| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Cancer Prevention Research Program, Fred Hutchinson Cancer Research Center, Seattle, Washington 98109-1024 [A. R. K., N. C. V., R. E. P., M. L. N., A. L. S., D. M.], and Department of Epidemiology [A. R. K., R. E. P.] and Nutritional Sciences Program [A. R. K., R. E. P., M. L. N.], University of Washington, Seattle, Washington 98195
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The majority of studies measuring fruit and vegetables have used some type of food frequency approach. One approach, commonly called the 5 A Day method, uses two summary questions to capture consumption of most fruits and vegetables ["How often did you eat a serving of fruit (not including juices)?" and "How often did you eat a serving of vegetables (not including salad and potatoes)"], to which usual consumption of juice, salad, and potatoes are added (8) . These questions are usually asked alone, as a very short FFQ,3 or can be extracted from items included in a longer, more comprehensive FFQ. This method is well suited for population-level surveillance and intervention evaluation because it is simple to administer and analyze. Another food frequency approach, which we call the summation method, is to add the frequencies of fruit and vegetable consumption from a list of relevant items in a comprehensive FFQ. This method is often used in epidemiological studies of diet and disease, for which long dietary questionnaires are more feasible. Lastly, instead of a food frequency approach, one can calculate servings of fruit and vegetables from 24-h dietary recalls or food records (9, 10, 11, 12, 13) . This requires either complex computer algorithms or hand coding of fruits and vegetables to extract the desired information. The computer expertise, costs, and high participant burden make this approach suitable only for small clinical studies or validation studies.
This report examines the bias and precision of the two commonly used FFQ-based approaches to measuring fruit and vegetable intakes: (a) the 5 A Day method; and (b) the summation method. We examine total fruit and vegetable consumption, as well as the consumption of components (e.g., fruit juice and salads) that comprise the total fruit and vegetable measure. Bias and precision are presented from three large studies that collected a FFQ-based measure and at least one independent criterion measure such as food records, dietary recalls, or serum carotenoids.
|
Materials and Methods |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
12 years of
education, 55% had completed 16 or more years of education), 57% were
female, 89% were white, and their mean age was 42 years (range, 20–67
years). The second data source was the Women’s Health Trial Feasibility Study in Minority Populations, a randomized trial to test the effectiveness of a low-fat dietary intervention program in minority and low socioeconomic status women. Details of the study design and participant characteristics (15) and dietary assessments (16) have been published. Dietary data are from baseline assessments, when participants completed both a 100-item FFQ that asked about diet over the past 3 months and a 4-day food record. Data from this analysis are from 1031 women with acceptable FFQs and analyzed food records (see Ref. 16 for details). Participants were of diverse race (28% black, 16% Hispanic, and 56% white) and mixed educational background (12% completed less than 12 years of education, 20% completed 12 years of education, and 37% completed 16 or more years of education), with a mean age of 60 years (range, 50–79 years).
The third data source was the Olestra Post-Marketing Surveillance Study, a population-based observational study of dietary patterns and serum concentrations of fat-soluble vitamins and carotenoids. Details on the design and participant characteristics of this study have been published (17) . Participants were recruited for a telephone interview using random-digit dialing (efficacy rate, 61.9%) and then invited into a clinic for further assessments (participation rate, 56.9%). Data for this analysis are from 342 adults from the sentinel site in Marion County (Indianapolis, IN). Participants completed four 24-h dietary recalls (three dietary recalls were administered by telephone, and one dietary recall was conducted in person) over a 1-year period. The last recall was administered at a clinic visit during which participants gave a fasting blood sample and completed a comprehensive 122-item FFQ that asked about diet over the past year. Participants were of diverse race (19% black, 77% white, and 4% other) and educational background (40% completed 12 or fewer years of education, 30% completed 13–15 years of education, and 30% completed 16 or more years of education), and 62% were female, with a mean age of 41 years (range, 18–84 years).
Measures
FFQs.
Table 1
gives details on how each FFQ was used to calculate servings of fruit
and vegetables. All FFQs used fixed categorized frequency responses. On
the 100- and 122-item FFQs, all food items included information on
portion size, categorized as small, medium, or large compared with a
stated medium portion size. Portion sizes matched those used in the 5 A
Day program (e.g., 1/2 cup cooked vegetable) and were
used when calculating total servings.
|
Summation Method.
Servings of fruit and vegetables were calculated as the simple sum of
servings of all relevant food items on the FFQ. There were 30 fruit and
vegetable items on the 100-item FFQ, and there were 35 fruit and
vegetable items on the 122-item FFQ. The summation method does not use
questions on either "servings of vegetables, not including salad or
potatoes" or "servings of fruit, not including juices."
Four-day Food Records/24-h Dietary Recalls.
Recalls were collected using the University of Minnesota Nutrition
Coordinating Center’s Nutrition Data System (versions 2.9 and 2.91),
and food records were entered into the Nutrition Data System for
analysis (18)
. One characteristic of this system is that
multiple-component foods (e.g., casseroles) are broken into
their component parts (e.g., meat, potatoes, carrots, and
onions) for nutrient analysis. We could therefore develop a set of
computer algorithms to calculate intake of all fruit and vegetables,
both those eaten alone and those eaten as part of multiple-component
foods (19)
. For this analysis, fruits were divided into
juice and all other forms of fruit. Vegetables were divided into
salads/raw, fried potatoes, and all other vegetables, excluding
condiments, legumes, nuts, pickled vegetables, and soybean products.
Intake was calculated in servings using definitions that match those
commonly consumed (e.g., a single piece of fruit, 1/4
cup of dried fruit, and 1/2 cup of cooked vegetable).
Serum Carotenoids.
Bloods were collected after a fast of at least 6 h and protected
from light throughout processing. Serum was stored at -70°C until
analysis at Quintiles Laboratories (Atlanta, GA). A high-performance
liquid chromatography method was used to measure concentrations of
-carotene, ß-carotene, lycopene, lutein, zeaxanthin, and
ß-cryptoxanthin, and an enzymatic method was used to measure serum
cholesterol. Details of laboratory methods and assay precision are
given elsewhere (20)
. Both total carotenoids and total
carotenoids minus lycopene were used as biological markers of fruit and
vegetable intake. Lycopene was removed from the total carotenoid
measure because serum lycopene is poorly correlated with other serum
carotenoids and fruit and vegetable intake (12
, 21)
.
Statistical Methods
We defined bias as the difference between mean FFQ-based measures
of fruit and vegetable intake and a self-reported criterion measure,
either multiple 24-h recalls or food records. Tests of significance
were based on paired t tests. We defined precision as the
Pearson correlation between FFQ-based measures and a criterion measure.
We examined bias and precision for: (a) total fruit and
vegetables; (b) total fruit and vegetables minus fried
vegetables (to match the 5 A Day guidelines); and (c)
subgroups of fruit and vegetables that could be defined similarly
between FFQ- and recall/record-based methods.
To normalize distributions, all values for fruits and vegetables were transformed using a (1+ln) transformation before analysis, and results are given back-transformed for ease of interpretation. Serum carotenoids were adjusted for total serum cholesterol following the residual method described by Willett (22) and log-transformed before analysis.
|
Results |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
gives comparisons between the 5 A Day method and the mean of three 24-h
dietary recalls. Compared with the 5 A Day method, recalls were 23%
higher for total fruits and vegetables, 15% higher for fruit, 72%
higher for salads, 61% higher for fried vegetables, and 5% higher for
cooked vegetables. Recalls were lower than the 5 A Day method only for
fruit juice. Correlations between the two measures for total fruit and
vegetables were modestly strong, and they were consistently higher for
fruit and fruit juices than for vegetables. There was some suggestion
that correlations were higher for salads than for cooked or fried
vegetables.
|
gives comparisons among the 5 A Day method, the summation method using
30 fruit and vegetable items, and food records. For total fruit and
vegetables and for all subgroups of fruits and vegetables where
comparisons are possible, intakes were higher for the summation method
than for the 5 A Day method. The summation method was 31% higher for
total fruit and vegetables, 70% higher for fruit, and 36% higher for
cooked vegetables. Intakes calculated from food records were
intermediate between the two FFQ-based methods. Correlations with food
records were slightly but not significantly higher with the summation
method, as compared with the 5 A Day method. For both FFQ-based
methods, correlations with food records were higher for fruit and fruit
juice than for vegetables. These results did not differ substantially
when analyses were stratified by race/ethnicity (black, white, or
Hispanic) or by education (12 years, 13–15 years, 16+ years; data
not shown).
|
gives correlations among the 5 A Day method, the summation method using
the sum of 35 fruit and vegetable items, 24-h dietary recalls, and
serum carotenoid concentrations. This table focuses on the relative
precision of all self-reported measures, using serum carotenoids as an
objective, biological measure. For this reason, and because the
patterns closely match those previously given, we do not give mean
intakes and correlations between recalls and the FFQ-based methods. The
correlation between the 24-h recall measure of total fruit and
vegetables and serum carotenoids was slightly higher than correlations
for either FFQ-based measure. The higher precision for the 24-h recalls
is due to stronger correlations for vegetables because correlations for
fruit were similar across measures. When the criterion measure was
total carotenoids minus lycopene, correlations increased for all
measures. Using this criterion measure, correlations are only slightly
higher for recalls than for the FFQ-based measures. Similar to the
findings shown in Tables 2
3
, there are virtually no differences in
precision between the 5 A Day method and the summation method.
|
|
Discussion |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
There are very few published reports on the precision and bias of the 5 A Day method using a self-reported criterion measure, and their results are somewhat inconsistent. Field et al. (23) examined the validity of FFQ-based measures in 102 high school students, using the mean of three 24-h recalls as the criterion measure; Thompson et al. (8) examined the validity of a 10-item FFQ using two 24-h dietary recalls as the criterion measure in 443 adults, and Serdula et al. (24) examined the 6-item FFQ used in the Behavioral Risk Factor Surveillance System using a variety of criterion measures in five sites (sample sizes ranged from 51–553 per site). There is consistency in the finding that the 5 A Day method underestimates total fruit and vegetable intakes, and some (23) but not all (24) studies show this to be due to underestimates of vegetable consumption. Several studies have compared the 5 A Day and summation methods (8 , 23, 24, 25, 26) , and almost all show lower intakes from the 5 A Day method. Two studies have compared the precision of the 5 A Day and summation methods, based on an independent criterion measure (8 , 23) , and both found no substantial differences between the two methods. Finally, there are many studies that have reported on the precision of FFQ-based measures of fruit and vegetables separately (23 , 24 , 27, 28, 29, 30, 31, 32) . Most of these studies used the summation method to calculate fruit and vegetable intakes, and most found that correlations with criterion measures were higher for fruit than vegetables.
There is no simple answer regarding whether it is preferable to use the 5 A Day method or the summation method to measure usual fruit and vegetable intakes. This report and other studies (8 , 23 , 24 , 32 , 33) clearly show that respondents overestimate fruit and vegetable intakes when they are measured as the sum of long lists of fruit and vegetable items. However, results here and in most other studies show that the 5 A Day method underestimates intake. Because there is no difference in precision between the 5 A Day and summation method, it should make little difference in epidemiological studies of associations between fruit and vegetable intakes and disease risk. However, in research where the objective is to measure absolute rather than relative intake, for example, for comparison with dietary recommendations, these two methods will lead to profoundly different conclusions. Other considerations are that the 5 A Day method is simpler to administer and has a lower respondent burden than the summation method. However, for some purposes, it may be preferable to have detailed information on which fruits and vegetables are consumed, for example, to better design or evaluate the effects of a dietary intervention program.
A related issue concerns using FFQs to calculate intakes of micronutrients concentrated in fruit and vegetables. Comprehensive FFQs typically ask about many different fruits and vegetables, grouped by similarities in nutrient content, to estimate intake of micronutrients (e.g., vitamin C, carotenoids, and folate) and other compounds (e.g., isoflavones). Most algorithms to calculate nutrients from FFQs first adjust the reported intakes of individual fruits and vegetables such that they add up to the answers given to the related summary question (e.g., the sum of the frequencies for all vegetable items would equal the answer to the summary question "How often did you eat a serving of vegetables, not counting salads or potatoes"). This almost always results in a downward adjustment of intake frequencies. Results here suggest that this adjustment procedure will result in estimates of nutrients from fruit and vegetables that are biased to be lower than those based on food records or dietary recalls, whereas calculations made without this adjustment will be higher. The truth appears to lie somewhere between the two approaches.
Differences in precision between components of the total fruit and vegetable measure may be due to both cognitive aspects of dietary recall and to the questions themselves. All self-reported measures were more valid for fruit and juices than for vegetables, and within vegetables, there was better agreement between FFQ-based and recall-based methods for salads than for cooked vegetables. One explanation of this could be that juices and salads tend to be eaten regularly, for example a glass of juice with breakfast and a salad with dinner. Therefore, they may be recalled with greater accuracy than the consumption of foods eaten irregularly. Another explanation is that the word "vegetable" has different interpretations, and vegetables in mixed foods, such as casseroles or sandwiches, may be forgotten and thus underreported (23) . One approach to improving the 5 A Day method would be to modify the questions addressing vegetables included in mixed foods, for example, "carrots, onions, or other vegetables including in mixed dishes." Research on ways to better measure vegetables is clearly needed.
These results have implications for all studies of fruit and vegetable consumption using FFQ-based approaches. First, because the precision of measuring usual vegetable intake is poor, it substantially limits study conclusions. Epidemiological studies of diet and disease risk will fail to detect significant associations or report associations that are weaker than true associations. Studies to evaluate intervention programs to increase vegetable intake will underestimate the true intervention effect or may find a significant effect for fruit but not for vegetables. Behavioral research to understand attitudes and other determinants of fruit and vegetable intakes may report that some factors more strongly affect fruit compared to vegetable intakes, when these differences may be due solely to measurement error (33) . Second, public health policy may be misguided by weak or negative study results. The generally accepted protective associations of high vegetable intakes with many cancers are likely to be much stronger than previously believed. Third, population-based surveillance studies of fruit and vegetable intakes are probably underestimating vegetable intake, and consumption of fruit and vegetables could be substantially higher than those reported based on programs such as the Behavioral Risk Factor Surveillance System (24 , 34) .
There are several strengths and limitations to the analyses reported here. Most importantly, we had no truly valid criterion measure of fruit and vegetable intakes. Although we used criterion measures believed superior to FFQ-based measures, it is best to consider our results not as a study of true validity but as a study of convergent validity. Both the FFQ and criterion measures of fruit and vegetable intakes differed across the three studies. This limits our ability to make statements about the actual amount of bias or to compare precision across studies. However, the consistency across studies, which are also based on samples that differed markedly in race/ethnicity and socioeconomic status, do suggest that our conclusions are generalizable. The time periods referenced by dietary assessment instruments used in each study were similar but did not match exactly. The results given here may therefore underestimate the true precision of FFQ-based measures. An important strength of the results of this report is that one study included serum carotenoids as an objective, biological measure of fruit and vegetable intake. Whereas serum carotenoids are only moderately correlated with fruit and vegetable intakes and are affected by many other dietary and lifestyle factors (20 , 21) , serum carotenoids are still one of the most broadly accepted and useful biomarkers of diet (35 , 36) .
We conclude that FFQ-based measures of fruit and vegetable intakes have significant limitations, due primarily to their imprecise assessments of vegetable intake. Research on ways to improve these measures is important because they are the only feasible methods for large studies or surveys. Whenever possible, studies should collect additional measures, such dietary recalls or biomarkers of fruit and vegetable intakes, to validate results based on FFQ-based measures alone.
|
Footnotes |
|---|
1 Supported by NIH Grants N01-CN25425,
N01-CN25426, N01-CN25427, N01-CN25343, R01-CA59731, and
P30-CA15704 and by the Procter and Gamble Co. (Cincinnati, OH). 
2 To whom requests for reprints should be
addressed, at Cancer Prevention Research Program, Fred Hutchinson
Cancer Research Center, 1100 Fairview Avenue N., MP-702, Seattle, WA
98109-1024. Phone: (206) 667-4686; Fax: (206) 667-5977; E-mail: akristal{at}fhcrc.org
3 The abbreviation used is: FFQ, food frequency
questionnaire.
Received 2/23/00; revised 6/ 6/00; accepted 6/26/00.
|
References |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
|
|
 |
|
  M. R. Saracino, J. Bigler, Y. Schwarz, J.-L. Chang, S. Li, L. Li, E. White, J. D. Potter, and J. W. Lampe Citrus Fruit Intake Is Associated with Lower Serum Bilirubin Concentration among Women with the UGT1A1*28 Polymorphism J. Nutr., March 1, 2009; 139(3): 555 - 560. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. S Maruti, C. M Ulrich, and E. White Folate and one-carbon metabolism nutrients from supplements and diet in relation to breast cancer risk Am. J. Clinical Nutrition, February 1, 2009; 89(2): 624 - 633. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 E. De Vet, J. De Nooijer, N. K. De Vries, and J. Brug Do the Transtheoretical Processes of Change Predict Transitions in Stages of Change for Fruit Intake? Health Educ Behav, October 1, 2008; 35(5): 603 - 618. [Abstract] [PDF]
|
|
|
|
 |
|
 S. S. Maruti, J. W. Lampe, J. D. Potter, A. Ready, and E. White A Prospective Study of Bowel Motility and Related Factors on Breast Cancer Risk Cancer Epidemiol. Biomarkers Prev., July 1, 2008; 17(7): 1746 - 1750. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C. Atkinson, K. M Newton, E. J A. Bowles, M. Yong, and J. W Lampe Demographic, anthropometric, and lifestyle factors and dietary intakes in relation to daidzein-metabolizing phenotypes among premenopausal women in the United States Am. J. Clinical Nutrition, March 1, 2008; 87(3): 679 - 687. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. R. Hebert, T. G. Hurley, K. E. Peterson, K. Resnicow, F. E. Thompson, A. L. Yaroch, M. Ehlers, D. Midthune, G. C. Williams, G. W. Greene, et al. Social Desirability Trait Influences on Self-Reported Dietary Measures among Diverse Participants in a Multicenter Multiple Risk Factor Trial J. Nutr., January 1, 2008; 138(1): 226S - 234S. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 E. de Vet, J. de Nooijer, N. K. de Vries, and J. Brug Comparing stage of change and behavioral intention to understand fruit intake Health Educ. Res., August 1, 2007; 22(4): 599 - 608. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. M. Troen, B. Mitchell, B. Sorensen, M. H. Wener, A. Johnston, B. Wood, J. Selhub, A. McTiernan, Y. Yasui, E. Oral, et al. Unmetabolized Folic Acid in Plasma Is Associated with Reduced Natural Killer Cell Cytotoxicity among Postmenopausal Women J. Nutr., January 1, 2006; 136(1): 189 - 194. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
B. Cartmel, D. Bowen, D. Ross, E. Johnson, and S. T. Mayne A Randomized Trial of an Intervention to Increase Fruit and Vegetable Intake in Curatively Treated Patients with Early-Stage Head and Neck Cancer Cancer Epidemiol. Biomarkers Prev., December 1, 2005; 14(12): 2848 - 2854. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. L. Neuhouser, R. E. Patterson, M. D. Thornquist, G. S. Omenn, I. B. King, and G. E. Goodman Fruits and Vegetables Are Associated with Lower Lung Cancer Risk Only in the Placebo Arm of the {beta}-Carotene and Retinol Efficacy Trial (CARET) Cancer Epidemiol. Biomarkers Prev., April 1, 2003; 12(4): 350 - 358. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. P. Bogers, P. C. Dagnelie, K. R. Westerterp, A. D. M. Kester, J. D. van Klaveren, A. Bast, and P. A. van den Brandt Using a Correction Factor to Correct for Overreporting in a Food-Frequency Questionnaire Does Not Improve Biomarker-Assessed Validity of Estimates for Fruit and Vegetable Consumption J. Nutr., April 1, 2003; 133(4): 1213 - 1219. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
N. K. Horner, A. R. Kristal, J. Prunty, H. E. Skor, J. D. Potter, and J. W. Lampe Dietary Determinants of Plasma Enterolactone Cancer Epidemiol. Biomarkers Prev., January 1, 2002; 11(1): 121 - 126. [Abstract] [Full Text]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| Cancer Research | Clinical Cancer Research |
| Cancer Epidemiology Biomarkers & Prevention | Molecular Cancer Therapeutics |
| Molecular Cancer Research | Cancer Prevention Research |
| Cancer Prevention Journals Portal | Cancer Reviews Online |
| Annual Meeting Education Book | Meeting Abstracts Online |
