risk estimates suffer from a loss in precision, i.e., wider confidence intervals, and are biased toward the null value. The extent of attenuation depends on the precision of the method relative to the heterogeneity of dietary exposures, i.e., the ratio of within-subject to between-subject variance. Estimates of this variance ratio may be obtained from replicate measurements and may be used to correct relative risk estimates obtained from logistic regression models (19,20). Measures of repeatability of dietary assessment methods reflect both the performance of the questionnaire and the true changes in individual dietary intake. A low degree of repeatability is a definite indication that the questionnaire does not provide a valid measure of long-term intake. A high level of repeatability may indicate low random measurement error relative to between-subject variation in dietary intake but may also reflect reproduction of systematic error (21). Our study was designed to test the reproducibility of the dietary questionnaire by obtaining repeat questionnaires covering the same time period from a sample of population controls who participated in the case-control study of carcinoma of the pancreas and biliary tract. Information on reproducibility was used to correct the relative risk estimates, obtained in the case-control study, for attenuation due to random error (1,2). Agreement between original and repeat questionnaires may be adversely affected by a time effect, i.e., the difference in the interval between the period of interest and the time of the interview. With the assumption of long-term stability of one's usual diet in a relatively elderly population, a second study was designed to examine the agreement between the original and repeat questionnaires covering a period approximately one year previously, thus keeping the aforementioned interval constant. Here we report the results of the reproducibility study of the SQFFQ for food groups, energy-providing nutrients, and selected micronutrients potentially involved in the causation

144

Nutrition and Cancer 1992

of cancer. We further present an example of correction for attenuation of relative risk estimates obtained in the case-control study on pancreatic cancer. Material and Methods

Downloaded by [New York University] at 04:04 10 June 2015

Design In the case-control study, the original SQFFQ, which was administered by a single trained dietitian, referred to a time period approximately one year previously (Figure 1). The original interview of controls was conducted at home. Because of the nature of the disease of cases (patients with pancreatic cancer are often seriously ill) and because it was assumed that a number of subjects (often men) would not be sufficiently aware of past cooking habits, the original interviews, which included questions on food preparation, were conducted in the presence of the person who usually prepared the meals (mainly women). In the present study, the repeated SQFFQ covered either the same time period (Study 1) or a more recent time period (Study 2) (Figure 1). Study 1 was initiated in the winter of 1985-86, whereas Study 2 started in the summer of 1986. Reinterviewing was done in the same season approximately one year after the original interview by the same dietitian. Reinterviews were conducted at home and, in the majority of cases, in the presence of the same individuals as the original interview. Study Population The study population was drawn from a group of 216 eligible controls, 35-79 years of age and residing in the province of Utrecht, who participated in an ongoing population-based case-control study between October 1984 and October 1987. In The Netherlands, it was assumed that seasonal changes in usual diet usually relate to two seasons, i.e., winter and summer. The original study period covered three winters and three summers. Study 1 started in the winter of 1985/86. Because Study 2 was initiated in the summer of 1986, controls study I In = 63)

-12 months

original SQFFQ r

0

repeated SQFFQ +12 months

study 2 In = 54) Figure 1. Time sequence of 2 studies on reproducibility of a semiquantitative food frequency questionnaire (SQFFQ). SQFFQ was administered over the same time period (Study 1) or over another time period (Study 2) to 63 and 54 population controls, respectively, residing in Utrecht (1984-87).

Vol. 18, No. 2

original SQFFQ -12 months

repeated SQFFQ +12 months

145

Downloaded by [New York University] at 04:04 10 June 2015

originally interviewed in the winter of 1984/85 were recruited for Study 1 only; eligible subjects interviewed in the remaining five seasons were randomly allocated to Study 1 or Study 2. For each season of the original interview, controls were divided into four strata according to sex (male; female) and year of birth (1905-24; 1925-49) and assigned a random order. For each study, we aimed to enroll three subjects from each stratum. This should have resulted in 72 participants in Study 1, i.e., 3 from each of the four strata over six seasons, and 60 participants in Study 2, i.e., 3 from each of the four strata over five seasons. Ultimately, we lacked 11 eligible controls for Study 1 and 7 for Study 2. Of the 130 eligible controls allocated to Study 1, 63 agreed to participate (including 2 extra in 2 strata), 14 did not respond, and 53 were not approached because the required number of participants had been reached in their strata. Of 86 eligible controls allocated to Study 2, 54 agreed to participate (including 1 extra in 1 stratum), 8 did not respond, and 14 did not need to be approached. The 63 participants of Study 1, aged 40-80 years, included 30 men and 33 women, whereas the 54 participants of Study 2, aged 36-80 years, comprised 24 men and 29 women. Questionnaire The core of the original dietary questionnaire consisted of an SQFFQ that was developed in cooperation with the TNO Toxicology and Nutrition Institute (Zeist, The Netherlands). The aim of the SQFFQ was to obtain a comprehensive assessment of diet by estimating the usual individual intake of 116 commonly used food items or food groups to rank subjects according to the level of intake approximately one year earlier. Participants could choose from 10 predefined frequencies of consumption, i.e., seldom or never, once a month, once in a fortnight, one, two, three, four, five, and six days per week, and daily. Frequencies of more than once daily were accounted for by estimating the number of standard portions of an item consumed daily, e.g., the number of glasses of milk consumed daily. Standard portions in household measures were used to estimate usual amounts. Samples of tableware and color photographs of approximately full-sized meals were used to clarify standard portions. Seasonal variation in the consumption of foods was taken into account by eliciting information on usual duration in months (3, 6, or 9 mo) and related usual frequency of consumption of a priori defined products. The interviewer used separate codes for the same 10 choices of frequencies for three, six, and nine months. For the following foods, questions on seasonal variation were asked: 1) subgroups of vegetables, i.e., stews such as curly kale and sauerkraut, legumes such as kidney beans, string beans, marrow fats, and lentils, rhubarb, tomatoes on bread, raw carrots, raw vegetables such as lettuce, endive, and cucumber; 2) subgroups of fruits, i.e., fresh tangerines, fresh grapes, fresh berries, fresh peaches, fresh cherries, fresh plums; and 3) ice. Information on the usual frequency of consumption and number of standard portions of added sugar, different types of milk, and added fat was included as well. The interview usually took about one hour. Statistical Methods Frequency and size of the portions of the food items were converted to daily intake of foods, energy, and nutrients. The composition of the food groups is given in Table 1. The nutritional values (the constituents) of the foods were based on an extended computerized version of the Dutch Food Composition Table of 1985 (22). For analysis of retinol, j8-carotene, vitamin D, and vitamin E, food tables of the CIVO Toxicology and Nutrition Institute (Zeist, The Netherlands) and other evaluations were used (23,24). Means, standard error of the means, and related coefficients of variation were used to describe the distributions of the intake of nutrients and food groups. Student's matchedpaired t test was used to assess the level of difference between the original and the repeated

146

Nutrition and Cancer 1992

Table 1. Description of Food Groups Description

Food Group Cereals Low fiber High fiber

Potatoes Vegetables Low fiber

Downloaded by [New York University] at 04:04 10 June 2015

High fiber Cooked Raw Cruciferous Fruits Low fiber

High fiber Citrus Noncitrus

White bread; current bread; white rolls; (Na-free) rusks; rice, macaroni, etc.; thickener (Na-free) whole wheat bread; rye bread; muesli; bran and wheat germs; whole grain rice; oat flakes; barley Boiled; fried; mashed Cooked beets; cooked carrots; cruciferous; rhubarb; raw vegetables such as lettuce, endive, and cucumber; tomato on bread; tomato or other vegetable juice; tofu Cooked leafy vegetables such as endive and spinach and Belgian endive; legumes such as red kidney beans, string beans, marrow fats, and lentils; other cooked vegetables such as green peas and green beans Cooked leafy vegetables such as endive, spinach, and Belgian endive; cooked carrots; cooked beets; rhubarb Tomatoes on bread; raw carrots; raw vegetables such as lettuce, endive, and cucumber. Stews such as curly kale and sauerkraut; red and white cabbage, brussels sprouts, and cauliflower Fresh tangerines; fresh oranges and grapefruits; fresh grapes; fresh cherries; fresh peaches; applesauce; canned fruit; orange juice; other fruit juice; fresh mango; fresh papaya; fresh kiwi; fresh melon; sugar-free applesauce; pears in syrup Apple and pear; fresh berries; fresh bananas; fresh plums; dried fruits; fruit jelly Fresh tangerines; fresh oranges and grapefruits Apple and pear; fresh berries; fresh grapes; fresh cherries; fresh peaches; fresh plums; fresh mango; fresh papaya; fresh kiwi; fresh melon

Fruit juices Cooked

Orange juice; other fruit juices (Sugar-free) applesauce; fruit jelly; pears in syrup

Eggs

Boiled egg; fried egg; raw egg; egg yolk; egg white

Meat

(Na-free) ham; lean meat products; cooked liver; (Na-free) sausages; chicken; minced meat; fat pork; lean pork; other pork; lean beef; other beef; veal; liver and kidney; croquette; sausage rolls; meats between meals; lamb; horse meat; rabbit; lean meat products (Na-free) ham; fat pork; lean pork; other pork; sausage rolls Lean beef; other beef

Pork Beef Cheese

Fat edammer; (Na-free) fat cheese; lean cheese; camembert; cottage cheese; lean curd cheese; moderately fat curd; fat curd; lean curd with fruits

Milk (products)

Whipped cream; moderately fat cream; evaporated milk; Becel; fat evaporated milk; coffee creamer; ice cream; raw milk; whole milk; fat yogurt; lean evaporated milk; fat chocolate milk; custard; ready made porridge; moderately fat milk; skimmed milk; lean yogurt; buttermilk; lean chocolate milk; fat yogurt with fruits; Na-free moderately fat milk; buttermilk curd; kefir; lean yogurt with fruits; yogurt drinks; Protifar; Fortimel (Continued)

Vol. 18, No. 2

147

Table 1. (Continued)

Downloaded by [New York University] at 04:04 10 June 2015

Food Group

Description

Fermented milk products

Cheese; fat yogurt (with fruits); lean yogurt (with fruits); yogurt drinks; buttermilk; buttermilk curd; kefir; yogonaise

Fish

Boiled cod; smoked herring; shrimps; tinned fish; herring; smoked fish; fish fingers; fried fillet

Oils and fats

Corn oil; soy bean oil; sunflower oil; olive oil; Becel oil; safflower oil; frying fats; Croma; Becel frying fat; fat for deep frying; lard; beef fat; butter; margarine; Becel margarine; vegetable margarine; tub margarine; low-fat margarine; mayonnaise; salad dressing; french fried sauce; Becel dressing; yogonaise

Ready-made meals

Fried rice/Chinese noodles; pancake; pizza; salad

Total sweets Sweet products

Cakes and biscuits

Granulated sugar; soft white sugar; milk chocolate; pure chocolate; chocolate strands; other chocolate sweets; cocoa; honey; jam; other sweets; syrup; rosebud syrup; dessert sauce; blancmange; coffee creamer; sorbitol; fructose; sugar-free jam; sugar-free chocolate; sugar-free sweets; diet rosebud syrup Apple cake; (sugar-free) pastry; cake; Dutch honey cake; other cakes; plain biscuits; fancy biscuits; sugar-free biscuits

Nuts and tasty snacks

Peanuts; peanut butter; peanut sauce; tahina; salty biscuits; toast with pate

Alcoholic beverages

Beer; wine; fortified wine; spirits and liquor; eggnog

Nonalcoholic beverages

Water; mineral water; coffee; tea; cola; other soft drinks; sugar-free soft drinks

dietary interview (absolute agreement). A p value < 0.05 (2 tailed) was considered statistically significant. Pearson's product-moment correlation coefficient was used to evaluate the relative individual agreement in nutrient and food group intake. Although even after log transformation most food groups and some nutrient intake distributions were skewed, we decided not to use nonparametric methods, because one of our objectives was to estimate the degree of attenuation due to random misclassification. Because epidemiological data are often analyzed categorically, the values for nutrient and food group intake were divided into tertiles to provide further information on the extent of misclassification. At the end of the repeat interview, the respondent was asked whether there had been a change in dietary and drinking habits since the original interview. Because "dietary change" was not further specified, the response depended on the personal judgment of the respondent about the importance (the extent) of changes in specific foods. Of the total study population, positive responses were obtained in 32 subjects. Because preliminary analyses showed only minor changes in results, we concluded that this question did not measure what it was supposed to measure, and therefore we did not exclude these 32 subjects from the analysis. Results Food Groups The means, standard error of the means, and coefficients of variation of the distributions of daily intake of food groups are listed in Table 2. The repeat estimates, obtained in Study 148

Nutrition and Cancer 1992

Downloaded by [New York University] at 04:04 10 June 2015

1, for the average consumption of vegetables, high-fiber vegetables, and cruciferous vegetables were significantly higher than the original estimates, whereas a significantly lower repeat estimate was found for ready-made meals. In general, the correlation coefficients for most of the food groups were high, with a median of 0.71. Correlations ranged from 0.36 for the consumption of high-fiber vegetables to 0.96 for alcoholic beverages. For several food groups, the correlations between original and repeat values were relatively low, i.e., vegetables, high-fiber vegetables, cooked and raw vegetables, cooked fruits, beef, cakes and biscuits, and nuts and snacks. In Study 2, the repeat questionnaires generally yielded somewhat lower values than the original reported intake of foods, with significant differences for consumption of meat, total sweets, sweet products, and alcoholic beverages. On average, correlation coefficients were high, with a median of 0.68, ranging from 0.28 for the consumption of eggs to 0.85 for the consumption of oils and fats and sweet products (Table 2). Relatively low correlations were obtained for the consumption of high-fiber vegetables, eggs, cheese, and nuts and snacks. Thus, in both studies, the correlation coefficients for vegetable intake, especially high-fiber vegetables, and nuts and snacks were relatively low. In Study 1, the proportion of subjects classified in the same fertile ranged from 47.6% (cooked vegetables) to 82.2% (alcoholic beverages) (data not shown). Only in the case of beef (11.1%) was the proportion of subjects grossly misclassified, i.e., in opposite tertiles, > 10%. In Study 2, 48.2% (eggs) to 83.4% (alcoholic beverages) of the participants were classified consistently in the same tertile (data not shown). Only for cheese (12.0%) was the proportion of subjects grossly misclassified > 10%. Nutrients In Study 1, the repeat estimate for the mean intake of polyunsaturated fat was significantly lower than the original estimate (Table 3). Pearson correlation coefficients were high, with a median of 0.77, ranging from 0.62 for /3-carotene to 0.85 for energy and 0.91 for ethanol intake. In Study 2, the subjects reported a lower intake of many nutrients at the repeated SQFFQ, with significant differences for intake of energy, vegetable protein, cholesterol, total carbohydrates, polysaccharides, and ethanol (Table 3). The correlation coefficients were relatively high again, with a median correlation of 0.75, ranging from 0.48 for /3-carotene to 0.76 for energy and 0.87 for ethanol intake. Thus, both studies consistently showed relatively low correlations for the intake of /3-carotene and retinol equivalents and a relatively high correlation for ethanol intake (Table 3). In Study 1, the proportion of subjects classified in the same tertile ranged from 42.8% (/3-carotene) to 84.2% (ethanol), with a median of 65.1% (data not shown). Only in the case of /3-carotene (12.7%) was the proportion grossly misclassified, i.e., in opposite tertiles, > 10%. In Study 2, the proportions correctly classified ranged from 48.2% (animal protein and vitamin C) to 81.5% (polyunsaturated fatty acids), with a median of 62.9% (data not shown). With the exception of animal protein (11.1%), the proportion of subjects grossly misclassified never exceeded 10%. Discussion Our findings indicate that the SQFFQ, used in the case-control study on pancreatic and biliary cancer, provides repeatable estimates for most of the food groups and nutrients of interest. The percentage of subjects reclassified in the opposite tertile was almost always < 10%. We found relatively low correlations for the consumption (subgroups) of vegetables. The latter finding was reflected in similar results for /3-carotene. The most repeatable estimates were obtained for the consumption of alcoholic beverages and consequently for the intake of ethanol. The similarity of the results of the two study designs indicates that, in our Vol. 18, No. 2

149

Downloaded by [New York University] at 04:04 10 June 2015

Table 2. Daily Intake of Selected Food Groups and Pearson Correlation Coefficients Derived From Original and Repeat Interviews in Studies on the Reproducibility of ai Food Frequency Questionnaire"'6 Study 2 (n = 54)

Study 1 (n = 63)

Food Group

VO

Original interview

C

C

Mean ± SE

CV, %

176 ± 9 48 ± 6 128 ± 8

39 97 52

Potatoes Vegetables Low-fiber High-fiber Cooked

127 220 109 112 93 28 59

± ± ± ± ± ± ±

9 8 6 6 6 3 3

58 30 41

42

248 125 124 45 140 9

± ± ± ± ± ±

16 10 11 5 11 2

50 63 73 91 65 134

52 ± 9

142

Cruciferous

p

Repeated interview

C

Cereal products Low-fiber High-fiber

Raw

i

Original interview

Fruits Low-fiber High-fiber Citrus Non-citrus Cooked Fruit juices

48 96

45

CV, Vo

Mean ± SEC

CV, %

r

34 100 55

157 ± 8 *45 ± 7 112 ± 8

37 108 55

0.79 0.58 0.54

9 9 7 6 5 4 4

54 26 40 38 40 86 44

0.81 0.63 0.65 0.47 0.56 0.69 0.64

20 13 12 8 13 4

57 69 74 130 71 135

0.69 0.74 0.66 0.73 0.73 0.54

59 ± 10

131

0.70

cv,%

r

40 107 48

0.87 0.61 0.75

166 ± 8 62 ± 9 104 ± 8

46

135 242 115 127 106 31 61

± ± ± ± ± ± ±

11 11 7 8 8 3 4

61 34 42 47

40

0.80 0.49 0.57 0.36 0.48 0.58 0.59

15 12 10 5 10 3

49 71 70 97 62 159

0.68 0.69 0.72 0.85 0.72 0.43

244 139 105 44 120 14

± ± ± ± ± ±

17 12 10 6 11 2

50 71 84 100 66 115

58 ± 11

146

0.73

62 ± 10

120

Mean ± SE 172 ± 9 44 ± 6 127 ± 8 122 ± •239 ± 111 ± *128 ± 101 ± 23 ± *68 ± 242 131 111 42 128 14

± ± ± ± ± ±

7 9 5 6 5 2 4

30 41 36

42 83

Mean ± SE

Repeated interview

55 81 53

125 ± 239 ± 120 ± 120 ± 96 ± 30 ± 65 ± 262 144 118 48 135 20

± ± ± ± ± ±

o

f Downloaded by [New York University] at 04:04 10 June 2015

to

Eggs Meat Beef Pork Cheese Milk (products) Fermented milk products Fish Oils and fats Ready made meals Total sweets Sweet products Cakes and biscuits Nuts and snacks Alcoholic beverages Nonalcoholic beverages

± 2 ± 5 ± 3 ± 4 ± 7 ± 26 ± 18 ± 4

83 32 81 72 133 61 91 194

13 122 31 45 37 341 144 14

± 1 ± 5 ± 3 ± 5 ± 6 ± 26 ± 16 ± 4

± 2 ± 3 ± 6 ± 4 ± 3 ± 1 ± 22 1,117 ± 45

36 116 58 76 66 128 177

41 *15 81 48 33 7 104

2 3 6 5 4 2

14 121 31 45 41 345 155 15 43 19 77 44 33 8 99

32

± ± ± ± ± ± ± 1 ,135 ±

24 49

84 34 81 80 121 59 86 199

0.71 0.72 0.51 0.71 0.96 0.80 0.75 0.95

15 130 32 54 31 417 150 10

38 138 63 75 103 181 182 34

0.77 0.77 0.77 0.84 0.59 0.43 0.96 0.58

42 17 80 45 35 9 130

a: Abbreviations are as follows: CV coeff of variation; r, Pearson correlation coeff. b: Statistical significance is as follows: 0.05 (Student's matched-paired 2-tailed t test). c: Values are expressed in grams.

±

1 5

± 4 ± 6 ± 2 ± 38 ± 15 ± 3

68 30 81 77 55 66 74 191

± 2 ± 3 ± 8 ± 6 ± 4 ± 2 ± 29 1,108 ± 61

39 128 75 98 75 207 166 40

12 ±

*121 ± 31 47 30 406 166 11

± ± ± ± ± ±

41 15 *71 *39 32 6 *83

± ± ±

± ± ± 1,098 ±

1 5 3 5 2 34 18 2

87 32 75 70 59 61 81 133

0.28 0.68 0.68 0.76 0.29 0.67 0.66 0.60

2 3 6 5 3 1 24 62

42 132 66 88 69 149 215 42

0.85 0.52 0.81 0.85 0.69 0.50 0.78 0.84

to

Table 3. Daily Intake of Selected Nutrients and Pearson (Correlation Coefficients From a Semiquantitative Food Frequency Questionnaire in an Original and Repeated Interview Using Different Study Designs"'* Study 2 (n = 54)

Study 1 (n = 63)

Downloaded by [New York University] at 04:04 10 June 2015

Original interview Mean ± SE

cv%

Mean ± SE

cv, %

r

Mean ± SE

Energy, k j

8,969 ± 256

23

8,869 ± 286

26

0.85

9,314 ± 350

2 1 2

24 26 31

2 1 2

, 25 28 31

0.79 0.78 0.79

P

o.

9

I

75 ± 26 ± 50 ±

74 ± 26 ± 49 ±

Repeated interview

Original interview

Nutrient

Protein, g Vegetable Animal

o e»

Repeated interview

77 ± 26 ± 51 ±

CV, % 28

2 1 2

23 30 26

Mean ± SE

CV, %

r

*8,756 ± 324

27

0.76

2 1 2

21

29 23

0.61 0.76 0.55

73 ± *24 ± 49 ±

Fat, g Saturated Monounsatured Polyunsaturated Cholesterol, mg

97 41 38 16 295

± ± ± ± ±

3 1 1 1 11

25 25 26 35 30

95 41 38 *15 286

± ± ± ± ±

4 2 2 1 12

30 32 31 38 33

0.79 0.77 0.73 0.70 O.70

99 42 39 17 298

± ± ± ± ±

4 2 2 1 11

30 29 34 47 28

96 41 37 16 *276

± ± ± ± ±

5 2 2 1 11

35 35 36 47 29

0.74 0.71 0.76 0.80 0.55

Carbohydrate, g Mono- and disaccharides Polysaccharides Dietary fiber, g Ethanol, g

225 105 120 27 9

± ± ± ± ±

8 5 4 1 2

28 35 29 24 139

225 106 118 27 9

± ± ± ± ±

8 5 5 1 2

30 39 31 22 145

0.88 0.86 0.87 0.77 0.91

234 113 119 27 12

± ± ± ± ±

11 8 5 1 2

35 53 28 27 119

*218 106 •112 26 *9

=fc ± ± ± ±

9 6 5 1 2

31 47 30 29 140

0.82 0.86 0.81 0.75 0.87

Calcium, mg Phosphorus, mg Potassium, mg

1,053 ± 42 1,545 ± 50 3,873 ± 96

32 26 20

1,043 ± 1,534 ± 3,863 ±

41 50 99

31 26 20

0.77 0.81 0.68

1,123 ± 56 1,616 ± 64 4,037 ± 141

36 29 26

1,098 ± 52 1,531 ± 55 3,885 ± 129

35 27 24

0.59 0.70 0.83

Retinol equivalents, jig /J-Carotene, /ig Vitamin C, mg Vitamin E, mg

1,020 2,960 90 10

270 310 42

0.69 0.62 0.72 0.70

1,080 3,120 94 11

1,070 3,190 100 10

30 34 47 40

0.53 0.48 0.71 0.82

± 30 ± 150 ± 5 ± 0. 4

270 40 41 29

1,050 3,060 91 10

± 30 ± 120 ± 5 ± 0.4

31

a: Abbreviations are as follows: CV, coeff of variation; r, Pearson correlation coeff. < 0.05 (Student's matched-paired 2-tailed t test). b: Statistical significance is as follows:

± 40 ± 140 ± 5 ± 0.6

273 32 40 39

± 40 ± 150 ± 6 ± 0.6

Downloaded by [New York University] at 04:04 10 June 2015

, Odds Ratios (log. scale)

Odds flatios (log. scale)

2 3 t Quintiles of daily intake of energy

2 3 i Quintiles of daily Make of vegetables

- IBs Energy

- ORs Vegetables

•*• Deattenuated Ills

Figure 2. Observed and deattenuated odds ratios (ORs) for usual daily intake of energy in case-control study on cancer of pancreas. , Odds Ratios (log. scale)

-*- Deattenuated IBs

Figure 3. Observed and deattenuated odds ratios (ORs) for usual daily consumption of vegetables in case-control study on cancer of pancreas.

1.00*

Odds Ratios llog. scale)

_.x

k_O5 0.71

- • • •

0 64

o.i? 0.39*

\

2

3

i

Quintiles of low-fiber vegetables — 1 — OHs LF vegetables

-*- DeatteraateJ ORs

Figure 4. Observed and deattenuated odds ratios (ORs) for usual daily consumption of low-fiber (LF) vegetables in case-control study on cancer of pancreas,

Vol. 18, No. 2

0.10

I 3 < Quintiles of high-fiber vegetables - + - ORs I f vegetables

-*- Deattenuated ORs

Figure 5. Observed and deattenuated odds ratios (ORs) for usual daily consumption of high-fiber (HF) vegetables in case-control study on cancer of pancreas.

153

Downloaded by [New York University] at 04:04 10 June 2015

study population, repeat measurements obtained in Study 1 were barely influenced by prolonging the interval between the period of interest and the time of the interview. Our findings compare well with the results of other studies that used a comparable dietary assessment method (4-12). The distribution of reported median correlation coefficients ranged from 0.45 (9) to 0.71 (8), with the majority of values between 0.6 and 0.7. The correlation coefficients obtained in our study were at the high end of the distribution. Differences in design may account for differences in outcome and thus will hamper interstudy comparisons. For instance, the time interval between repeated measurements influences intrasubject variation and consequently intersubject variation in dietary intake. Reported intervals varied from one (12) to three months (5,11), six months (4), approximately one year (6-9), and several years (3). Some studies tested self-administered questionnaires (6,7,10-12), and others used interviewer-based questionnaires (4,5,8,9). The fairly high correlation coefficients obtained for our study may be attributed to a higher intersubject variation. However, in those studies in which standard deviations of intake were reported, the variation was larger (8,11,12) or comparable (7). The fairly high correlation coefficients may also result from the relatively small number of food items in our questionnaire. Only Willett and co-workers (7) listed fewer food items. The use of the same trained dietitian for the interview, the assistance of other household members, and the attention to seasons are probably also factors that had a favorable effect on the reproducibility. A poor reproducibility of the consumption of vegetables and consequently of /3-carotene has been encountered in most other studies as well. The reproducibility of vegetables or /3-carotene was among the lowest in four other studies (6,8-10), whereas two studies (7,12) did not report the reproducibility of /3-carotene or vegetable intake. A poor reproducibility will prevent identification of the weak associations so common in nutritional epidemiology. Yet, for a wide range of cancers, inverse relationships with the consumption of vegetables are consistently found (2,25). If the consumption of vegetables truly has a protective effect, the actual associations will be even stronger than those observed. Another noteworthy finding was the high reproducibility of ethanol intake, which may be attributed to the high intersubject variation of alcohol intake. The high proportion of subjects correctly reclassified in the same tertile (84.2% in Study 1) and the high rank correlation for ethanol (r = 0.92), however, suggest small random error. Ethanol intake was estimated on the basis of a limited number of alcoholic beverages, which may have reduced random error. Three of the previously mentioned studies also found the highest correlation coefficient for ethanol intake (6,10,11), with the Finnish study reporting comparable intersubject variation (11). Although assessment of the intake of ethanol appears to be reproducible, the validity of ethanol intake is reported to be poor (26). For most items, we observed similar agreement in both studies indicating that the repeat measurements obtained in Study 1 were barely influenced by prolonging the interval between the period of interest and the time of interview one year. In our study population, the stability of usual diet, at least over the past three years, may explain such an effect. No studies with a comparable dietary assessment method have examined the effect of different intervals. Bloemberg and co-workers (27) found similar results using time intervals of 3 and 12 months. In that study the cross-check dietary history method was used. Although the degree of agreement was comparable in both studies, agreement for several items was less in Study 2 than in Study 1. The lower agreement in Study 2 may be related to true changes in mean consumption over time due to increased health awareness and/or aging. An example of increased health awareness may be the apparent shift from low- to high-fiber cereals and the decrease in consumption of eggs (possibly related to the lower agreement for cholesterol) and alcoholic beverages. The lower agreement for nutrients such as energy and calcium and foods such as cereal products, cheese, and milk products (other

154

Nutrition and Cancer 1992

Downloaded by [New York University] at 04:04 10 June 2015

than fermented milk products) may in part be related to a decreased energy need due to aging. As mentioned in the Introduction, validation of a retrospective method is difficult. Measuring reproducibility is a minimum requirement for determination of the precision of dietary assessment methods. The precision depends on the random error variance and the ratio of intra- and intersubject variation. With the assumption that errors in the original and repeated interviews are independent of each other, the unattenuated risk estimation can be obtained from this ratio (19,20). The Pearson correlation coefficient is based on this ratio. In general, correlations between repeated assessments tend to be on the order of 0.6 for most nutrients. Biological variables, such as blood pressure, have a comparable or even lower reproducibility (28). The degree to which nondifferential misclassification was due to real measurement error or a true variation in diet remains unclear. Using the linear approximation method, the "true" logistic regression coefficient /?, is estimated by dividing the observed regression coefficient /3O by ro, i.e., the observed correlation between original and repeat estimations of intake (20). We assume that the observed correlation coefficient ro does not change with increasing exposure and is equal among cases and controls. Given a ro of 0.85 from Study 1, the odds ratio for the highest quintiles of energy intake, reported in the case-control study on cancer of the pancreas (1), increases from 3.44 to an unattenuated true odds ratio of 4.29 (Figure 2). Similarly, given ro's of 0.49, 0.57, and 0.36, respectively, the odds ratio for the highest quintile of vegetables is reduced from 0.34 to 0.11 (Figure 3), that for low-fiber vegetables from 0.27 to 0.10 (Figure 4), and that for high-fiber vegetables from 0.64 to 0.29 (Figure 5) (2). These values indeed show that, after correction for random error, both low- and high-fiber vegetables exert a protective effect. In general, an observed odds ratio of 1.5 and a median correlation coefficient of 0.7 yield an unattenuated true odds ratio of 1.8. Our overall results therefore suggest relatively little attenuation of the risk estimates due to random error and, consequently, a greater chance of detecting weak associations. In conclusion, test-retest results indicate that our food frequency questionnaire provides reliable estimates for most of the main food groups, energy-providing nutrients, calcium, and phosphorus. Reliability for vegetables, cooked fruits, beef, cakes and biscuits, nuts and snacks, nonalcoholic beverages, and 0-carotene was substantially lower. Gross misclassification, i.e., > 10% in opposite tertiles, was observed for beef and 0-carotene. This study does not provide information on the validity of our questionnaire, and cautious interpretation of results is therefore warranted. Acknowledgments and Notes The authors thank F. de Waard (Dept. of Public Health and Epidemiology, University of Utrecht, The Netherlands), C. J. Moerman, E. Rontgen-Pieper, and M. C. E. Stam-Rademaker (Dept. of Epidemiology of the National Institute of Public Health and Environmental Protection, Bilthoven, The Netherlands) for participation in the study and D. Kromhout, B. Bloemberg, E. Feskens, and C. De Lezenne Coulander for evaluation of the draft. Address reprint requests to H. B. Bueno de Mesquita, PO Box 1, 3720 Bilthoven, The Netherlands. Submitted 7 May 1991; accepted in final form 14 May 1992.

References 1. Bueno de Mesquita, HB, Moerman, CJ, Runia, S, and Maisonneuve, P: "Are Energy and Energy-Providing Nutrients Related to Exocrine Carcinoma of the Pancreas?" Int J Cancer 46, 435-444, 1990. 2. Bueno de Mesquita, HB, Maisonneuve, P, Runia, S, and Moerman, CJ: "The Intake of Foods and Nutrients and Cancer of the Exocrine Pancreas: A Population-Based Case-Control Study in the Netherlands." Int J Cancer 48, 540-549, 1991. 3. Bingham, SA, Nelson, M, Paul, AA, Haraldsdottir, J, Bjtfrgen Ltfken, E, et al.: "Methods for Data Collection

Vol. 18, No. 2

155

4. 5. 6. 7. 8. 9. 10.

Downloaded by [New York University] at 04:04 10 June 2015

11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28.

at an Individual Level." In Manual on Methodology for Food Consumption Studies, ME Cameron and WA Van Staveren (eds). Oxford, UK: Oxford University Press, 1988, pp 95-97. Jain, M, Howe, GR, and Miller, AB: "Evaluation of a Diet History Questionnaire for Epidemiologic Studies." Am J Epidemiol 111, 212-219, 1980. Hankin, JH, Nomura, AMY, Lee, J, Hirohata, T, and Kolonel, LN: "Reproducibility of a Diet History Questionnaire in a Case-Control Study of Breast Cancer." Am J Clin Nutr 37, 981-985, 1983. Rohan, TE, and Potter, JD: "Retrospective Assessment of Dietary Intake." Am J Epidemiol 120, 876-887, 1984. Willett, WC, Sampson, L, Stampfer, MJ, Rosner, B, Bain, C, et al.: "Reproducibility and Validity of a Semiquantitative Food Frequency Questionnaire." Am J Epidemiol 122, 51-65, 1985. McKeown-Eyssen, GE, Yeung, KS, and Bright-See, E: "Assessment of Past Diet in Epidemiologic Studies." Am J Epidemiol 124, 94-103, 1986. Byers, TE, Marshall, J, Anthony, E, Fiedler, R, and Zielezny, M: "The Reliability of Dietary History From the Distant Past." Am J Epidemiol 125, 999-1011, 1987. Rohan, TE, Record, SI, and Cook, MG: "Repeatability of Estimates of Nutrient and Energy Intake: The Quantitative Food Frequency Approach." Nutr Res 7, 125-137, 1987. Pietinen, P, Hartman, AM, Haapa, E, Rasanen, L, Haapakoski, J, et al.: "Reproducibility and Validity of Dietary Assessment Instruments. I. A Self-Administered Food Use Questionnaire With a Portion Size Picture Booklet." Am J Epidemiol 128, 655-666, 1988. Engle, A, Lynn, L, Koury, K, and Boyar, AP: "Reproducibility and Comparability of a Computerized, Self-Administered Food Frequency Questionnaire." Nutr Cancer 13, 280-292, 1990. Block, G: "A Review of Validations of Dietary Assessment Methods." Am J Epidemiol 115, 492-505, 1982. Lee-Han, H, McGuire, V, and Boyd, NF: "A Review of the Methods Used by Studies of Dietary Measurement." J Clin Epidemiol 42, 269-279, 1989. Van't Veer, P: Dietary Habits and Breast Cancer (Doctoral thesis). Zeist, The Netherlands: TNO-CIVO Toxicology and Nutrition Institute, 1990. Wat eet Nederland. Resultaten van de voedselconsumptiepeiling 1987-1988 (in Dutch). Rijswijk, The Netherlands: Ministry of Welfare, Health and Culture, 1988. Van Staveren, WA, Burema, J, Deurenberg, P, and Katan, MB: "Weak Associations in Nutritional Epidemiology. The Importance of Replication of Observations on Individuals." Int J Epidemiol 16, 964-969, 1988. Willett, WC: "Nutritional Epidemiology: Issues and Challenges." Int J Epidemiol 16, 312-317, 1987. Willett, WC: "An Overview of Issues Related to the Correction of Nondifferential Exposure Measurement Error in Epidemiologic Studies." Stat Med 8, 1031-1040, 1989. Rosner, B, Willett, WC, and Spiegelman, D: "Correction of Logistic Regression Relative Risk Estimates and Confidence Intervals for Systematic Within-Person Measurement Error." Stat Med 8, 1051-1069, 1989. Willett, W: Nutritional Epidemiology. New York: Oxford University Press, 1990. Kommissie Uniforme Codering Voedingsmiddelen Tabel: Uitgebreide Voedingsmiddelen Tabel 1985. The Hague, The Netherlands: Voorlichtingsbureau voor de Voeding (Bureau of Nutrition Education), 1985. Bovens, M, Hulshof, KFAM, and Kistemaker, C: Estimation, Analysis and Evaluation of the Nutrient Composition of 130 Food Items for the Purpose of the SEARCH Project. Zeist, The Netherlands: TNO-CIVO Toxicology and Nutrition Institute, 1984. (Rep. No. V 84.320/080156) Runia, S, and Bueno de Mesquita, HB: Nutrient Analyses of Foods and Computation of Nutrient Intake for the Purpose of a Case-Control Study on Diet and Cancer. Bilthoven, The Netherlands: National Institute of Public Health and Environmental Hygiene, 1989. (Rep No. 52 84 78 001-6) Negri, E, La Vecchia, C, Franceschi, S, D'Avanzo, B, and Parazzini, F: "Vegetable Intake and Fruit Consumption and Cancer Risk." Int J Cancer 48, 350-354, 1991. Lemmens, PHHM, Tan, FES, and Knibbe, RA: "Measuring Quantity and Frequency of Drinking in a General Population Survey. A Comparison of 5 Indices." In Measurement and Distribution of Alcohol Consumption, PHHM Lemmens (ed). The Hague, The Netherlands: CIP Gegevens Koninklijke Bibliotheek, 1991, pp 51-69. Bloemberg, BPM, Kromhout, D, Obermann-De Boer, GL, and Van Kampen-Donker, M: "The Reproducibility of Dietary Intake Data Assessed With the Cross-Check Dietary History Method." Am J Epidemiol 130, 1047-1056, 1989. Shepard, DS: "Reliability of Blood Pressure Measurements: Implications for Designing and Evaluating Programs to Control Hypertension." J Chron Dis 34, 191-209, 1981.

156

Nutrition and Cancer 1992