Preventive Medicine 38 (2004) S24 S33 www.elsevier.

com/locate/ypmed

Validity and reliability of a behavior-based food coding system for measuring fruit, 100% fruit juice, vegetable, and sweetened beverage consumption: results from the Girls Health Enrichment Multisite Studies
Karen W. Cullen, Dr.P.H., R.D., L.D., a,* John H. Himes, Ph.D., M.P.H., b Tom Baranowski, Ph.D., a Janet Pettit, b Mary Stevens, b Deborah Leachman Slawson, Ph.D., R.D., c Eva Obarzanek, Ph.D., R.D., d Maureen Murtaugh, Ph.D., R.D., b Donna Matheson, Ph.D., e Wanjie Sun, M.S., f and James Rochon, Ph.D. g
a

Department of Pediatrics, Childrens Nutrition Research Center, Baylor College of Medicine, Houston, TX 77030-2600, USA b Division of Epidemiology, School of Public Health, University of Minnesota, Minneapolis, MN 55454, USA c University of Memphis, Memphis, TN 38152, USA d Division of Epidemiology and Clinical Applications, National Heart, Lung, and Blood Institute, Bethesda, MD 20892, USA e Center for Research in Disease Prevention, Stanford University, Stanford, CA 94305, USA f George Washington University Biostatistics Center, Rockville, MD 20852, USA g Duke Clinical Research Institute, Durham, NC 27715, USA

Abstract Background. This paper presents the rationale, reliability, and validity of a behavior-based food coding system for measuring fruit (F), juice (J), vegetable (V), sweetened beverage, and water consumption in children. Methods. Coding algorithms for FJV, sweetened beverages, and water were developed for use with the Nutrition Data System for Research (NDS-R). Two hundred and ten 8- to 10-year-old African American girls at four field centers completed two 24-h dietary recalls at baseline and at 12 weeks follow-up after a weight gain prevention intervention. Differences in mean baseline consumption of selected food variables and other selected nutrients across the four field centers were analyzed. Intraclass correlation coefficients (ICCs) for reliability across 2 days of food recalls and 12-week test retest reliability correlations were calculated. For the purposes of this paper, nutrient intake estimates were considered construct validators of food intake, and validity was assessed by correlating the coded food variable servings with nutrient intake. Results. ICCs varied from zero (0.001 for beta carotene equivalents) to moderate (0.44 for sucrose), indicating substantial instability in consumption or reporting. Twelve-week test retest correlations were slight to moderate (0.09 for lycopene to 0.49 for folate). FJV consumption was negatively related to percent energy from fat (r = 0.28; P = 0.001) and positively related to other nutrients. Sweetened beverage consumption was positively related to energy, sucrose, fructose, and vitamin C consumption. Conclusions. This behavior-based food coding system demonstrated construct validity among 8- to 10-year-old African American girls and can measure the desired food groups. D 2004 The Institute For Cancer Prevention and Elsevier Inc. All rights reserved.
Keywords: Obesity; Prevention; Female; African American; Children; Diet; Fruit; Vegetables; Juice; Sweetened beverages; Water; Validity; Reliability

Introduction Healthy dietary behaviors such as increased consumption of fruit (F), 100% fruit juice (J), and vegetable (V) and reduced consumption of sweetened beverages have been related to lower chronic disease risk [1 3] and lower body weight [4 6]. Assessing the impact of behavior change

* Corresponding author. Department of Pediatrics, Childrens Nutrition Research Center, Baylor College of Medicine, 1100 Bates Street, Houston, TX 77030-2600. Fax: +1-713-798-6764, +1-713-798-7098. E-mail address: kcullen@bcm.tmc.edu (K.W. Cullen).

0091-7435/$ - see front matter D 2004 The Institute For Cancer Prevention and Elsevier Inc. All rights reserved. doi:10.1016/j.ypmed.2003.05.001

K.W. Cullen et al. / Preventive Medicine 38 (2004) S24S33

S25

programs focusing on FJV or sweetened beverages requires valid and reliable measurement of dietary intake. Several coding systems currently exist for measuring FJV consumption, and the inclusion of FJV items to summarize intakes reflects the emphases of programs with which they are associated [7 10]. Targeting particular fruits and vegetables for inclusion in coding systems is appropriate and should depend on the research question [8,9] because it can influence whether the project is determined to be a success [8,11]. Similarly, the coding procedure used for sweetened beverages should reflect the research question and be clearly defined. Behavior-based coding systems are designed to measure targeted dietary changes. Thus, a major difference between food coding methods is the definition of particular foods and preparatory practices reflected within composite categories of FJV. For example, the 5 A Day program, whose goal is to encourage behavior change for healthy diets that are lower in fats and sweets and higher in FJV, discounts condiments (e.g., catsup), high-fat and sweetened fruits (e.g., apple pie), and high-fat vegetables (e.g., potato chips and French fries) [9]. Only 100% fruit juice was counted within the juice composite. Other, especially analytic epidemiologic, studies measure total nutrient intake to assess nutrient chronic disease relationships and thereby often count all food sources of the nutrients of interest for inclusion in coding systems. The Girls Health Enrichment Multisite Studies (GEMS), conducted at four field centers, was a collaborative effort designed to develop and evaluate interventions to prevent excessive weight gain among 8- to 10-year-old African American girls. The major dietary goals of three of the GEMS interventions were to increase FJV and water consumption and to decrease sweetened beverage consumption. The other intervention primarily targeted physical activity and had minimal emphasis on dietary change. The GEMS investigators selected a behavior-based coding system approximating the national 5 A Day guidelines [9] and developed coding algorithms for FJV for use with the Nutrition Data System for Research (NDS-R) software (version 4.02_30, 1999, Nutrition Coordinating Center, University of Minnesota, Minneapolis, MN). New coding algorithms were also developed for sweetened beverages and water. This report describes the procedures used for developing the coding system to define servings of FJV, sweetened beverages, and water and examines these derived food serving composite variables for their reliability and validity.

Memphis; University of Minnesota, Minneapolis; and Stanford University, Palo Alto. A coordinating center, located at the George Washington University Biostatistics Center, provided support and coordination for key study activities. The National Heart, Lung, and Blood Institute sponsored the program and their staff also participated. A total of 210 African American girls, age 8 10 years, were randomly assigned to an active intervention (n = 115) or comparison group (n = 95). Each field center designed and developed its own 12-week intervention targeting a healthy diet and/or engaging in physical activity. Girls assigned to comparison groups were given health education materials or counseled on topics other than diet and physical activity. All four field centers had several common key measures, including dietary intake, which were measured the same way across field centers. Outcome measurements were taken at baseline, just before randomization, and at follow-up, soon after the 12week intervention period concluded. All parents or guardians gave written informed consent, and girls gave their assent to participate in the study. Each field centers study was approved by its IRB. Other details of the design are described elsewhere [11]. Dietary assessment methods At baseline, each girl completed two 24-h dietary recalls that were administered 1 2 weeks apart. Recalls covering one weekend day and one weekday were encouraged but not required. GEMS staff obtained the information and recorded it using a laptop computer and NDS-R software. The first interview was conducted in person with online interview prompts and a food amounts booklet available to provide dimensional and volume reporting aids for prompting the girl to recall the foods and amounts eaten. The second interview was conducted by telephone, also using NDS-R along with a food amounts booklet given to the girls at the first interview. The telephone interview method was used to minimize participant burden of traveling to some data collection place for the second recall. According to NDSR procedures, a multiple-pass approach was utilized to provide the girls with several opportunities to recall food and beverage consumption for the previous day. An initial brief listing of meals and foods was entered into the computer. This was followed by prompts for additional foods not previously recalled and for more detailed information regarding each food listed. Additional foods consumed were often remembered and recorded during the probing. After probing, the data collector read what was recorded and queried for verification and completeness. At each center, a lead nutritionist reviewed each recall and sent all to NCC. Another review was conducted and questionable values were marked and the data returned to each center. The final data cleaning procedure was a reconciliation between the NCC and each center and this data file was used in all analyses. After the 12-week intervention was concluded, the girls completed another pair of dietary recalls using the same procedures as during baseline. For analyses,

Methods Study design Pilot studies were conducted at four field centers, located at Baylor College of Medicine, Houston; University of

S26

K.W. Cullen et al. / Preventive Medicine 38 (2004) S24S33

Table 1 Description of GEMS behavior-based food groups Category Fruit Group name Inclusion 1. Fresh, frozen, cooked, canned, and dried 2. Caramel apple 3. Fruit in recipes (e.g., salads) 4. Raisins in cereal 5. Jell-O with fruit Exclusion 1. Baked goods, desserts, and pies 2. Fruit other than raisins in cereal 3. Avocado 4. Maraschino cherries 5. Fruit in cookies 6. Granola bars with fruit 7. Trail mix 8. Banana chips 9. Fruit leather or fruit roll-ups 10. Fruit in candy (e.g., chocolate or yogurt covered raisins) 11. Fruit in ice cream and ice cream recipes 12. Fruit in commercial entrees 13. Fruit in jam, jelly, and marmalade 1. Drinks with <100% juice 2. Frozen concentrate 3. Fruit nectars 4. Cranberry drinks 5. Not consumed as a beverage (i.e., part of a recipe 6. <100% fruit juice bars 1. Drinks with <100% juice 2. Not consumed as a beverage (i.e., part of a recipe) 1. Fried and/or breaded (e.g., home fries, hash browns, pan-fried potatoes) 2. French fries 3. Tater tots 4. Cranberry orange relish and other relish 5. Potato skins 6. Potato chips and other snack chips 7. Catsup 8. Onion rings 9. Olives 10. Pickles and pickled vegetables (e.g., sauerkraut) 11. Tomato sauce 12. Tomato-based sauce (e.g., spaghetti and pizza sauce) 13. Tomato paste 1. Soy-based desserts (e.g., Toffutti) 2. Textured vegetable protein (TVP) 3. Soy nuts 4. Products with TVP (e.g., veggie burgers, hamburger with TVP) Serving size 1. Fresh, frozen, canned, or cooked = 1/2 cup chopped or default density 2. One medium piece when appropriate (e.g., apple, orange) 3. Dried = 1/4 cup 4. Cereal = to be determined (e.g. 1 cup cereal = 1/4 cup raisin) Unit serving/day

Fruit juice

1. 100% juice (sweetened or unsweetened) 2. Consumed as a beverage 3. 100% juice bars

6 fluid ounces

serving/day

Vegetable juice

Vegetable

1. 100% juice (sweetened or unsweetened) 2. Consumed as a beverage 1. Raw, frozen, cooked, and canned 2. Vegetables in recipes 3. Soup 4. Salsa 5. Immature lima beans

6 fluid ounces

serving/day

1. Raw, cooked, or canned = 1/2 cup chopped or default density 2. Raw leafy vegetables = 1 cup 3. Salsa = 1/2 cup 4. Potato = 1 medium

serving/day

Legume

1. Dried beans 2. Mature lima beans 3. Tofu 4. Refried beans 5. Tempeh 6. Beans in sauce

1. Cooked dry beans = 1/2 cup 2. Tofu + tempeh = 85 g 3. Refried beans = 1/2 cup 4. Beans in sauce = 1/2 cup

serving/day

K.W. Cullen et al. / Preventive Medicine 38 (2004) S24S33 Table 1 (continued) Category Water-based beverages Group name Sweetened soft drinks Inclusion 1. Sweetened carbonated drink 2. Consumed as a beverage Exclusion 1. Artificially sweetened carbonated drinks 2. Ice 3. Not consumed as a beverage (i.e., part of a recipe) 1. Artificially sweetened noncarbonated drinks 2. Dry powder 3. Ice 4. Not consumed as a beverage (i.e., part of a recipe) 1. Artificially sweetened sports drinks 2. Dry powder 3. Ice 4. Not consumed as a beverage (i.e., part of a recipe) 1. Unsweetened or artificially sweetened tea 2. Dry powder 3. Ice 4. Not consumed as a beverage (i.e., part of a recipe) 1. Unsweetened or artificially sweetened coffee 2. Dry powder 3. Coffee drinks 4. Not consumed as a beverage (i.e., part of a recipe) 1. Unsweetened or artificially sweetened water 2. Ice 3. Not consumed as a beverage (i.e., part of a recipe) Serving size 8 fluid ounces Unit

S27

serving/day

Water-based beverages

Sweetened fruit drinks

1. Sweetened noncarbonated drink 2. <100% juice 3. Consumed as a beverage 1. Sweetened sports drink (e.g., Gatorade, Powerade) 2. Consumed as a beverage

8 fluid ounces

serving/day

Water-based beverages

Sweetened sports drinks

8 fluid ounces

serving/day

Water-based beverages

Sweetened tea

1. Sweetened tea 2. Consumed as a beverage

8 fluid ounces

serving/day

Water-based beverages

Sweetened coffee

1. Sweetened coffee 2. 100% coffee 3. Consumed as a beverage

8 fluid ounces

serving/day

Water-based beverages

Sweetened water

1. Sweetened water 2. Consumed as a beverage

8 fluid ounces

serving/day

the 2 days of collection were averaged for each assessment period, baseline, and postintervention, respectively. The University of Minnesota Nutrition Coordinating Center staff trained at least one dietary assessment staff from each field center centrally; subsequently, all other measurement staff were trained by the NCC-trained staff and certified by NCC after passing standard certification tests. Behavioral food coding system Decisions regarding food grouping were based on the content and intentions of the interventions. In this way, changes in consumption could be linked to the intervention effects. Fruit. Only fruits that appeared like fruits when eaten, in the fresh, canned, or frozen forms, were included in the

coding system. Fruits from high-fat and sugar-containing dishes such as fruit in baked desserts and baked grain products (e.g., pie, granola bar), fruits that had a high concentration of added sugar (e.g., jam, roll-up, fruit leather), and deep-fried fruit chips (e.g., apple chips, banana chips) were not included (Table 1). Fruits suspended in flavored gelatin and raisins in cereals were counted. Juice. Only 100% fruit or vegetable juices in the fresh, canned, or frozen forms were included (Table 1). Amounts less than 100% dilute the nutritional value of fruits and vegetables and were not targeted in this intervention. Vegetables. Only vegetables that appear like vegetables, or vegetable pieces, in the fresh, frozen, or canned forms, were included (Table 1). Vegetables high in fat like olives and avocados were excluded. Fried vegetables with high-fat

S28

K.W. Cullen et al. / Preventive Medicine 38 (2004) S24S33

content were not counted (e.g., French fries, hash browns, other fried potato products, fried onion rings). Tomato paste and sauce in mixed dishes as well as vegetable condiments were excluded because they are found mostly in foods that were not promoted in the interventions (e.g., pizza, catsup). Within mixed dishes, the amount of each component vegetable was included to provide a total number of servings of vegetable for the entire amount eaten (e.g., carrots, potatoes, and onion in beef stew.) Legumes were defined as mature beans and soy products and were considered vegetables. However, they were categorized separately from the vegetable group and only included in the summary composite of all fruits, juices, and vegetables. Textured vegetable protein products (e.g., veggie burgers) were not counted as legumes because they are processed protein products. Vegetables with sauces and cheese (e.g., baked potato with toppings, broccoli-cheese casserole) were counted since the interventions focused on substituting lower fat cheeses and sauces for the typical higher fat versions. Sweetened beverages. The sweetened beverages category was operationalized as water-based beverages. The interventions encouraged girls to choose water and thereby hopefully decrease sweetened beverage consumption (Table 1). The coding categories were soft drinks (carbonated), fruit drinks (noncarbonated), sports drinks, sweetened tea, sweetened coffee, and sweetened water. Water. Noncarbonated, nonsweetened water from the tap or in bottles that was consumed as a beverage was counted. Hypotheses. Nutrients included in the analyses were selected to relate to the behavioral coding of food groups (F, J, V, water, and sweetened beverages) in specific directions and included total energy intake, percent energy from fat, sucrose, fructose, dietary fiber, folate, vitamin C, beta carotene equivalents, and lycopene. Support for the construct validity of the food group coding system would be provided in two ways: the presence of a significant correlation between a food group and nutrients related to the food group (e.g., between vegetables and beta carotene equivalents; fruits and vitamin C) and by dietary displacement (e.g., a negative relationship between water and sweetened beverages or between F, J, V, and fat intake). The hypotheses included the following: (1) F, J, and V consumption would be positively related to consumption of energy, dietary fiber, folate, vitamin C, beta carotene, and lycopene (dietary correlation) and negatively related to percentage of energy from fat (dietary displacement); (2) sweetened beverage consumption would be positively related to energy intake and the major components of the sugars, sucrose, and fructose (dietary correlation) and negatively related to dietary fiber, folate, vitamin C, beta carotene, and lycopene (dietary displacement); and (3) water consumption would be negatively related to energy, sucrose, fructose, and percent energy from fat (dietary displacement).

Data analysis. Demographic characteristics were compared between the four centers using chi-squares for categorical variables and ANOVA for continuous variables. Mean daily nutrient totals for all girls at baseline were calculated, averaged across the 2 days, and compared between field centers using ANOVA. Mean daily consumption of all food group servings for all girls at baseline was calculated using GEMS behavior-based food coding system and compared between field centers using ANOVA. Intraclass correlation coefficients (ICCs) were calculated across the 2 days of baseline dietary intake for all food categories to assess reliability across days of intake for each girl. Differences among field centers were assessed using Fishers Z transformation and chi-square test statistic (n = 210 girls). Bivariate plots of Days 1 and 2 nutrient and food group consumption values were examined for outliers and potential bias between reporting days. Mean daily nutrient totals and food group servings were calculated for girls in the comparison groups at follow-up. Test retest reliability was assessed by calculating Spearman correlations for all food groups and nutrient variables between the mean of the two recalls at baseline and mean of the two recalls at follow-up for girls in the comparison groups only. Spearman correlations were used because not all variables were normally distributed. To assess construct validity or how well the food coding system worked, Spearman correlation coefficients were calculated between all food group servings and nutrients at baseline. Validity correlations were first conducted separately by field center and compared between centers using Fishers Z transformation and the chi-square test statistic. The validity correlations did not consistently differ by field center. Therefore, data were pooled across the sites for this report. Variables with significant site differences in

Table 2 Baseline characteristics of randomized girls Baylor, n = 35 Mean (SE) Girl age*** Parent education (%)*** Some high school High-school graduate Tech school or some college College or graduate degree Household income (%) < US$20,000 US$20,000 $39,999 z US$40,000 Body mass index (kg/m2)*** 8.0 (0.1) 0.0 3.0 3.0 93.9 Memphis, n = 60 Mean (SE) 8.9 (0.1) 10.3 17.2 8.6 63.8 Minnesota, n = 54 Mean (SE) 8.8 (0.1) 10.2 26.5 12.2 51.0 Stanford, n = 61 Mean (SE) 9.1 (0.1) 11.1 25.9 0.0 63.0

12.1 36.4 51.5 23.8 (1.0)

35.6 33.9 30.5 23.7 (0.7)

26.0 44.0 30.0 20.4 (0.8)

31.5 42.6 25.9 21.3 (0.8)

*** P < 0.001 for differences among field center.

K.W. Cullen et al. / Preventive Medicine 38 (2004) S24S33 Table 3 Mean (SD) daily intakes at baseline by field center from 2 days of 24-h recalls All girls, n = 210 Total energy** Energy from fat* Sucrose (g)** Fructose (g)** Fiber (g) Folate (Ag) Vitamin C (mg) Beta carotene equivalents (Ag) Lycopene (Ag) Fruit (serving)** 100% fruit juice (serving)** Vegetables (serving)* Legumes (serving) Total FJV (serving) Water (serving) Sweetened beverages (serving)*** 1,590 34.6 42.8 23.5 9.7 266 75.9 1,318 (590) (5.9) (31.1) (11.9) (4.5) (127.1) (56.0) (1,885) Baylor, n = 35 1,914 33.3 62.6 29 11 318.8 93 1,732 (831) (5.7) (55.7) (14.1) (4.7) (185.2) (76.0) (2,025) Memphis, n = 60 1,498 36.3 34.8 19.4 9.1 234.5 68 1,292 (472) (5.1) (19.8) (8.7) (4.2) (117.7) (47.2) (1,857) Minnesota, n = 54 1,468 32.9 41 23.9 8.7 258.7 81.9 1,100 (448) (6.1) (16.4) (9.9) (3.5) (95) (50.2) (1,710) Stanford, n = 61 1,603 35.1 41 23.9 10.3 272.9 68.5 1,300 (583) (6.1) (26.3) (13.5) (5.2) (112.1) (54.0) (1,985) Post hoc comparison Baylor > all Memphis > Minnesota Baylor > all Baylor, Minnesota, Stanford > Memphis

S29

3,883 (4,627) 0.4 (0.7) 0.7 (0.7) 0.9 0.1 1.9 0.8 1.2 (0.8) (0.1) (1.4) (0.9) (0.9)

5,790 (5,316) 0.3 (0.5) 0.6 (0.6) 1.2 0.0 2.2 0.9 1.9 (1.0) (0.1) (1.5) (1.4) (1.1)

3,293 (5,737) 0.3 (0.4) 0.5 (0.4) 0.8 0.1 1.6 0.9 1.0 (0.8) (0.1) (1.1) (0.9) (0.8)

3,290 (3,288) 0.5 (0.7) 0.9 (0.8) 0.7 0.0 2.1 0.8 0.9 (0.7) (0.1) (1.4) (0.8) (0.7)

3,893 (3,725) 0.6 (0.9) 0.6 (0.7) 0.8 0.1 1.9 0.7 1.1 (0.8) (0.2) (1.7) (0.7) (0.8)

Minnesota and Stanford > Baylor; Stanford > Memphis Minnesota > Baylor, Memphis, Stanford Baylor > all

Baylor > all

* P < 0.05 for differences among centers by ANOVA. ** P < 0.01 for differences among centers by ANOVA. *** P < 0.001 for differences among centers by ANOVA.

the pattern of validity correlations are described in the text.

Results Demographic characteristics of the GEMS participants by field center are shown in Table 2. Significant differences for

age, body mass, and parental education were found between centers. The girls randomized at Baylor were 8 years old; the other three field centers enrolled 8, 9, and 10 year old girls. The majority of parents at Baylor reported a college degree or higher. Girls from Baylor and Memphis had significantly higher body mass index than girls from Minnesota. All girls completed the baseline assessment. Because differences in BMI by field center were identified, analyses

Table 4 Intraclass correlation coefficients across 2 days of dietary intake at baseline and daily mean (SD) consumption for Days 1 and 2 recalls Intraclass correlations All girls, n = 210 Kilocaloriesa % Energy from fat Sucrose (g)a Fructose (g) Fiber (g) Folate (Ag) Vitamin C (mg) Beta carotene equivalents Lycopene (Ag)a Fruit (serving) 100% fruit juice (serving)a Vegetables (serving) Legumes (serving) Total FJV (serving) Water (serving) Sweetened beverages (serving) ICCs different between field centers. * P < 0.05. *** P < 0.001.
a

Mean intakes Memphis, n = 60 0.19 0.05 0.42 0.01 0.42 0.55 0.20 0.10 0.69 0.15 0.08 0.22 0.08 0.0005 0.38 0.10 Minnesota, n = 54 0.28 0.17 0.02 0.11 0.28 0.31 0.20 0.16 0.01 0.36 0.43 0.12 0.10 0.32 0.41 0.03 Stanford, n = 61 0.46 0.11 0.05 0.21 0.48 0.34 0.13 0.05 0.04 0.08 0.20 0.04 0.02 0.24 0.21 0.12 Day 1, n = 210 1,629 34.7 44.9 24.1 9.8 258.0 76.7 1,114 4,225 0.44 0.63 0.75 0.08 1.88 0.74 1.23 (737) (7.7) (39.3) (15.6) (5.4) (34.3) (66.9) (2,121) (6,005) (0.88) (0.86) (0.89) (0.24) (1.71) (1.02) (1.25) Day 2, n = 210 1,551 34.5 40.8 22.8 9.5 274.0 75.1 1,523 3,540 0.39 0.67 0.96 0.04 2.02 0.86 1.1 (647) (7.8) (33.3) (14.5) (5.3) (169.4) (81.4) (3,095) (5,806) (0.88) (0.83) (1.31) (0.14) (1.92) (1.21) (1.1)

Baylor, n = 35 0.66 0.23 0.82 0.48 0.37 0.43 0.13 0.12 0.06 0.11 0.09 0.04 0.11 0.26 0.45 0.20

0.43* 0.13 0.44*** 0.21 0.42 0.38 0.13 0.001 0.22*** 0.17 0.16* 0.06 0.05 0.20 0.38 0.04

S30

K.W. Cullen et al. / Preventive Medicine 38 (2004) S24S33

Table 5 Test retest reliability of nutrients and food serving variables for comparison girls at baseline and 12 weeks follow-up Baseline mean, n = 92 Kilocalories % Energy from fat Sucrose (g) Fructose (g) Fiber (g) Folate (Ag) Vitamin C (mg) Beta carotene equivalents (Ag) Lycopene (Ag) Fruit (serving) 100% fruit juice (serving) Vegetables (serving) Legumes (serving) Total FJV (serving) Water (serving) Sweetened beverages (serving) * P < 0.05. ** P < 0.01. *** P < 0.001. 1,624 34.0 45.1 22.4 10.1 281.7 77.5 1,473 (620) (5.3) (37.6) (11.4) (4.8) (147.1) (52.8) (2,168) 12 weeks follow-up mean, n = 92 1,573 34.3 41.8 24.0 9.2 260.7 69.2 1,764 (531) (6.4) (20.8) (12.6) (3.9) (104) (46.2) (3,186) Test retest correlations, n = 92 0.24* 0.22* 0.30** 0.17 0.36*** 0.49*** 0.30** 0.19 0.09 0.19 0.30** 0.23* 0.13 0.41*** 0.29** 0.28**

4,197 (5,610) 0.5 (0.8) 0.6 (0.7) 0.9 0.1 2.0 0.8 1.0 (0.9) (0.1) (1.7) (1.0) (0.8)

3,545 (3,268) 0.3 (0.5) 0.6 (0.7) 0.9 0.1 1.8 0.7 1.4 (0.9) (0.3) (1.2) (0.8) (1.0)

were adjusted for BMI. Adjusting for BMI did not change results (data not presented). Some differences in dietary intakes across field centers were identified (Table 3). Baylor girls reported consuming more energy ( P < 0.01), sucrose ( P < 0.01), vegetables ( P < 0.05), and sweetened beverages ( P < 0.001) than girls from the other three field centers. Memphis girls consumed greater percent energy from fat ( P < 0.05) than girls in Minnesota and the least fructose ( P < 0.01) of all field centers. Minnesota and Stanford girls consumed more fruits ( P < 0.01) than Baylor girls, and Stanford girls consumed more fruits than Memphis girls. The ICCs across the 2 days of baseline dietary intake for the whole group were moderate to poor (Table 4). There

were significant differences in ICCs between field centers for energy ( P < 0.05), sucrose ( P < 0.001), lycopene ( P < 0.001), and 100% juice ( P < 0.05). Bivariate plots comparing Days 1 2 recalls by field center for energy, fructose, lycopene, fruits, 100% fruit juice, total FJV, and sweetened beverages were examined because of the large differences in ICCs across the field centers (not shown). In general, the differences in ICC reflected outliers in various field centers for various food groups that artificially increased the ICC in a small sample (e.g., energy for Baylor girls). The only plot that showed evidence of potential systematic bias was for sucrose reported by the Stanford field center. More girls reported lower consumption for the second recall compared with the first recall. Consistent with this finding was the lower intake of sweetened beverages reported on Day 2 compared to Day 1. Most test retest reliability correlations between baseline and 12 weeks follow-up for the control girls for all nutrients and food groups were significant and ranged from a low for lycopene (r = 0.09) to moderate for folate (r = 0.49) (Table 5). Total energy intake was significantly correlated with all food and nutrient variables except legumes and water (Table 6). There was a negative relationship between percent energy from fat and juice and total FJV consumption (both P < 0.001). Fruit consumption was significantly ( P < 0.001) and positively (r > 0.25) correlated with fructose, fiber, and vitamin C. Juice consumption was significantly and positively correlated with all nutrients with the exception of a negative correlation with percent energy from fat (Table 6). Vegetable consumption was significantly related to the consumption of all nutrients, except sucrose and percent energy from fat, all in the positive direction. Legume consumption was significantly and positively related to fiber ( P < 0.01) and folate ( P < 0.05) consumption. Total FJV consumption was positively correlated with all nutrients, with the highest correlations found for fiber (r = 0.60), beta carotene equivalents (r = 0.62), and vitamin C (r = 0.64). Water consumption was not related to any

Table 6 Spearman correlations among food group servings and nutrients for the mean of two 24-h dietary recalls at baseline (n = 210) Fruit Energy % Energy from fat Sucrose Fructose Fiber Folate Vitamin C Beta carotene equivalents Lycopene * P < 0.05. ** P < 0.01. *** P < 0.001. 0.15* 0.06 0.05 0.30 *** 0.34 *** 0.10 0.25 *** 0.12 0.01 Juice 0.20** 0.32*** 0.14* 0.46*** 0.29*** 0.32*** 0.55*** 0.36*** 0.17* Vegetables 0.26*** 0.11 0.13 0.25*** 0.47*** 0.38*** 0.39*** 0.62*** 0.20** Legumes 0.02 0.05 0.03 0.03 0.20** 0.17* 0.03 0.11 0.14 Total FJV 0.32*** 0.28*** 0.17* 0.50*** 0.60*** 0.47*** 0.64*** 0.62*** 0.19** Water 0.09 0 0.0 0.04 0.10 0.12 0.06 0.02 0.06 Sweetened beverages 0.37*** 0.01 0.42*** 0.57*** 0.12 0.03 0.15* 0.04 0.06

K.W. Cullen et al. / Preventive Medicine 38 (2004) S24S33

S31

nutrient. Sweetened beverage consumption was positively correlated with energy ( P < 0.001), sucrose ( P < 0.001), fructose ( P < 0.001), and vitamin C ( P < 0.05). Few differences across field centers in correlations between nutrients and food groups were found, and field centerspecific correlations were generally in the same direction as the pooled data correlations.

Discussion This report described the development of a coding system to define servings of FJV, sweetened beverages, and water from 24-h dietary recalls collected via NDS-R and examined these derived food servings variables for their reliability and validity. Two types of construct validity (dietary correlation and dietary displacement) for the derived food group variables were evident from the significant correlations between specific nutrients and food groups (Table 6). Total FJV and juice consumption were negatively related to the percentage of energy from fat, providing some support for the dietary displacement hypothesis that eating a greater amount of FJV may displace foods higher in fat [12]. Alternatively, juice consumption may have increased energy intake, but not fat, thereby reducing the percent energy from fat without displacing high-fat foods. This result also could reflect substitution of juice as a meal beverage instead of milk. Mean daily juice intake, however, was only 0.7 servings per day or about four fluid ounces, perhaps minimizing the likelihood of the last two alternatives to the dietary displacement hypothesis. The dietary correlation type of construct validity was found for FJV. Fruit consumption was positively related to nutrients found in fruits (fructose, fiber, and vitamin C), with the exception of folate. The latter finding might reflect alternative sources of folate, particularly from fortification of grain products. Juice was positively correlated to all the nutrients in FJV, plus sucrose. Vegetable consumption was significantly correlated with nutrients commonly found in vegetables (fructose, fiber, folate, vitamin C, beta carotene equivalents, and lycopene). Dietary correlation was also moderate for sweetened beverages (Table 6). Significant positive correlations were obtained between sweetened beverages and energy, sucrose, fructose, and vitamin C. Because the major sweetener of soft drinks and fruit drinks is high fructose corn syrup and sucrose, and vitamin C is often added to fruit drinks, a component of sweetened beverages, these correlations are not surprising. The lack of a negative relationship between sweetened beverages and percent energy from fat (dietary displacement) may reflect a moderate consumption of sweetened beverages by these girls, which ranged from 0.9 to 1.9 servings (7.2 15 fluid ounces) per day (Table 3). Water consumption was very low and the data did not indicate dietary displacement of other foods and beverages.

Low variability in water consumption may account for why it was not related to consumption of any nutrient. Little is known about water consumption by children in the United States. Water consumption is an important area of research because potential savings in energy may be an important component of obesity prevention and weight management programs [21]. The poor to moderate ICCs revealed substantial variability in intake over 2 days, which may be particularly evident for nutrients and food groups in which consumption is very low (Table 4). Many of the recalls were conducted on both a weekday and a weekend, that is, Saturdays and Sundays. Because weekend consumption for some foods may differ systematically from consumption on week days [13], this may influence the ICC. For example, sweetened beverages may only be allowed on weekends in some homes. The reliability ICCs of sweetened beverages, legumes, and vegetables were poor. The low reliability of vegetable servings (r = 0.06) is consistent with the low ICC for beta carotene equivalents (0.001) because vegetables are a major source of this nutrient. Higher ICCs would be expected with more days of assessment. To achieve an ICC of 0.4, approximately 10 days of collection would be needed for vegetable intake. An ICC of 0.32 was obtained for vegetables from 5 days of food records completed by 137 African American, Hispanic, and White 9- to 12-year-old girls [14], whereas an ICC of 0.53 for vegetables was obtained for two 24h dietary recalls from 10- to 14-year-old African American boy scouts [15]. The ICCs for fruit (0.17) and juice (0.16) reported here were also lower than those obtained in previous work with 9- to 12-year-old girls (0.65 for fruit and 0.51 for juice) [14] and 11- to 14-year-old boys (0.31 for fruit and 0.33 for juice) [15]. These results suggest that this younger group of African American girls (mean age about 9 years) may not provide reliable self-report of dietary intake and that more than 2 days of dietary assessment may be needed to provide more reliable estimates of usual consumption from younger girls. Alternatively, GEMS dietary intake was collected in 2002, 4 5 years after the previous studies [14,15]. Family dietary patterns experienced by the GEMS participants may be more variable from day to day, especially from weekday to weekend day, than the dietary patterns from 5 years ago. Reported dietary intake of FJV, sweetened beverages, water, energy, percentage of energy from fat, and selected nutrients varied among 8- to 10-year-old girls across the four field centers in the GEMS program (Table 3). For three of the field centers, total energy intake was lower than the 1,812 energy reported by a large nationally representative sample of 6- to 11-year-old African American girls [16]. The percentage of energy from fat exceeded the recommended 30% but was similar to the 33.7% reported by 6 11 years old, a large nationally representative sample of African American girls [16]. Despite reporting one of the lowest overall energy intake, girls from Memphis also

S32

K.W. Cullen et al. / Preventive Medicine 38 (2004) S24S33

reported the highest consumption of fat as a percent of total energy. The recommended intake of grams of dietary fiber for children is 5 plus their age [17]. Girls from all field centers reported about half the recommended value (9.7 g). These fiber intakes were somewhat lower than the 12 g reported by a national sample of 6- to 11-year-old African American girls via two in person 24-h dietary recalls [16]. Girls from all field centers reported consumption of vitamin C (75.9 mg) beyond the recommended range of 25 mg for 8 year olds and 45 mg for girls ages 9 10 but less than the 111 mg reported by a national sample of 6- to 11-year-old African American girls [16]. Folate intakes (266 Ag) were lower than the 300 Ag recommended for 9- to 13-year-old girls but similar to national data for 6- to 11-year-old African American girls (233 Ag) [16]. In general, FJV consumption by GEMS girls (1.9 serving/day) was very low compared to national surveys. However, the behavior-based food coding system excluded high-fat vegetables, fruits in desserts, and juice in fruit drinks that may account for the low consumption values. Consumption of 2.13 servings of FJV was reported by 210 African American, Hispanic, and White 9- to 12-year-old students in Houston using a behavior coding system and student-completed food records [14]. Girls from Baylor consumed the greatest amount of vegetables (1.2 servings per day), which was still lower than the 2.4 servings reported by a national sample of 6- to 11-year-old African American girls [16]. However, 25% of vegetables were French fries in the national data set, and our coding scheme excluded fried potatoes [16] so our values are actually much lower than national estimates. Memphis and Baylor girls reported the lowest fruit consumption (0.3 serving/day), but the other field centers also reported low consumption (0.5 0.6 serving/day) compared with the 1.4 serving reported by a national sample of 6- to 11-year-old African American girls [16]. The national data reflect an epidemiological FJV coding system whereby all FJV counts are not directly comparable to the current analyses. However, it is clear that girls assessed in this study are consuming inadequate amounts of fruits and vegetables and interventions are needed to improve consumption of these important foods. More 100% fruit juice was consumed by Minnesota girls (about 5.3 fluid ounces) than the other three field centers. About 2.1 fluid ounces of citrus juices were consumed by 6to 11-year-old African American girls [16]. Sweetened beverage consumption varied from the highest amount (1.9 servings or 15.2 fluid ounces) for the Baylor field center to a low of 7.2 fluid ounces reported by girls in Memphis. Previously published data reported consumption of about 11.6 fluid ounces of sweetened beverages for 9- to 12-year-old African American boys and girls in Houston [18] and 11.8 fluid ounces by a national sample of African American girls [16]. The Houston data were obtained from up to 7 days of food records completed daily in classrooms and might have provided more reliable estimates of con-

sumption compared with other reported sweetened beverage consumption data. Because sweetened beverages supply only energy, consumption at any level may increase risk for weight gain. The variability in sweetened beverage consumption across field centers may reflect true differences in consumption and patterns, perhaps due to regional differences in food amounts consumed and in region-specific differences in reporting. Test retest correlations of average intake between baseline and conclusion of the intervention for girls in the comparison condition were in the poor to moderate range (Table 5). This could reflect variation due to seasonal availability since up to 5 months elapsed between baseline and follow-up recalls. Differences in the proportion of weekend versus weekday meals between baseline and follow-up might have contributed to the lower values. There are several important limitations to these results. The low reliability across the 2 days of recalls for some of the food groups and nutrients may have reduced the ability to detect real differences between the test retest measures. Obtaining more days of dietary recall would reduce the standard error and increase the likelihood of detecting significant relationships in the data. Seasonality and day of the week effects could impact the results. All data were from self-report and are thereby subject to memory and recall errors [19]. To minimize this problem, all data collectors attended a rigorous training program and were certified in using NDS-R to collect GEMS dietary recalls. While some biomarkers would help to provide an objective measure of some of the nutrients found in FJV [20], measuring biomarkers is invasive, expensive, and difficult to justify with healthy young girls volunteering for a health promotion program. Also, there is no biomarker adequate for validating consumption of fruits or vegetables.

Conclusion A behavior-based coding system for FJV, sweetened beverages, and water for NDS-R was found to have construct validity among 8- to 10-year-old African American girls. Reliability across 2 days of recall was low for some nutrients and food groups, suggesting that more than 2 days of dietary recalls would be needed to provide more reliable estimates of usual consumption. Use of the coding system with children from other ethnic groups and adults and further validation against observational or other procedures are warranted.

Acknowledgments This work is a publication of the USDA/ARS Department of Pediatrics, Childrens Nutrition Research Center, Baylor College of Medicine and Texas Childrens Hospital,

K.W. Cullen et al. / Preventive Medicine 38 (2004) S24S33

S33

Houston, TX. This study was sponsored by the National Heart, Lung, and Blood Institute (U01 HL65160, U01 HL62662, U01 HL62663, U01 HL62732, and U01 HL62668). This project has also been funded in part by federal funds from the USDA/ARS under Cooperative Agreement No.58-6250-6001. The contents of this publication do not necessarily reflect the views or policies of the USDA, nor does mention of trade names, commercial products, or organizations imply endorsement by the U.S. Government.

[9]

[10]

[11]

[12]

References
[13] [1] Hubert HB, Feinleib M, McNamara PM, Castelli WP. Obesity as an independent risk factor for cardiovascular disease: a 26-year followup of participants in the Framingham Heart Study. Circulation 1983;67:968 77. [2] Pi-Sunyer FX. Medical hazards of obesity. Ann Intern Med 1993; 119:655 60. [3] Potter JD. Food, nutrition and the prevention of cancer: a global perspective. Washington, DC: World Cancer Research Fund and American Institute for Cancer Research; 1997. [4] Lloyd T, Chinchilli VM, Rollings N, Kieselhorst K, Tregea DF, Henderson NA, et al. Fruit consumption, fitness, and cardiovascular health in female adolescents: the Penn State Young Womens Health Study. Am J Clin Nutr 1998;67:624 30. [5] Ludwig DS, Peterson KE, Gortmaker SL. Relation between consumption of sugar-sweetened drinks and childhood obesity: a prospective, observational analysis. Lancet 2001;357:505 8. [6] Raynor HA, Epstein LH, Gordy CC. Effects of increased fruits and vegetables and decreasing high-fat and or high sugar during obesity treatment. Annals Behav Med 1999; 21:S019 [Suppl.]. [7] Cleveland LE, Cook DA, Krebs-Smith SM, Friday J. Method for assessing food intakes in terms of servings based on food guidance. Am J Clin Nutr 1997;65:1254S 63S. [8] Cullen K, Baranowski T, Baranowski J, Hebert D, deMoor C. Behavioral or epidemiological coding of fruit and vegetable consumption

[14]

[15]

[16] [17] [18]

[19]

[20]

[21]

from 24-h dietary recalls: research question guides choice. J Am Diet Assoc 1999;99:849 51. Eldridge AL, Smith-Warner SA, Lytle LA, Murray DM. Comparison of 3 methods for counting fruits and vegetables for fourth-grade students in the Minnesota 5 A Day Power Plus Program. J Am Diet Assoc 1998;98:777 82 [quiz 783-4]. Smith SA, Campbell DR, Elmer PJ, Martini MC, Slavin JL, Potter JD. The University of Minnesota Cancer Prevention Research Unit vegetable and fruit classification scheme (United States). Cancer Causes Control 1995;6:292 302. Rochon J, Klesges RC, Story M, Robinson TN, Baranowski T, Obarzanek E, et al. Common design elements of the Girls health Enrichment Multi-site Studies (GEMS). Ethn Dis 2003;13:S6 S14. Havas S, Heimendinger J, Reynolds K, Baranowski T, Nicklas T, Bishop D, et al. 5-a-day for better health: a new research initiative. J Am Diet Assoc 1994;94:32 6. Baranowski T, Smith M, Hearn MD, Lin LS, Baranowski J, Doyle C, et al. Patterns in childrens fruit and vegetable consumption by meal and day of the week. J Am Coll Nutr 1997;16:216 23. Cullen KW, Baranowski T, Rittenberry L, Cosart C, Hebert D, de Moor C. Child-reported family and peer influences on fruit, juice, and vegetable consumption. Reliability and Validity of Measures. Health Educ Res 2001;1:187 200. Baranowski T, Baranowski J, Cullen KW, de Moor C, Rittenberry L, Hebert D, et al. 5 A day achievement badge: results of a pilot study among African american boy scouts. Cancer Epidemiol Biomark Prev [submitted for publication]. U.S. Department of Agriculture ACNRG. Food and Nutrient Intakes by Race, 1994-94 (Table set 11). 1994 1996. Williams CL, Bollella M, Wynder EL. A new recommendation for dietary fiber in childhood. Pediatrics 1995;96:985 8. Cullen KW, Ash DM, Warneke C, de Moor C. Intake of soft drinks, fruit-flavored beverages, and fruits and vegetables by children in grades 4 through 6. Am J Public Health 2002;92:1475 8. Domel S, Thompson WO, Baranowski T, Smith AF. How children remember what they have eaten. J Am Diet Assoc 1994;94: 1267 72. Resnicow K, Odom E, Wang T, Dudley WN, Mitchell D, Vaughn R, et al. Validation of three food frequency questionnaires and twentyfour hour recalls with serum carotenoids in a sample of African American adults. Am J Epidemiol 2000;152:1072 80. Rolls B, Barnett RA. Volumetrics. New York: Harper Collins; 2000.