Nutritional Epidemiology

Transcription

1 Nutritional Epidemiology Portion Size Adds Limited Information on Variance in Food Intake of Participants in the EPIC-Potsdam Study 1 Ute Noethlings, 2 Kurt Hoffmann, Manuela M. Bergmann and Heiner Boeing German Institute of Human Nutrition, Department of Epidemiology, Bergholz-Rehbruecke, Germany ABSTRACT Food-frequency questionnaire (FFQ) data should reflect interindividual variation and therefore measure variance in intake among populations. We conducted this analysis to evaluate the relevance of separate portion size questions to the interindividual variation in food intake. The contribution of portion size questions to the variance in food intake was quantified and compared with the variance when group-specific portion sizes would be assigned, using 26,764 FFQ of the European Investigation into Cancer and Nutrition (EPIC)-Potsdam Study. Groups were defined according to gender, age ( 50y, 50y) or body mass index (BMI) ( 26 kg/m 2, 26 kg/m 2 ). The FFQ inquired about both consumption frequency and portion size. Linear regression models for each food item were fit with intake (g/d) as dependent variables and frequency of intake as independent variables. The mean coefficient of determination (R 2 ) for the different food items explained by frequency only was 84.0% ( %). The R 2 for gender-, age- and BMI-specific frequencies of intake did not markedly alter the overall results. We conclude that the omission of individual portion size information would probably result in a notable reduction of interindividual variance. However, to reduce the respondents burden and to increase data completeness in self-administration in large epidemiologic studies, the assignment of a constant portion size seems to be adequate. The variance was not increased markedly when constant gender-, age- and BMI-specific portion sizes were applied, thus supporting the assignment of an overall portion size. J. Nutr. 133: , KEY WORDS: food-frequency questionnaire portion size EPIC-Potsdam Study 1 The EPIC-Potsdam Study is supported by the Deutsche Krebshilfe (Grant no Ha I) and the European Community (Grant No. SI F02). 2 To whom correspondence should be addressed. noethlin@mail. dife.de. 3 Abbreviations used: BMI, body mass index; CL, confidence limits; EI, energy intake; EPIC, European Investigation into Cancer and Nutrition; ER, energy requirement; FFQ, food-frequency questionnaire. The food-frequency questionnaire (FFQ) 3 has become the primary instrument for dietary assessment in large-scale epidemiologic studies (1,2). The main reason for this development during the last decades is the agreement among scientists that FFQ data are sufficiently valid for etiologic studies (3). This is accompanied by feasibility and budget considerations. In studying diet and disease relationships, comparison of disease risk for quintiles of intake is often used as a measure of association (4,5). Therefore, the dietary assessment instrument has to rank individuals and reflect variance in food intake rather than estimate absolute amounts of individual intake. If the aim of dietary assessment by FFQ is to quantify intake, recent considerations have suggested conducting standardization studies as an integral part of dietary assessments (5 8). FFQ data are nevertheless sometimes used to assess mean values and distributions of nutrient intake among groups without standardization studies (9). Whether the main focus of a FFQ is to provide information about quantitative nutrient or food intake or about variance in intake in a population is still under discussion (10,11). Considerations about compliance with follow-up procedures and data completeness led us to the modification of the existing FFQ in the European Investigation into Cancer and Nutrition (EPIC)-Potsdam Study (12). The revised instrument will be incorporated in the follow-up of the cohort study for repeated dietary exposure assessment in addition to the routine ascertainment of incident cases (13). In FFQ, information about amounts consumed is required to calculate daily intake. The two possibilities of incorporating this information into the analysis are either to ask the participants about individual portion sizes or to assign portion sizes. With the aim of simplifying the FFQ, we evaluated the need to ask separate questions about portion size. Information on average portion sizes in FFQ can be asked in separate questions ( How often did you eat apples? and What was the usual portion size? (e.g., 0.5, 1), be incorporated into the question about the consumption frequency of the food item itself ( How often did you eat one apple? ) (semiquantitative questionnaire) or can be omitted ( How often did you eat apples? ). In the last two instances, assigned portion sizes are used in the analysis (14). The assigned portion sizes can either be the same for all subjects or they can be chosen specifically for men and women or age groups, because amounts consumed differ by age and gender (15). We therefore investigated the effect of separate portion size questions on variance in reported food intake in our previous FFQ. Furthermore, we analyzed whether assignment of groupspecific portion sizes in contrast to assignment of an overall portion size would increase interindividual variance in food intake /03 $ American Society for Nutritional Sciences. Manuscript received 23 August Initial review completed 24 September Revision accepted 4 November

2 PORTION SIZES IN FOOD-FREQUENCY QUESTIONNAIRES 511 SUBJECTS AND METHODS Study population. The study population was the EPIC-Potsdam cohort, which is one of the two German cohorts contributing to the EPIC study, a multicenter cohort study into diet and chronic diseases, especially cancer (16). A total of 27,548 study participants was recruited between 1994 and Recruitment procedures have been reported elsewhere (17). Approval for all study procedures was given by the Ethical Committee of the State of Brandenburg, Germany, and written informed consent was obtained from all study participants. Participants with missing dietary data (n 109) and body mass index [(BMI) weight (kg)/height (m) 2 ](n 212) were excluded from the analysis. The lower and upper 1% of participants were excluded also due to implausible energy intake in relation to energy requirement [energy intake (EI)/energy requirement (ER)] (n 549). ER was calculated as basal metabolic rate 1.35 (18,19). Lower and upper bounds were EI/ER 0.31 and EI/ER 1.65, respectively. In total, data of 784 participants were excluded and data of 26,764 participants were available. Food-frequency questionnaire. The FFQ was designed and validated for the application in EPIC-Germany (12,20,21). The FFQ consisted of a food list containing 148 food items accompanied by questions about preparation methods and preferred fat content of specific products (such as dairy and meat). For each food item, participants were asked about both frequency of consumption and average portion size. Frequency of intake was measured using a scale of 10 categories ranging from never, one time per month or less to five times per day or more. Photographs or, if available, household measures were used to define portion sizes for each food item. Respondents were either asked whether their average portion size was half, the same, double or three times the amount shown (relative portion sizes; 103 items), or respondents had to choose one out of three pictures showing different amounts of foods or dishes (absolute portion sizes; 38 items). The latter are used to visualize small, medium and large portions. The actual portion sizes were derived from a representative national nutrition survey (Nationale Verzehrsstudie). Daily consumed amount was calculated by multiplying frequency per day and portion size. In several cases of similar food items, portion size was obtained only for the generic item, resulting in 141 items used in this analysis. These 141 food items were divided into 24 food groups, each consisting of 1 24 food items (Table 1). Statistical analysis. Mean food intake for men and women and for age ( 50 y; 50 y) and BMI ( 26 kg/m 2 ; 26 kg/m 2 ) groups were calculated. Median values for age and BMI were chosen as cut-off points. Gender, age and BMI were considered the most discriminating variables concerning food intake and were therefore focused on to potentially assign group-specific portion sizes. We used the food group cereals as an example to analyze whether differences in mean intake could be either attributed to the variation in portion sizes or to the variation in consumption frequencies or both. We therefore compared the average reported portion sizes and the average reported consumption frequencies between men and women and according to age and BMI groups for five items. We chose the food group cereals as an example because of the heterogeneity in consumption patterns of the five items, i.e., consumption at various times a day and different percentage of nonconsumers. To estimate the contribution of frequency of intake to variance in food intake, linear regression models for each food item were fit (model 1). Intake (I i ) (g/d) of item i (i 1,...141) was used as the dependent variable. The independent variable was defined by frequency F i of intake per day of food i (i ). No intercept was allowed in the models because, by definition, daily intake has to be zero if frequency of intake is zero. I i F i (1) We calculated the coefficient of determination, R 2, to estimate the proportion of variance explained by frequency alone. Because the product of portion size and food intake completely explains variance of intake, the loss of variance, caused by omission of portion size information, is equal to 1 R 2. The single R 2 -values were combined in arithmetic means for food groups. To analyze whether explained variance increases markedly by Food group TABLE 1 Food items belonging to 24 food groups, EPIC-Potsdam Study Food item No. of items Potatoes Potatoes (as side dish), mashed and roast 7 potatoes, French fries, potato fritter, potato salad Meat Poultry, pork, beef, hamburger, minced meat, 16 liver, lamb, roast hare Soup Vegetable or potato stew, vegetable, meat and 2 fish soup, broth Cereals Cereal flakes, grains, muesli, cornflakes, crisps, 5 pasta, rice, vegetarian dishes Desserts Pudding, fruity curd, sweet soufflé, canned fruit 3 Bread Bread, rolls (white or whole-wheat), croissant, 6 pretzel Fats Butter, margarine 3 Eggs Eggs, omelette, fried and scrambled eggs 2 Salty snacks Pizza, quiche, potato chips, cracker 3 Sweets Chocolates, pralines, candy bars, ice-cream 4 Legumes Beans, peas, lentil, -stew 3 Fish Fish, canned fish, smoked fish 2 Coffee, tea Coffee, black, green, fruit and herbal tea 5 Cake Fruitcake, layer cake, biscuits, pancake, 7 cookies Fruits All kinds of fresh fruits 14 Alcoholic Beer, wine, champagne, liqueur, aperitifs, 6 beverages spirits Vegetables All kinds of fresh and cooked vegetables, 24 vegetarian paste Milk, milk Milk, milk shakes, yogurt, (herbal) curd, (high 4 products and low fat) Soft drinks Mineral and tap water, fruit juices, lemonade, 8 cola, multivitamin drinks Processed Salami, cold-cut sausage, ham, fried sausage, 8 meat liver sausage Spreads Honey, jam, jelly, peanut butter, chocolate 3 spread Cheese All kinds of cheese (high and low fat) 4 Sauce Ketchup 1 Nuts All kinds of nuts 1 assignment of group-specific portion sizes, we fit linear regression models incorporating BMI-, gender- and age-specific frequencies of intake as independent variables. The number of independent variables equaled the number of groups under investigation in the different models. In detail, we used group-specific intake frequencies F ij. Here, F ij F j if the i th individual belongs to group j and F ij 0 otherwise. In the case of two groups, e.g., men and women, the model equation has the form I i 1 F i1 2 F i2 (2) The two parameters I and 2 can be interpreted as portion sizes of individuals of groups 1 and 2, respectively. Frequency of intake has been stratified in this way for gender, for BMI groups, for age groups and for combinations of age and gender, BMI and gender and of age, gender and BMI groups. To evaluate the effect of group- specific portion sizes, we compared R 2 of model 2 with that of model 1. The difference in R 2 would be the increase of explained variance attributed to the assignment of age-, gender- and BMI-specific portion sizes. To investigate the cumulative effect of a predefined portion size vs. variable portion sizes on the nutrient intake, we analyzed data of our FFQ validation study (21). The procedure of the validation was reported elsewhere. In brief, 134 participants (75 men and 59 women) completed h dietary recalls throughout a year and filled in the FFQ at the end of this year. We simulated the assignment of predefined portion sizes by replacing all absolute portion sizes (i.e.,

3 512 NOETHLINGS ET AL. TABLE 2 Mean food intake measured by FFQ in the EPIC-Potsdam Study1 Food group2 Men n 10,596 Women n 16,168 n 14,011 n 12,753 Age 50 y n 12,872 Age 50 y n 13,892 Potatoes (134.9; 138.1) (107.6; 109.7) (114.2; 116.7) (123.0; 125.7) (109.0; 111.4) (127.1; 129.8) Meat 84.7 (83.6; 85.7) 61.1 (60.5; 61.7) 64.8 (64.0; 65.5) 76.7 (75.8; 77.5) 72.4 (71.6; 73.2) 68.6 (67.9; 69.4) Soup 48.8 (47.9; 49.7) 43.2 (42.5; 43.8) 43.1 (42.4; 43.8) 47.9 (47.1; 48.7) 41.9 (41.2; 42.5) 48.7 (47.9; 49.4) Cereals 27.0 (26.4; 27.5) 30.0 (29.5; 30.4) 31.9 (31.4; 32.4) 25.4 (24.9; 25.8) 34.3 (33.7; 34.8) 23.7 (23.3; 24.1) Desserts 36.8 (36.1; 37.6) 32.9 (32.3; 33.4) 33.2 (32.6; 33.8) 35.8 (35.1; 36.4) 32.7 (32.1; 33.3) 36.1 (35.4; 36.7) Bread (246.5; 250.8) (194.0; 197.0) (212.3; 215.8) (217.4; 221.1) (213.0; 216.8) (216.4; 219.8) Fats 27.0 (26.7; 27.3) 20.4 (20.2; 20.6) 23.3 (23.1; 23.6) 22.7 (22.4; 23.0) 23.9 (23.7; 24.2) 22.2 (21.9; 22.4) Eggs 19.3 (19.0; 19.7) 16.1 (15.9; 16.4) 17.0 (16.8; 17.3) 17.8 (17.6; 18.1) 19.1 (18.8; 19.4) 15.9 (15.6; 16.1) Salty snacks 11.4 (11.1; 11.7) 10.7 (10.5; 10.9) 12.0 (11.8; 12.3) 9.7 (9.5; 10.0) 14.3 (14.0; 14.6) 7.8 (7.6; 8.0) Sweets 16.8 (16.5; 17.2) 16.2 (15.9; 16.6) 17.1 (16.7; 17.4) 15.8 (15.5; 16.2) 18.8 (18.4; 19.2) 14.3 (14.0; 14.6) Legumes 41.3 (40.6; 42.0) 29.7 (29.3; 30.2) 32.0 (31.5; 32.6) 36.8 (36.3; 37.4) 33.3 (32.7; 33.8) 35.3 (34.7; 35.8) Fish 28.1 (27.5; 28.7) 21.0 (20.7; 21.4) 21.9 (21.6; 22.3) 25.9 (25.4; 26.4) 23.3 (22.9; 23.7) 24.4 (23.9; 24.8) Coffee, tea (691.5; 708.0) (733.0; 747.1) (709.8; 724.1) (723.8; 740.1) (714.6; 730.9) (718.2; 732.4) Cake 67.7 (66.5; 69.0) 58.0 (57.2; 58.8) 63.4 (62.4; 64.4) 60.1 (59.1; 61.2) 60.6 (59.5; 61.6) 63.1 (62.0; 64.1) Fruits (246.9; 252.8) (348.0; 358.4) (302.2; 311.7) (311.3; 322.5) (296.7; 307.1) (315.6; 325.9) Alcoholic beverages (443.7; 462.8) (110.8; 115.9) (213.1; 224.3) (272.7; 287.3) (241.0; 254.1) (241.9; 254.5) Vegetables (199.7; 205.2) (233.6; 238.9) (212.1; 217.2) (228.9; 234.9) (219.9; 225.6) (220.3; 225.7) Milk, milk products (171.4; 179.8) (204.5; 211.3) (189.9; 196.9) (193.1; 201.0) (189.3; 197.0) (193.3; 200.6) Soft drinks (765.8; 794.1) (914.4; 938.9) (781.4; 805.1) (936.9; 965.8) (872.8; 899.6) (839.3; 865.2) Processed meat 97.4 (95.8; 99.1) 53.8 (53.0; 54.6) 63.6 (62.5; 64.7) 79.3 (78.0; 80.6) 73.2 (71.9; 74.5) 69.1 (67.9; 70.2) Spreads 12.5 (12.3; 12.8) 11.1 (10.9; 11.3) 12.3 (12.1; 12.6) 11.0 (10.8; 11.2) 9.4 (9.2; 9.6) 13.8 (13.6; 14.0) Cheese 46.4 (45.5; 47.3) 45.9 (45.2; 46.6) 46.2 (45.4; 46.9) 46.0 (45.2; 46.8) 45.1 (44.3; 45.8) 47.0 (46.2; 47.8) Sauce 2.5 (2.4; 2.6) 2.1 (2.1; 2.2) 2.2 (2.1; 2.3) 2.4 (2.3; 2.4) 3.0 (2.9; 3.1) 1.6 (1.6; 1.7) Nuts 3.7 (3.6; 3.9) 2.9 (2.8; 3.0) 3.3 (3.2; 3.5) 3.1 (3.0; 3.2) 3.5 (3.3; 3.6) 3.0 (2.8; 3.1) 1 Values are means and 95% confidence limits. 2 Food group intake in g/d. g/d small, medium, large) by the medium portion sizes and compared nutrient intake derived from FFQ correlated with 24-h recalls with simulated FFQ data with 24-h recalls. Spearman correlation coefficients are given for the comparison between nutrient intakes according to the mean of 24-h recalls, FFQ with variable portion sizes (original data) and FFQ with assigned portion sizes (simulated data). We used SAS for Windows V8 (SAS Institute, Cary, NC) to conduct all statistical analyses. RESULTS The study population comprised 10,596 men (40%) and 16,168 women (60%); 52% of the participants had BMI 26 kg/m 2 and 48% of participants were younger than 50 y. Due to the sampling frame of EPIC (16), the men were older with a mean age of 52 y and had a higher mean BMI (27 kg/m 2 ). Women had a mean age of 49 y and a BMI of 26 kg/m 2. Mean intake of food groups and 95% confidence limits (CL) are presented in Table 2. Food consumption showed significant gender differences in all food groups except for sweets and cheese. Cereals, coffee, tea, fruits, vegetables, milk, milk products and soft drinks were consumed in higher amounts by women, whereas men consumed higher amounts of potatoes, meat, soup, desserts, bread, fats, eggs, salty snacks, sweets, legumes, fish, cake, alcoholic beverages, processed meat, spreads, cheese, sauce and nuts. Consumption of most of the food groups was different for participants with a BMI 26 kg/m 2 and those with a BMI 26 kg/m 2. No differences according to BMI were observed for coffee, tea, fruits, milk, milk products, cheese, sauce and nuts. In most cases, the amounts consumed were higher in individuals with higher BMI. However, compared with participants with lower BMI, consumption of cereals, fats, salty snacks, sweets, cake, spreads was lower in this group. Comparison of age groups revealed significantly different amounts consumed of all food groups except for bread, coffee, tea, alcoholic beverages, vegetables and milk and milk products. Gender was the most discriminating variable with regard to the mean food consumption. Percentage of nonconsumers varied across food items. In general, across all food groups, the proportion of nonconsumers was 21.5% on average, with a range of 94.1% (vegetarian spreads) to 0.2% (boiled potatoes). Differences in amounts consumed of most of the cereal items were due to variation in reported portion sizes as well as reported frequencies of intake (Table 3). Mean portion sizes and mean consumption frequencies differed between men and women for all food items of this food group. Differences in reported portion size and reported consumption frequency according to BMI and age groups were also observed, but these were not as consistent as for gender. Taking all 15 comparisons into account, 12 portion sizes differed and 14 consumption frequencies showed differences. Because both portion sizes and frequencies of intake contributed to variation, we separated out the variance due to portion size and frequency. In Table 4, explained variance of food intake is shown for all food groups. On average, 84.0% of variance in food intake was explained by frequency of intake solely (model 1), with the highest values for French fries (95.7%) and the lowest for nuts (71.2%). In general, compared with other food groups, the proportion of variance in intake explained only by fre-

4 PORTION SIZES IN FOOD-FREQUENCY QUESTIONNAIRES 513 TABLE 3 Proportion of participants consuming cereals, reported portion sizes and frequencies of intake in the EPIC-Potsdam Study1,2 Food item Men n 10,596 Women n 16,168 n 14,011 n 12,753 Age 50 y n 12,872 Age 50 y n 13,892 Cereal flakes, grains, muesli A B 52.4 (51.6; 53.2) 47.7 (47.3; 48.2) 49.4 (48.9; 49.9) 48.7 (48.2; 49.3) 50.6 (50.0; 51.1) 47.5 (46.9; 48.0) C (0.081; 0.090) (0.115; 0.123) (0.124; 0.133) (0.077; 0.085) (0.098; 0.106) (0.105; 0.114) Cornflakes, crisps etc. A B 36.4 (35.8; 37.0) 34.3 (33.9; 34.6) 34.4 (34.0; 34.8) 35.7 (35.2; 36.1) 35.8 (35.4; 36.2) 33.6 (33.1; 34.0) C (0.035; 0.040) (0.052; 0.056) (0.052; 0.057) (0.038; 0.042) (0.054; 0.059) (0.037; 0.041) Pasta A B (125.4; 127.0) (108.4; 109.6) (114.4; 115.8) (115.5; 116.9) (123.1; 124.5) (107.1; 108.4) C (0.084; 0.087) (0.098; 0.100) (0.099; 0.102) (0.085; 0.087) (0.110; 0.114) (0.076; 0.078) Rice A B (109.9; 111.3) 97.0 (96.5; 97.5) (100.6; 101.8) (102.8; 104.0) (105.3; 106.6) 98.6 (98.0; 99.2) C (0.067; 0.070) (0.083; 0.086) (0.080; 0.083) (0.073; 0.076) (0.085; 0.088) (0.069; 0.072) Vegetarian dishes A B (133.6; 137.2) (127.1; 129.5) (129.9; 132.5) (128.1; 131.3) (131.5; 134.0) (125.1; 128.3) C (0.008; 0.009) (0.014; 0.016) (0.015; 0.017) (0.007; 0.009) (0.016; 0.018) (0.007; 0.008) 1 Values are (A) percentage of consumers, (B) mean portion sizes (g) and 95% confidence limits and (C) mean frequencies of intake (times/d) and 95% confidence limits. 2 Participants who reported no consumption are excluded. quency of intake was greatest for potatoes. Variance in intake of nuts was explained least by frequency alone. The range of the single R 2 -values for items of one food group in most cases was rather small. Loss of variance due to omission of portion size information ranged between 4.3 and 28.8%. Allowing group-specific portion sizes in regression models could only marginally improve values of explained variance in food intake compared with an overall portion size. Stratification for gender, age and BMI yielded the highest values, but in comparison to model 1, in general 1% of R 2 was gained. Table 5 shows correlation coefficients of nutrient intake between the original FFQ data and the 24-h recall, and the simulated data and 24-h recalls of 134 participants of the validation study. The range of the correlation coefficients was r 0.36 and r 0.62 with a mean value of r 0.52 for FFQ vs. 24-h recall and r 0.38 to r 0.56 with a mean of r 0.48 for simulated FFQ vs. 24-h recall. DISCUSSION Information on individual portion sizes adds a notable amount to the explanation of variance in food intake, but the major part of variance is explained by frequency of intake alone. The latter was already noted by Heady 40 years ago (22). Although we determined that amounts consumed showed gender differences and varied across age and BMI groups, calculations that allowed group-specific portion sizes according to gender, age or BMI did not markedly improve explained variance. The assignment of portion sizes compared with inquiring about variable portion sizes resulted in a moderate decrease in correlation coefficients for nutrient intake between FFQ and repeated 24-h recalls. A limitation of our study in this context was that information on portion sizes and frequency of intake was assessed simultaneously. If necessary, participants could interchange larger portion size with a higher frequency of intake and vice versa. We do not know how frequently participants made use of such substitutions. However, we assume that substitution was rarely present because portion size and frequency options were very detailed. Therefore, these data may be a good approximation of real effects. A further limitation concerns our study population, which was a selected sample out of the general population, probably with higher motivation for epidemiologic research. Generalizability of our results is therefore limited. Results may not be valid for other populations with other dietary habits. Several researchers tried to measure the effect of portion size questions on food intake data by calculating correlation coefficients. Correlations between the same FFQ, inquiring about portion size or applying standard portion sizes, showed values around 0.9 for nutrients (23). In a Danish study, FFQ data with and without individually estimated portion sizes were compared with weighed diet records (24). Mean correlation coefficients for food groups and nutrients changed only slightly, indicating that little extra information could be obtained by additional questions about portion size. Clapp et al. (15) found correlation coefficients of 0.73 to 0.92 for retinol and folacin, respectively. Our findings correspond to these results. Stratification of standard portion sizes according to age and gender has been suggested and is already in practice (15,25). In a recent validation study, the authors stated that low correlation coefficients for nutrient intake could be due to assignment

5 514 NOETHLINGS ET AL. TABLE 4 Variation in food intake explained by frequency of consumption, according to age, gender or body mass index (BMI) in the EPIC-Potsdam Study (n 26,764)1 Food group Overall Age2 BMI3 Sex Age; Sex4 Sex; BMI5 Age; Sex; BMI6 Potatoes Meat Soup Cereals Desserts Bread Fats Eggs Salty snacks Sweets Legumes Fish Coffee, tea Cake Fruits Alcoholic beverages Vegetables Milk, milk products Soft drinks Processed meat Spreads Cheese Sauce Nuts Overall (mean) Values are means. 2 Two age categories ( 50 y, 50 y). 3 Two BMI-categories ( 26 kg/m2, 26 kg/m2). 4 Categories by age and gender. 5 Categories by gender and BMI. 6 Categories by age, gender and BMI. 7 R2 100%. %7 of an overall portion size instead of gender-specific portion sizes (26). However, our empirical results regarding the assignment of group-specific portion sizes indicated a minor benefit with regard to variance in food intake. Furthermore, Willett (11) pointed out that models of disease and diet relationships would always be adjusted for age and gender to account for the confounding effect. The collection of valid individual portion size data requires the individuals to be able to estimate the amounts consumed correctly. However, this seems to be a questionable postulate. A study to validate individual portion size estimates compared FFQ using photos to 14-d weighed food records and revealed only a small relationship between estimated and measured portion size (27). Participants selecting small portion sizes seemed to underestimate and those selecting large portion sizes seemed to overestimate amounts actually consumed (27,28). The existence of a usual portion size for an individual is a further assumption that is implicitly made when inquiring about consumed amounts. The data of the proportion of intraand interperson variability of portion sizes shed doubts on this concept. In a study by Hunter et al. (29), the intra-individual variability in food intake in 61 of 68 items exceeded the interindividual variability. The ratio of intra-individual to interindividual variance was 3.4 on average, indicating a smaller contribution of interindividual variance to total variance in food intake. In another study, variance ratios ranged from 0.67 to 1.60 (27). The inclusion of separate questions inquiring about portion sizes in a FFQ introduces one additional question for each food item into the questionnaire (14), and thus expands the length of the FFQ. In addition to the accuracy of information on food and nutrient intake, questionnaire length has to be considered and, consequently, respondent burden. Questionnaires extended in length by extra nondietary questions and portion size questions resulted in a 20% higher total nonresponse rate compared with short forms, whereas the inclusion of portion size questions alone was not significantly associated with the nonresponse rate (30). A short FFQ including 97 items without questions on portion size except for a few items resulted in a 20-min completion time (31), and response rates for a semiquantitative FFQ were higher than for questionnaires inquiring about portion size (26). However, Subar et al. (32), who designed a questionnaire to be cognitively easier for study participants, concluded that shorter questionnaires are not always better in large-scale epidemiologic settings. In general, information about portion sizes in FFQ is important in measuring variance in food intake, and our findings might have different implications in different research contexts. However, depending on the purpose of the data, the omission of separate portion size questions in favor of a semiquantitative FFQ can be of advantage, especially in large epidemiologic studies in which a questionnaire should be kept simple. Because group-specific portion sizes did not markedly

6 PORTION SIZES IN FOOD-FREQUENCY QUESTIONNAIRES 515 TABLE 5 Correlations between nutrient intake based on the foodfrequency questionnaire (FFQ) with and without variable portion size and based on 24-h recalls, in the EPIC-Potsdam Validation Study1,2 Food group FFQ variable portion sizes vs. 24-h recall3 increase explained variance, the assignment of an overall portion size is recommended. LITERATURE CITED n 134 FFQ assigned portion size vs. 24-h recall4 Energy Total protein Total fat Saturated fatty acids Monounsaturated fatty acids Polyunsaturated fatty acids Cholesterol Total carbohydrates Dietary fiber Carotin Retinol Calcium Iron Nutrient intake was calculated as the mean of repeated 24-h recalls. 2 Values are Spearman correlation coefficients. 3 Original data. 4 Simulated data. 5 P for all correlation coefficients. 1. Willett, W. C. (1994) Future directions in the development of foodfrequency questionnaires. Am. J. Clin. Nutr. 59: 171S 174S. 2. Thompson, F. E. & Byers, T. (1994) Dietary assessment resource manual. J. Nutr. 124: 2245S 2317S. 3. Byers, T. (2001) Food frequency dietary assessment: how bad is good enough? Am. J. Epidemiol. 154: Shils, M., Olson, J. & Shike, M. ed (1994) Modern Nutrition in Health and Disease. Lea & Febiger, Malvern, PA. 5. Kushi, L. H. (1994) Gaps in epidemiologic research methods: design considerations for studies that use food-frequency questionnaires. Am. J. Clin. Nutr. 59: 180S 184S. 6. Hoffmann, K., Kroke, A., Klipstein-Grobusch, K. & Boeing, H. (2002) Standardization of dietary intake measurements by nonlinear calibration using short-term reference data. Am. J. Epidemiol. 156: Liu, K. (1994) Statistical issues related to semiquantitative food-frequency questionnaires. Am. J. Clin. Nutr. 59: 262S 265S. 8. Briefel, R. R., Flegal, K. M., Winn, D. M., Loria, C. M., Johnson, C. L. & Sempos, C. T. (1992) Assessing the nation s diet: limitations of the food frequency questionnaire. J. Am. Diet. Assoc. 92: Block, G. & Subar, A. F. (1992) Estimates of nutrient intake from a food frequency questionnaire: the 1987 National Health Interview Survey. J. Am. Diet. Assoc. 92: Block, G. (2001) Invited commentary: another perspective on food frequency questionnaires. Am. J. Epidemiol. 154: Willett, W. (2001) Invited commentary: a further look at dietary questionnaire validation. Am. J. Epidemiol. 154: Bohlscheid-Thomas, S., Hoting, I., Boeing, H. & Wahrendorf, J. (1997) Reproducibility and relative validity of food group intake in a food frequency questionnaire developed for the German part of the EPIC project. European Prospective Investigation into Cancer and Nutrition. Int. J. Epidemiol. 26: S Bergmann, M. M., Bussas, U. & Boeing, H. (1999) Follow-up procedures in EPIC-Germany data quality aspects. European Prospective Investigation into Cancer and Nutrition. Ann. Nutr. Metab. 43: Willett, W., ed (1998) Nutritional Epidemiology. Oxford University Press, Oxford, UK. 15. Clapp, J. A., McPherson, R. S., Reed, D. B. & Hsi, B. P. (1991) Comparison of a food frequency questionnaire using reported vs standard portion sizes for classifying individuals according to nutrient intake. J. Am. Diet. Assoc. 91: Riboli, E. & Kaaks, R. (1997) The EPIC Project: rationale and study design. European Prospective Investigation into Cancer and Nutrition. Int. J. Epidemiol. 26: S Boeing, H., Korfmann, A. & Bergmann, M. M. (1999) Recruitment procedures of EPIC-Germany. European Investigation into Cancer and Nutrition. Ann. Nutr. Metab. 43: Schofield, W. N. (1985) Predicting basal metabolic rate, new standards and review of previous work. Hum. Nutr. Clin. Nutr. 39: Goldberg, G. R., Black, A. E., Jebb, S. A., Cole, T. J., Murgatroyd, P. R., Coward, W. A. & Prentice, A. M. (1991) Critical evaluation of energy intake data using fundamental principles of energy physiology: 1. Derivation of cut-off limits to identify under-recording. Eur. J. Clin. Nutr. 45: Bohlscheid-Thomas, S., Hoting, I., Boeing, H. & Wahrendorf, J. (1997) Reproducibility and relative validity of energy and macronutrient intake of a food frequency questionnaire developed for the German part of the EPIC project. European Prospective Investigation into Cancer and Nutrition. Int. J. Epidemiol. 26: S Kroke, A., Klipstein-Grobusch, K., Voss, S., Moseneder, J., Thielecke, F., Noack, R. & Boeing, H. (1999) Validation of a self-administered food-frequency questionnaire administered in the European Prospective Investigation into Cancer and Nutrition (EPIC) Study: comparison of energy, protein, and macronutrient intakes estimated with the doubly labeled water, urinary nitrogen, and repeated 24-h dietary recall methods. Am. J. Clin. Nutr. 70: Heady, J. A. (1961) Diets of bank clerks: development of a method of classifying the diets of individuals for use in epidemiologic studies. J. R. Stat. Soc. 124: Block, G., Hartman, A. M. & Naughton, D. (1990) A reduced dietary questionnaire: development and validation. Epidemiology. 1: Tjonneland, A., Haraldsdottir, J., Overvad, K., Stripp, C., Ewertz, M. & Jensen, O. M. (1992) Influence of individually estimated portion size data on the validity of a semiquantitative food frequency questionnaire. Int. J. Epidemiol. 21: Subar, A. F., Midthune, D., Kulldorff, M., Brown, C. C., Thompson, F. E., Kipnis, V. & Schatzkin, A. (2000) Evaluation of alternative approaches to assign nutrient values to food groups in food frequency questionnaires. Am. J. Epidemiol. 152: Subar, A. F., Thompson, F. E., Kipnis, V., Midthune, D., Hurwitz, P., McNutt, S., McIntosh, A. & Rosenfeld, S. (2001) Comparative validation of the Block, Willett, and National Cancer Institute food frequency questionnaires: the Eating at America s Table Study. Am. J. Epidemiol. 154: Haraldsdottir, J., Tjonneland, A. & Overvad, K. (1994) Validity of individual portion size estimates in a food frequency questionnaire. Int. J. Epidemiol. 23: Faggiano, F., Vineis, P., Cravanzola, D., Pisani, P., Xompero, G., Riboli, E. & Kaaks, R. (1992) Validation of a method for the estimation of food portion size. Epidemiology. 3: Hunter, D. J., Sampson, L., Stampfer, M. J., Colditz, G. A., Rosner, B. & Willett, W. C. (1988) Variability in portion sizes of commonly consumed foods among a population of women in the United States. Am. J. Epidemiol. 127: Kuskowska-Wolk, A., Holte, S., Ohlander, E. M., Bruce, A., Holmberg, L., Adami, H. O. & Bergstrom, R. (1992) Effects of different designs and extension of a food frequency questionnaire on response rate, completeness of data and food frequency responses. Int. J. Epidemiol. 21: Wakai, K., Egami, I., Kato, K., Lin, Y., Kawamura, T., Tamakoshi, A., Aoki, R., Kojima, M., Nakayama, T., Wada, M. & Ohno, Y. (1999) A simple food frequency questionnaire for Japanese diet Part I. Development of the questionnaire, and reproducibility and validity for food groups. J. Epidemiol. 9: Subar, A. F., Ziegler, R. G., Thompson, F. E., Johnson, C. C., Weissfeld, J. L., Reding, D., Kavounis, K. H. & Hayes, R. B. (2001) Is shorter always better? Relative importance of questionnaire length and cognitive ease on response rates and data quality for two dietary questionnaires. Am. J. Epidemiol. 153: