International Journal of Epidemiology © International EpWemiological Association 1992

Vol. 21, No. 4 Printed In Great Britain

Influence of Individually Estimated Portion Size Data on the Validity of a Semiquantitative Food Frequency Questionnaire TjfSnneland A (Danish Cancer Registry, Institute of Cancer Epidemiology, Danish Cancer Society, Rosenvaagets Hovedvej 35, PO Box 839, DK-2100 Copenhagen, Denmark), Haraldsd6ttir J, Overvad K, Stripp C, Ewertz M and Jensen O M. Influence of individually estimated portion size data on the validity of a semiquantitative food frequency questionnaire. International Journal of Epidemiology 1992; 21: 770-777. Using data from a Danish dietary validity study, we evaluated the influence of including individually estimated portion size data on the validity of a semiquantitative food frequency questionnaire. A total of 144 subjects, aged 40-64 years were included. Correlation coefficients and classification of subjects into quintiles according to their intake of foods and nutrients were used to compare questionnaire data, with and without individually estimated portion size data, with data obtained by 2 x 7 days weighed diet records. For men, the mean correlation coefficient for food group comparisons was 0.47 when information about individually estimated portion sizes for food items that do not come in natural units was included in the analysis. Using a common average portion size reduced the mean correlation coefficient to 0.45. For women, similar changes were observed (0.36 to 0.35). For nutrients the mean values for the comparison changed from 0.51 to 0.49 and from 0.39 to 0.40 for men and women, respectively. For both men and women the classification into quintiles according to their intake of foods and nutrients showed only minor differences. We conclude that little extra information was gained by including individual portion size information for food items that do not come in natural units. This may reflect that portion sizes are of minor importance compared with frequencies or that the relevant individual portion sizes were not estimated correctly.

The semiquantitative food frequency questionnaire has become a widely accepted method for measuring dietary intake in large scale epidemiological studies. However, it remains to be determined whether the collection of additional information on portion sizes will improve the validity of the method. There is probably little doubt that the incorporation of a portion size into the question on frequency will add clarity for foods which come in natural units such as a slice of bread, a cup of coffee etc. The problem really arises for foods which are conceptually difficult

to quantify, e.g. cooked dishes.1 Furthermore, portion sizes seem to vary less among individuals than frequencies of use and portion sizes may also correlate positively with frequencies.2^ Using data from a Danish dietary validity study,56 we were able to evaluate the importance of including portion size estimates in a semiquantitative food frequency questionnaire,

• Danish Cancer Registry, Institute of Cancer Epidemiology, Danish Cancer Society, Rosenvsengeu Hovedvej 35, PO Box 839, DK-2100 Copenhagen, Denmark. •• Research Department of Human Nutrition, The Royal Veterinary and Agricultural University, Roiighedsvej 25, DK-1958 Frederiksberg C, Denmark. t Institute of Epidemiology and Social Medidne, Aarhus University, HUegh-Guldbergs Gade 8, DK-8O0O Arhus C, Denmark.

' . The « « d y Comprised 144 subjects (59 men and 85 women aged 4 0 - 6 4 years), who filled in a selfadministered Semiquantitative food frequency questionnaire and completed 2 x 7 days of weighed diet , records. For most foods the questionnaire included questions

MATERIAL AND METHODS Subject selection, data collection, and methods used to calculate nutrient intakes have been reported in detail

t Deceased.

on portion sizes. These were estimated as follows: 770

Downloaded from http://ije.oxfordjournals.org/ at Thomas Jefferson University, Scott Library on March 11, 2015

ANNE TJ0NNELAND,* JOHANNA HARALDSDOTTIR," KIM OVERVAD.t CONNIE STRIPP,* MARIANNE EWERTZ* AND OLE M0LLER JENSEN*'*

VALIDITY OF A FOOD FREQUENCY QUESTIONNAIRE

Statistical Analysis Sample means and standard deviations of intake of food groups and nutrients were computed for men and women separately. The Spearman's rank order correlation coefficient was used for food group comparison. Pearson's product-moment correlation coefficients were used to compare nutrient intake from the questionnaire data sets with the diet records. Because the intake of most nutrients was skewed towards higher values, loge transformations were used to improve normality. Correlations for nutrients were also adjusted for total energy intake.9 The degree of

misclassification was assessed by dividing food group and nutrient intakes into quintiles. RESULTS For most food groups lower mean intakes were found in the food frequency questionnaire than in the diet records (Table 1). Individual portion size information (A) did not result in consistently higher mean values corresponding more closely to the diet records than the means obtained using standard portions (B), except for vegetables. The mean intake of nutrients calculated from the food frequency questionnaire using photo portion size information (A) and by replacement with standard portion sizes (B) is presented in Tables 2 and 3 for men and women separately. Since the questionnaire covered known proportions of total nutrient intake, means were adjusted up to 100%, as previously described.3'6 For men, total energy, protein and fat intake in data set A showed values closer to the diet records than in data set B. For carbohydrate and some of the micronutrients the means of data set B were closer to the diet records. For women little difference was found between data sets A and B. As to food groups, the mean correlation coefficient for men was 0.47 and 0.45 for the comparison of the dietary records and data sets A and B, respectively (Table 4). The values were slightly higher for data set A except for meat. For women, quite similar correlation coefficients were obtained in the two data sets with the mean correlation changing from 0.36 for data set A to 0.35 for data set B. Correlation coefficients from the comparison of nutrient intakes estimated from the food frequency questionnaire and from the diet records are presented in Table 5. Overall seven nutrients showed better values in data set A than B, while four nutrients showed poorer values in A than B, and two nutrients showed unchanged values in A and B. This was consistent for both men and women. Mean correlation coefficients for data sets A and B weTe in men 0.51 and 0.49 and in women 0.39 and 0.40, respectively. Classification of subjects into quintiles according to their intake of foods and nutrients was used to evaluate the agreement between categorizations based on the diet records and the food frequency questionnaire (data sets A and B). For nutrients, on average, about 70% of the subjects were classified in the same and adjacent quintile in both data sets A and B. For food groups, the greatest difference between A and B was found for table spread for men, where the agreement decreased from 71% to 64%. Similarly for

Downloaded from http://ije.oxfordjournals.org/ at Thomas Jefferson University, Scott Library on March 11, 2015

I. By selecting a photo (with four options of portion sizes). This was used for meat, fish, potatoes, cooked vegetables, mixed salad, shredded vegetables and butter/margarine on bread; ten series of photos were induded in the questionnaire. For similar food items, the same photo was used to estimate several portion sizes i.e. photos of hamburger patties were used to calculate the portion size of meat balls, and photos of mixed vegetables were used for all types of cooked vegetables. II. The second way to estimate portion sizes was by multiplying the number of units consumed (slices of bread, glasses of milk etc.) assuming a standard size for a slice and a glass. This was used for foods which usually come in such natural units. III. For the remaining foods (cakes, desserts etc.) a single standard portion size was used. The standard portion sizes and the four different portion sizes used to develop the colour photographs were derived from the 1985 Danish National Dietary Survey.7'8 In this survey, the study subjects described their portion sizes using food models and photographs in a diet history interview. The four colour photographs were constructed to cover the whole range of portion sizes from the National Survey. Two sets of data were created. Data set A included all available information about portion sizes, while in data set B the information from the photos was ignored and instead a standard portion equal to the median portion from the National Survey7'8 was used. In data set B, the individual portion indications for foods like bread, milk and fruit, which come in natural units, were maintained. For comparisons at the food group level, the food items from both the food frequency questionnaire and the diet records were grouped. In the food groups individually estimated portion size information was available for table spread, meat, fish, potatoes and vegetables. Certain food items occurring in the dietary records with no equivalent question in the food frequency questionnaire were omitted.

771

772

INTERNATIONAL JOURNAL OF EPIDEMIOLOGY

TABLE 1 Mean and standard deviation (SDJ daily Intake of food groups estimated by the food frequency questionnaire using photo portion size information (A), or standard portion sizes (B) compared with diet records Women (No. - 85)

g/day

g/day

Data set A

Data let B

Diet records

Dati set A

Data set B

Diet records

Mean (SD)

Mean (SD)

Mean (SD)

Mean (SD)

Mean (SD)

Mean (SD)

18 (13)

19 (11)

31 (15)

11 ( 9)

12 ( 9)

19(11)

Meat

138 (74)

121 (51)

158 (63)

91 (51)

94 (46)

105 (37)

Fun

29 (23)

33 (27)

33 (30)

23 (18)

28 (19)

21 (17)

115 (74)

93 (48)

96 (73)

74 (49)

89 (49)

53 (29)

75 (50)

56 (34)

108 (62)

95 (94)

71 (51)

104 (54)

Butter/margarine

Potatoes Vegetables

TABLE 2 Mean and standard deviation (SD) of daily intake of nutrients for men estimated by the food frequency questionnaire using photo portion size information (A) or standard portion sizes (B) and by weighed diet records

Nutrient

Mean*

Total energy* (kj)

9996

Diet records

Data set B

Data set A

(SD)

(2149)

Mean* 9532

(SD)

Mean

(SD)

(1993)

10 178

(2075)

Protein (g)

74.9

(13.2)

72.0

(11.8)

Total fat (g)

87.5

(15.5)

80.5

(13.7)

Saturated fat (g)

37.6

(8.1)

34.2

(7-3)

Polyunsaturated fat (g)

14.0

(4.6)

13.3

(4.0)

81.5 101 39.0 15.2

(13.1) (15.8) (8.1) (5.0)

Cholesterol (total) (g)

305

(86)

290

(82)

496

(118)

Carbohydrate (g)

263

(31)

254

(28)

231

(41)

Sucrose (g)

116

(30)

114

(29)

Dietary fibre (g) Vitamin A (meg) Vitamin E (mg) Vitamin C (mg) Calcium (mg) Iron (mg)

23.2 1205 5.6

(3.9) (403) (2.0)

21.5 1079 5.4

(3.4) (350) (1.7)

115

(92)

111

(91)

1352

(388)

1330

(375)

14.2

(2.4)

13.7

* Mean values from food frequency questionnaire were adjusted to 100% of nutrient intake. b Alcohol included.

(2.5)

88.0 21.1 1090

(31.2) (5.2) (465)

6.7

(2.4)

75.5

(60.0)

1269 15.1

(416) (2.7)

Downloaded from http://ije.oxfordjournals.org/ at Thomas Jefferson University, Scott Library on March 11, 2015

Food group

Men (No. - 59)

773

VALIDITY OF A FOOD FREQUENCY QUESTIONNAIRE

TABLE 3 Mean and standard deviation (SD) of daily intake of nutrients for women estimated by the food frequency questionnaire using photo portion size information (A) or standard portion sizes (B) and by weighed diet records Data set A

Nutrient

Mean 1

Total energy" (kJ)

6868

Data set B (SD)

Mean*

(1762)

6906

Diet records (SD)

(1729)

Mean

7017

(SD)

(1578)

(7.9)

58.4

(7.4)

62.5

(9.1)

Total fat d )

62.1

(9.7)

62.0

(9.1)

71.7

(8.9)

Saturated fat (g)

26.6

(6.0)

26.3

(5.4)

28.3

(4.9)

11.0

(2.8)

10.9

Poryunsaturated fat (g)

(3.0)

Cholesterol (total) (g)

249

(56)

255

Carbohydrate (g)

198

(23)

199

Sucrose (g)

(27.0)

89.5

(4.1)

18.2

Dietary fibre (g) Vitamin A (meg)

4.7

Vitamin C (mg) Calcium (mg)

1188

(1.4)

4.7

(2.6)

(54)

(98)

(22)

162

(24)

(26.5) (3.5) (421) (1.4)

60.9 15.4 1049 5.2

(18.4) (4.7) (544) (1.1)

110

(66)

110

(66)

73

(34)

1059

(237)

1050

(233)

1046

(241)

11.0

Iron (mg)

17.8

(690)

1300

Vitamin E (ing)

88.0

9.8 334

11.0

(1.6)

(8.9)

11.1

(1.8)

* Mean values from food frequency questionnaire were adjusted to 100% of nutrient intake. b Alcohol included.

TABLE 4 Spearman correlation coefficients for comparison of the dietary intake of food groups from diet records and either photo portion size information (A) or standard portion size food frequency questionnaire information (B)

Men (No. = 59)

Food group

Women (No. - 8 5 )

A

B

A

B

Butter/margarine

0.51

0.46

0.24

0.28

Meat

0.31

0.37

0.16

0.14

Fish

0.49

0.42

0.46

0.44

Potatoes

0.40

0.38

0.46

0.37

Vegetables

0.63

0.60

0.50

0.51

women, the value for potatoes decreased from 74ge transformed values.

for other foods the options were obviously not optimal as practically all the answers fell in two options (e.g. hamburger patties for women, potatoes for men). DISCUSSION This study showed that little extra information was gained by including individual portion size information for food items that do not come in natural units in a self-administered semiquantitative food frequency questionnaire. This agrees well with several studies reported in the literature but contradicts others. Hernandez-Avila et al.i0 calculated the intakes of 18 nutrients from a semiquantitative food frequency questionnaire with and without portion size information for foods which do not come in natural units, and compared it with intakes obtained from a 1-week diet record. After adjustment for energy intake mean correlation was 0.54 with portion size data and 0.53 without. The number of foods which were individually estimated was not specified. The portion size data were obtained by a personal interview, using food models, after the questionnaire had been filled in.

Downloaded from http://ije.oxfordjournals.org/ at Thomas Jefferson University, Scott Library on March 11, 2015

A

Margetts et al.u found no improvement in the association between nutrient intakes from a 24-hour diet record and a food frequency interview, taking into account individually estimated portion sizes for three foods (bread, milk and potatoes) out of a total of 65 food items. Two studies have evaluated questionnaires aiming at one specific nutrient only. Cummings et al.n found a somewhat reduced correlation between calcium intake from a food frequency questionnaire and a 7-day diet record, when all portions were rated as 'medium' instead of an individual choice between the specifications 'small', 'medium', and 'large' portions (correlation decreased from 0.76 to 0.64). In this study Cummings et al. also found that using an inappropriate way of indicating individual portions reduced the performance of the questionnaire below the value of using the medium portion only (when ounces and cups were used for indication, correlation was reduced to 0.49). The other one-nutrient study was done by Samet et al.1 They did not do a validity study but compared vitamin A intakes calculated from a food frequency interview with and without individual portion estimates (using food models). They found correlations of almost 0.90 and exact agreement between quartile classifications for 64% of the participants. Hankin et alP examined the validity of food intakes for frequencies alone and frequencies with individual portion estimates, compared to one 7-day diet record. Photos were used for estimating portions. Slightly higher mean correlations were found for the quantitative estimates than for frequencies alone (0.59 compared to 0.55). Chu et a/.14 did not do a validity study but compared frequency and quantitative data (using photos), and did this for both food and nutrient intakes. Their regression analysis showed that the results were not identical and therefore not interchangeable. Using this strict criterion Chu et al. concluded that frequencies alone could not be relied on to yield the same dietdisease associations as would the quantitative data—for food groups or for nutrients. In two recent publications Block et a/.13 and Jackson et al.16 found what they judged as a nontrivial increase in the relative validity of nutrient intakes, when taking individually estimated portion sizes into account. In the study of Block et al.ls 16 out of 18 nutrients showed somewhat higher correlations with a 12-day diet record, when individually variable (and age-sexspecific) indications of portion sizes as 'large', 'medium' and 'small' were used instead of only one medium size. The average correlation was increased from 0.49 to 0.54.

775

VALIDITY OF A FOOD FREQUENCY QUESTIONNAIRE

TABLE 6 Agreement between food group distribution quintilesfrom diet records and photo portion size information (A) or standard portion size (B) food frequency questionnaire (FFQ) Lowest quintflc on food records (B)

Lowen quintile on food records(A) Lowest qulntile on FFQ (A) (ft)

Lowest two quint iles on FFQ (A) (ft)

Highest quintile on FFQ (A) (ft)

Lowest quintflc on FFQ (B) (ft)

Lowest two quint iles on FFQ (B) (ft)

Highest quintile on FFQ (B) (ft)

Butter/margarine

42

75

0

50

58

0

Meat

42

67

8

42

67

0

Fish

75

92

0

67

83

0

Potatoes

50

75

0

50

67

8

Vegetables

50

83

0

58

83

0

Butter/margarine

24

41

18

24

41

12

Meat

24

47

24

24

42

12

Fish

41

59

6

41

59

12

Potatoes

53

53

12

35

53

6

Vegetables

53

59

53

71

0

Food group Men (No. - 59)

0

TABLE 7 Agreement between lowest quintile of nutrient Intake from diet records and either photo portion size information (A) or standard portion size (B) food frequency questionnaire (FFQ) in men (No. •= 59)

Highest quintile on FFQ (B)

Lowest two quintiles on FFQ (A) (ft)

Highest quintile on FFQ (A) (ft)

Lowest quintflc on FFQ (B)

Nutrient

Lowest quintile on FFQ (A) (ft)

(ft)

Lowest two quintues on FFQ (B) (ft)

Protein

50

75

0

42

67

8

Total fat

42

83

0

50

67

0

Saturated fat

42

75

0

42

75

0

Polyunsaturated fit

58

75

17

50

67

8

Cholesterol (total)

42

67

17

42

67

17

Carbohydrate

25

50

8

33

67

8

Sucrose

42

75

8

50

67

8

Dietary fibre

33

58

8

17

50

17

Vitamin A

25

67

17

33

67

8

Vitamin E

33

67

17

50

67

17

(ft)

Vitamin C

58

67

8

50

75

8

Calcium

42

83

0

75

92

0

Iron

75

83

8

50

83

8

Mean value

44

71

8

45

70

8

Downloaded from http://ije.oxfordjournals.org/ at Thomas Jefferson University, Scott Library on March 11, 2015

Women (No. - 85)

776

INTERNATIONAL JOURNAL OF EPIDEMIOLOGY

TABLE 8 Percentage of study subjects usint each of the photo portion sizes in the food frequency questionnaire

Men %

Women ft

Hamburger patties

80 160 240 320

20 56 20 4

34 64 2 0

Meat stew

75 200 300 500

10 13 37 40

11 37 45 7

Liver, fried

50 75 150

24 27 34 15

15 64 20 1

Fish, fried

75 150 225 300

29 61 4 6

46 51 3 0

Poutoes, boiled

50 100 200 350

0 3 60 37

4 31 61 4

Mixed vegetables, boiled

40 80 150 250

2 11 58 29

1 13 75 11

Mixed salad

25 50 80 130

2 10 41 47

3 11 55 31

Carrots, shredded, raw

25 50 100 150

20 47 27 6

12 38 39 11

too

In the study of Jackson et al." only the intakes of fat, calcium and dietary fibre were calculated from the food frequency questionnaire, and compared to intakes measured by a dietary history interview. For fat and calcium agreement was better when individual portions were included in the calculations (r = 0.45 instead of 0.30 for fat) than when only medium portions were used for all foods. For dietary fibre, however, the correlations did not increase (0.49 instead of 0.51). Most correlation coefficients obtained in the present study were between 0.4 and 0.5. In a cohort study a true relative risk of 3.0 will show up as a relativeriskof 1.6 if the validation study had produced a correlation coefficient of 0.5. For a correlation coefficient of 0.4 the relative risk would have been estimated to be 1.5. In the present data the differences between the correla-

Downloaded from http://ije.oxfordjournals.org/ at Thomas Jefferson University, Scott Library on March 11, 2015

Portion size (g)

Nutrient

tion coefficients obtained between data sets A and B and the diet records were almost all less than 0.1. Thus, most authors report minor changes in the relative validity of a food frequency questionnaire when standard portions are used instead of individually estimated portion sizes. This is remarkable, especially when the large differences between the studies are taken into account. First, the number of foods for which alterations were introduced, varied from all foods16 to only three out of 65 foods." Obviously, the largest impact would be expected where standard portions for all foods were substituted by individual estimates for all foods. Secondly, the method used for the individual portion estimates varies from ticking off a written description in the questionnaire (small, medium, large), to a personal interview using photos and models showing several options of size. Thirdly, the questionnaires varied from covering 90% of the total energy intake" to only aiming at one specific nutrient.2 This lack of influence of individual portion estimates can be interpreted as either portion sizes being unimportant, compared to frequencies, or that individual usual portion sizes have not been properly measured in these studies. Some studies on the relationship between portion sizes and frequencies indicate that portions may vary less among individuals than do frequencies of use, and that portions may also be positively correlated to frequencies.2"4 Hunter et al." found that the day-to-day variation in portion size, for the same individual, was much larger than the variation in portion size between individuals (in a group of 34-59 year old women). These results suggest that using standard portions instead of individual estimates may not introduce a large error. As to the estimates of individual portions it is obvious that an inappropriate measurement may do more harm than good.12 Further, individual portion estimates can only give results which are different from calculations with standard portions if there is a fair degree of variation in the estimates of portions. Therefore, it is important that the portion size photos, food models or descriptions cover the range of portion sizes used by the participants. In our study the photo options did not cover the variation sufficiently for all food items. A standard portion size would therefore have little impact on the relative intake, compared with the wide variation in reporting of frequencies. This problem of variation of the individual estimates has received little attention, except for Chu et al.1* who reported that portion sizes tended to cluster in the categories bounded by one-half to twice the one-size food model used. Several other studies had three different options of portion sizes, leading to tendency

VALIDITY OF A FOOD FREQUENCY QUESTIONNAIRE 1

ACKNOWLEDGEMENTS The study was funded by the Danish Cancer Society. We are grateful to Ms Anette Feddersen and Ms Katja Boll for technical assistance. REFERENCES 1

2

3

Cuthrie H A. Selection and quantification of typical food portions by young adults. J Am Diet Assoc 1984; 84: 144

Influence of individually estimated portion size data on the validity of a semiquantitative food frequency questionnaire.

Using data from a Danish dietary validity study, we evaluated the influence of including individually estimated portion size data on the validity of a...
469KB Sizes 0 Downloads 0 Views