- Email: [email protected]
Dietary intake estimates in the National Health Interview Survey, 2000: Methodology, results, and interpretation
RESEARCH Current Research
Continuing Education Questionnaire, page 487 Meets Learning Need Codes 3000, 3010, 4000, and 9070
Dietary Intake Estimates in the National Health Interview Survey, 2000: Methodology, Results, and Interpretation FRANCES E. THOMPSON, PhD, MPH; DOUGLAS MIDTHUNE, MS; AMY F. SUBAR, PhD, MPH, RD; TIMOTHY MCNEEL; DAVID BERRIGAN, PhD, MPH; VICTOR KIPNIS, PhD
ABSTRACT Objectives To describe the implementation of the Multifactor Screener in the 2000 National Health Interview Survey (NHIS); to provide intake estimates for fruits and vegetables, fiber, and percentage of energy from fat by various demographic and behavioral characteristics; and to discuss the strengths and weaknesses of the method. Design/Subjects The 2000 NHIS was conducted in 38,632 households in a clustered sample representative of the 48 contiguous states in the United States. The Cancer Control Module was administered to adults (18 years and older) and included 17 dietary intake questions. Analyses After excluding individuals with missing data or unlikely values on the diet questions, we computed individual intake of servings of fruits and vegetables, percentage of energy from fat, and grams of fiber. We estimated median intakes and distributions of intakes using adjusted variance estimates. We present bivariate relationships between diet and demographics and
F. E. Thompson is an epidemiologist, A. F. Subar is a research nutritionist, D. Berrigan is a cancer prevention fellow, and D. Midthune and V. Kipnis are mathematical statisticians, National Cancer Institute, Bethesda, MD. T. McNeel is with Information Management Services, Inc, Silver Spring, MD. Address correspondence to: Frances E. Thompson, MPH, PhD, Risk Factor Monitoring and Methods Branch, Applied Research Program, Division of Cancer Control and Population Sciences, National Cancer Institute, EPN 4016, 6130 Executive Blvd, MSC 7344, Bethesda, MD 20892-7344. E-mail: [email protected]. GOV Copyright © 2005 by the American Dietetic Association. 0002-8223/05/10503-0005$30.00/0 doi: 10.1016/j.jada.2004.12.032
352
Journal of the AMERICAN DIETETIC ASSOCIATION
diet and behavioral characteristics in almost 30,000 adults in the United States. Results In general, intakes of these dietary factors were closer to recommendations among well-educated individuals, those engaged in other healthful behaviors, and underweight and normal weight individuals. Latinos had higher intakes of fruits and vegetables (men: 6 servings; women: 4.8 servings) and fiber (men: 23 g; women: 17 g), and generally a lower percentage of energy intake from fat (men: 33.7%; women: 32.1%) than did non-Latino whites (men: 5.4 servings; women: 4.5 servings; men: 19 g; women: 14 g; men: 33.9%; women: 32.0%) and non-Latino blacks (men: 5.4 servings; women: 4.4 servings; men: 19 g; women: 13 g; men: 34.7%; women: 33.5%). The strengths and limitations of the short dietary assessment method are discussed. Conclusions The Multifactor Screener in NHIS, when used in conjunction with external reference data, provides reasonable estimates for three dietary factors and suggests relationships between intakes and other characteristics that are consistent with other data. Thus, these NHIS estimates could provide useful national benchmarks for local surveys using the same instrument. J Am Diet Assoc. 2005;105:352-363.
T
he interviewer-administered 24-hour dietary recall continues to be generally recognized as the most accurate and complete self-reported information about an individual’s diet for a given day (1). Thus, recalls have been used in national surveys to characterize the US population’s dietary intake and to track changes over time. However, because the 24-hour recall necessitates highly-trained interviewers, a heavy respondent burden, and complex coding, it is prohibitively expensive for many research applications. In some situations, self-administered food frequency questionnaires, which often may be optically scanned for inexpensive data entry, have been used. However,
© 2005 by the American Dietetic Association
Questions asked on Multifactor Screener: NHIS 2000 How many times per day, week, or month did you USUALLY eat (or drink): . . . cold cereals? What kind of milk did you usually use? . . . milk, either to drink or on cereal? Response categories were: whole milk, 2% fat, 1% fat, 1⁄2% milk, nonfat or skim milk, or other. . . . bacon or sausage, not including low-fat, light, or turkey varieties? . . . hot dogs made of beef or pork? . . . whole-grain bread including toast, rolls, and in sandwiches? Whole-grain breads include whole wheat, rye, oatmeal, and pumpernickel. . . . 100% fruit juice such as orange, grapefruit, apple, and grape juices? Do NOT count fruit drinks such as Kool-Aid,a lemonade, cranberry juice cocktail, Hi-C,b and Tang.a . . . fruit? COUNT fresh, frozen, or canned fruit. Do NOT count juices. . . . regular-fat salad dressing or mayonnaise, including on salad and sandwiches? . . . lettuce or green leafy salad, with or without other vegetables? . . . french fries, home fries, or hash brown potatoes? . . . other white potatoes? COUNT baked potatoes, boiled potatoes, mashed potatoes, and potato salad. . . . cooked dried beans, such as refried beans, baked beans, bean soup, and pork and beans? . . . not counting what you just indicated (about lettuce salads, white potatoes, cooked dried beans), and not counting rice . . . OTHER vegetables? . . . any kind of pasta? COUNT spaghetti, noodles, macaroni and cheese, pasta salad, and any other kind of pasta. . . . peanuts, walnuts, seeds, or other nuts, not including nut butters. . . . regular fat potato chips, tortilla chips, or corn chips? Do NOT include low-fat chips. (Additional help was given for individual questions as the need arose.) Response categories were recorded for 1) Number of times (0 to 94; 95⫽95⫹) and 2) time period (day, week, month, year). Figure. Questions asked on Multifactor Screener: National Health Interview Survey 2000 (3). aKraft Foods, Northfield, IL. bCoca-Cola Co, Atlanta, GA.
complete quantitative food frequency questionnaires themselves often require up to an hour to complete, and thus are unsuitable for many studies. Various shorter tools that measure a limited number of dietary factors rather than the entire diet have been developed (1). One of these—the Multifactor Screener, developed by Thompson and colleagues (2)—assesses intakes of fruits and vegetables, fat, and fiber, and was included in the 2000 National Health Interview Survey (NHIS). The purpose of this paper is to describe the implementation of the screener in the 2000 NHIS; to provide intake estimates of intakes of fruits and vegetables, fiber, and percentage of energy from fat by various demographic and behavioral characteristics; and to discuss the strengths and weaknesses of the method. METHODS Study Sample and Design The NHIS is conducted annually by the National Center for Health Statistics to ascertain a variety of self-reported health behaviors and conditions. Periodically, a Cancer Control Module is included to obtain information pertinent to cancer researchers and not asked in the other US national health survey, the National Health and Nutrition Examination Survey (NHANES). The 2000 Cancer Control Module supplement (3) consisted of questions about diet, physical activity, tobacco use, sun exposure, cancer screening, genetic testing, family history of cancer, drug use, and Latino acculturation. Because the 2000 Cancer Control Module was constrained to take no longer
than 20 minutes, only a limited number of questions about diet were included. The goal of the diet questions was to allow characterization of the diets of subgroups of the population in terms of servings of fruits and vegetables, grams of fiber, and percentage of energy from fat for examination of interrelationships among other cancerrelated behaviors and disease risk factors. The NHIS consists of a clustered, randomized sample of households from the 48 contiguous states. The 2000 survey was administered in the home by computerassisted personal interviewing. The 2000 Cancer Control Module supplement was administered to one adult sample person selected randomly from men and women age 18 and older living in the households participating in the NHIS. In 2000, information for the NHIS was collected from 38,632 households, 88.9% of those eligible for interview. From the participating households, 32,374 adults were interviewed, 82.6% of those eligible. Thus, the overall response rate for the sample adults was 72.1% (3). Multifactor Screener The Multifactor Screener (shown in the Figure) asks frequency of use information for 16 food groupings and one question about type of milk consumed; portion-size information is not asked. Development of the Multifactor Screener reflects a novel approach to the construction and analysis of short instruments. Using nationally representative food intake data, the foods selected for inclusion in the screener were those “explaining” the most variability in dietary intakes. Scoring algorithms to translate fre-
March 2005 ● Journal of the AMERICAN DIETETIC ASSOCIATION
353
quency of use responses into estimated daily intakes using portion-size information from an external source were also developed. We evaluated this screener’s performance with several external datasets (2), principally in the Observing Protein and Energy Nutrition Study conducted in 1999-2000 (4) and the Eating at America’s Table Study conducted in 1997-1998 (5). The Multifactor Screener in the NHIS differed slightly from that in the Observing Protein and Energy Nutrition Study in that an additional category of other milk to the query about type of milk usually consumed was present in NHIS. Data Processing Individuals who did not give substantive responses to at least 50% of the questions in selected fields of the Cancer Control Module were excluded from further analyses (n⫽1,152, or 3.6% of sample adults) (6). The data collected on the Cancer Control Module for each food were in the form of a rate (times per day, week, month, or year), coded as frequency and time unit (day, week, month, or year). We used the US Department of Agriculture’s Continuing Survey of Food Intakes of Individuals (CSFII) 1994-1996 data on reported intakes over two days of 24-hour recalls (7) to make judgments about reasonable frequencies of consumption that were reported on a perday basis. We accepted frequency values reported in NHIS that were reported on a per day basis up to the maximum daily (averaged over 2 days) values (rounded to the next whole number) reported among adults (age 18 and older) in the CSFII. The maximum acceptable values for the daily report of each food were: cold cereals, 10; whole milk, 5; 2% fat milk, 6; 1% fat milk, 6; fat-free milk, 5; other milk, 5; bacon or sausage, 3; hotdogs, 2; wholegrain bread, 5; 100% fruit juice, 4; fruit, 12; salad dressing, 3; salad, 5; fried potatoes, 3; other white potatoes, 3; dried beans, 3; other vegetables, 9; pasta, 3; nuts, 3; and chips, 3. In addition, we evaluated the acceptability of frequency reports when reported in weekly and monthly time periods. For example, a report of 25 times may be most logically associated with a month or year time period, but is not so logically associated with a week time period. Therefore, missing values were assigned to reports indicating consumption more than 14 times per week and more than 60 times per month. For results presented here, we excluded individuals only for the dietary factors for which they had missing data. The total number of excluded individuals was 2,842 (8.8%) for fruits and vegetables, 3,554 (11.0%) for percentage of energy from fat,
and 3,535 (10.9%) for fiber. These numbers include those excluded for not giving substantive responses to at least 50% of the questions; those with refused or don’t know responses on individual food items asked; and those with unreasonable values. Analytical Algorithms After individuals’ foods with extreme values were assigned missing values, we proceeded to estimate each individual’s dietary factor. First, we converted the reported frequency category to mean daily number of times consumed. Second, we applied external information about portion size from CSFII 1994-1996 data. Median portion size estimates for each food asked on the screener were computed for sex and 10-year age-specific subgroups. For the fruit and vegetable servings variable, portion size was in terms of servings as reflected in the Food Guide Pyramid (8). For fruits, a serving is defined as a whole fruit, such as a medium apple; ½ cup of cut-up fruit; or ¾ cup fruit juice. For vegetables, a serving is defined as 1 cup of raw leafy vegetables, like lettuce; ½ cup of other vegetables; or ¾ cup of vegetable juice. For percentage of energy from fat and fiber variables, portion sizes of the individual food groups were in terms of grams. [Portion size estimates are available on (9).] For the NHIS data, for each food the portion-size estimates for the appropriate sex and age group of each individual were multiplied by that individual’s reported daily frequency of intake. Finally, we used linear regression models to predict the dietary factors [servings of fruits and vegetables with and without french fries, percentage of energy from fat, and fiber (g)] for each individual based on his or her portion-size adjusted responses to the Multifactor Screener. The regression coefficients in the prediction formula were estimated from CSFII 19941996 data, after square-root transformation of fruits and vegetables and cube-root transformation of fiber, to better approximate normal distributions. The resulting “screener estimates of intake” estimate intake on these transformed scales. These procedures are described more fully in the paper by Thompson and colleagues (2) and regression coefficients are reported on (9). Estimation of Distributions Under various model assumptions, the distribution of the screener estimate of intake should have approximately the same mean and median as the distribution of “true” usual intake (as estimated from 24-hour recalls). The
Table 1. Variance-adjustment factorsa for the NHISb Multifactor Screener
Sex
Square-root fruit and vegetable
Square-root fruit and vegetable (excluding french fries)
% Energy from fat
Cube-root fiber
Male Female
1.3 1.1
1.3 1.2
1.5 1.3
1.2 1.2
a
True intake standard deviation⫼screener standard deviation. NHIS⫽National Health Interview Survey.
b
354
March 2005 Volume 105 Number 3
Table 2. Estimated median (and 95% CIa) intake of fruits and vegetables (servings/day) by demographic characteristics: NHISb 2000 Fruits and Vegetables (servings) Demographic characteristic
n
Men (nⴝ12,704)
All subjects ⱖ18 y Age (y) 18-39 40-59 ⱖ60
29,532
4™™™™™™™™™ median (95% CI) ™™™™™™™™™3 5.46 (5.43-5.50) 4.52 (4.50-4.55)
12,217 10,167 7,148
5.55 (5.49-5.61) 5.42 (5.36-5.48) 5.34 (5.28-5.41)
4.48 (4.44-4.52) 4.57 (4.53-4.62) 4.54 (4.50-4.58)
4,060 4,940 19,526 1,006
5.38 (5.27-5.49) 5.99 (5.87-6.11) 5.39 (5.35-5.43) 5.63 (5.45-5.80)
4.36 (4.29-4.43) 4.79 (4.73-4.86) 4.51 (4.48-4.55) 4.52 (4.39-4.66)
Education completed (among respondents aged 25 or older) Less than high school graduate High school graduate Some college College graduate
5,390 7,677 6,936 6,190
5.33 (5.24-5.43) 5.32 (5.25-5.38) 5.39 (5.32-5.46) 5.62 (5.56-5.67)
4.40 (4.34-4.45) 4.38 (4.33-4.42) 4.57 (4.52-4.62) 4.77 (4.71-4.82)
Household income (as percent of poverty) ⬍1.25 1.25-3.49 ⬎3.49
4,654 9,264 9,459
5.52 (5.41-5.64) 5.44 (5.37-5.51) 5.46 (5.41-5.51)
4.42 (4.36-4.49) 4.49 (4.44-4.54) 4.62 (4.57-4.66)
Region Northeast Midwest South West
5,520 6,812 10,728 6,472
5.60 (5.52-5.67) 5.38 (5.30-5.46) 5.41 (5.35-5.48) 5.53 (5.46-5.60)
4.66 (4.60-4.72) 4.56 (4.52-4.61) 4.39 (4.35-4.43) 4.59 (4.53-4.66)
Urbanization Metropolitan area Nonmetropolitan area
23,594 5,938
5.47 (5.43-5.51) 5.41 (5.33-5.50)
4.54 (4.51-4.57) 4.46 (4.41-4.52)
Race/ethnicity Black non-Latino Latino White non-Latino Other non-Latino
a
Women (nⴝ16,828)
CI⫽confidence interval. NHIS⫽National Health Interview Survey.
b
variance of the screener, however, is expected to be smaller than the variance of true intake because the screener prediction formula estimates the conditional expectation of true intake given the screener responses, and in general the variance of a conditional expectation of a variable X is smaller than the variance of X itself. As a result, the screener estimates of intake cannot be used to estimate quantiles (other than median) or prevalence estimates of true intake without adjustment. In this paper we suggest a simple method for adjusting the screener estimates to obtain more realistic estimates of the variance of true usual intake; this method may be used to estimate quantiles and prevalences more or less than given cutoffs (such as the proportion meeting dietary recommendations, provided here). Using repeat 24-hour recall and screener data from the Observing Protein and Energy Nutrition Study (4) and the Eating at America’s Table Study (5), we estimated the standard deviation of true usual intake and the standard
deviation of the screener estimate of intake for each dietary factor, by sex and on the transformed scale (square root for fruits and vegetables, cube root for fiber). (In the Eating at America’s Table Study we were only able to do this for fruits and vegetables because the screener administered in that study only assessed fruit and vegetable intake). The standard deviation of true usual intake was estimated from the repeat 24-hour dietary recalls, using the National Research Council method (10,11) to adjust for day-to-day variability in the 24-hour recall. Table 1 presents our estimates of the ratio of estimated standard deviation of true intake to standard deviation of screener estimate of intake for each dietary factor. We call these ratios variance adjustment factors because they are the factors by which the screener estimate must be multiplied to have its variance equal the estimated variance of true intake. We computed a variance-adjusted screener estimate of true intake in NHIS by multiplying the screener estimate for each individual by the appropriate
March 2005 ● Journal of the AMERICAN DIETETIC ASSOCIATION
355
Table 3. Estimated mean (and 95% CIa) intake of percentage of energy from fat by demographic characteristics: NHISb 2000 Percent Energy from Fat Demographic characteristic
n
Men (nⴝ12,374)
All subjects ⱖ18 y Age (y) 18-39 40-59 ⱖ60
28,844
4™™™™™™™™™ mean (95% CI) ™™™™™™™™™3 34.0 (33.9-34.1) 32.2 (32.1-32.3)
11,925 9,964 6,955
34.4 (34.3-34.5) 34.2 (34.0-34.3) 32.6 (32.5-32.7)
32.4 (32.3-32.6) 32.6 (32.5-32.8) 31.0 (30.8-31.1)
3,954 4,814 19,093 983
34.7 (34.5-35.0) 33.7 (33.5-33.9) 33.9 (33.8-34.0) 33.4 (33.0-33.8)
33.5 (33.2-33.7) 32.1 (31.9-32.3) 32.0 (31.9-32.1) 31.7 (31.3-32.1)
Education completed (among respondents aged ⱖ25) Less than high school graduate High school graduate Some college College graduate
5,242 7,485 6,793 6,082
34.5 (34.3-34.7) 34.5 (34.4-34.7) 34.2 (34.0-34.4) 32.7 (32.6-32.9)
32.7 (32.5-32.9) 32.7 (32.6-32.9) 32.3 (32.2-32.5) 31.0 (30.8-31.1)
Household income (as percent of poverty) ⬍1.25 1.25-3.49 ⬎3.49
4,552 9,089 9,306
34.4 (34.2-34.7) 34.4 (34.3-34.6) 33.6 (33.5-33.8)
33.1 (32.8-33.3) 32.4 (32.3-32.6) 31.7 (31.5-31.8)
Region Northeast Midwest South West
5,401 6,621 10,530 6,292
33.0 (32.8-33.2) 33.8 (33.6-34.0) 34.6 (34.5-34.8) 33.9 (33.7-34.1)
31.2 (31.0-31.4) 31.8 (31.6-32.0) 32.9 (32.7-33.0) 32.3 (32.0-32.5)
Urbanization Metropolitan area Nonmetropolitan area
23,066 5,778
33.8 (33.6-33.9) 34.8 (34.6-35.0)
31.9 (31.8-32.0) 33.0 (32.8-33.3)
Race/ethnicity Black non-Latino Latino White non-Latino Other non-Latino
a
Women (nⴝ16,470)
CI⫽confidence interval. NHIS⫽National Health Interview Survey.
b
variance adjustment factor and adding a constant so that the overall mean was unchanged: Variance-adjusted screener ⫽(variance adjustment factor)⫻(unadjusted screener ⫺meanunadjusted screener)⫹meanunadjusted screener For fruits and vegetables or fiber, these adjustments were performed on the transformed scale (square root for fruits and vegetables, cube root for fiber). The resulting variables were squared or cubed, respectively, to obtain estimates on the original scale. Assuming that the variance adjustment factors appropriate to NHIS are similar to those in the Observing Protein and Energy Nutrition Study (and Eating at America’s Table Study), the variance-adjusted screener estimate of intake should have a variance closer to the estimated variance of true intake that would have been obtained from repeat 24-hour recalls using the National Research Council method.
356
March 2005 Volume 105 Number 3
Data Analysis For the NHIS data, we used the unadjusted screener estimates of intake to estimate means and medians of intake by demographic and behavioral characteristics. Means and confidence intervals for fruits and vegetables and fiber were estimated on the transformed scale and subsequently back-transformed to the original units, and as such should be considered estimates of median rather than mean intake. (This is because the distribution of intake is symmetric on the transformed scale so that median⫽mean on that scale; under the back-transformation, the mean/median loses its character as mean but retains its character as median). We used the varianceadjusted screener estimates to estimate prevalence of attaining certain levels of dietary intake: at least 5 servings of fruits and vegetables and 20 g or more fiber daily and 30% or less of total energy as fat. Behavioral variables were constructed in the following way. Leisure-time physical activity questions included fre-
Table 4. Estimated median (and 95% CIa) intake of fiber (g/day) by demographic characteristics: NHISb 2000 Fiber (g) Demographic characteristic
n
Men (nⴝ12,379)
All subjects ⱖ18 y Age (y) 18-39 40-59 ⱖ60
28,839
4™™™™™™™™™ median (95% CI) ™™™™™™™™™3 19.2 (19.1-19.3) 14.4 (14.3-14.5)
11,921 9,968 6,950
20.3 (20.0-20.5) 18.8 (18.6-19.0) 17.7 (17.5-17.9)
14.8 (14.7-15.0) 14.1 (14.0-14.3) 14.1 (13.9-14.2)
3,964 4,816 19,081 978
18.5 (18.1-18.9) 22.5 (21.9-23.1) 18.9 (18.8-19.1) 18.0 (17.5-18.5)
13.4 (13.1-13.6) 16.6 (16.3-17.0) 14.3 (14.2-14.4) 13.8 (13.4-14.3)
Education completed (among respondents aged 25 or older) Less than high school graduate High school graduate Some college College graduate
5,234 7,494 6,789 6,076
18.7 (18.4-19.1) 18.5 (18.2-18.7) 18.8 (18.6-19.1) 19.3 (19.1-19.6)
14.3 (14.1-14.5) 13.6 (13.4-13.7) 14.3 (14.1-14.4) 15.3 (15.1-15.5)
Household income (as percent of poverty) ⬍1.25 1.25-3.49 ⬎3.49
4,547 9,090 9,296
20.0 (19.5-20.5) 19.5 (19.2-19.8) 19.1 (18.9-19.2)
14.6 (14.4-14.9) 14.4 (14.3-14.6) 14.6 (14.5-14.8)
Region Northeast Midwest South West
5,399 6,616 10,533 6,291
18.8 (18.6-19.1) 19.2 (18.9-19.5) 19.0 (18.7-19.2) 20.0 (19.7-20.3)
14.4 (14.1-14.6) 14.5 (14.3-14.7) 14.0 (13.8-14.1) 15.2 (14.9-15.4)
Urbanization Metropolitan area Nonmetropolitan area
23,062 5,777
19.2 (19.1-19.4) 19.2 (18.8-19.5)
14.5 (14.4-14.6) 14.1 (13.9-14.4)
Race/ethnicity Black non-Latino Latino White non-Latino Other non-Latino
a
Women (nⴝ16,460)
CI⫽confidence interval. NHIS⫽National Health Interview Survey.
b
quency and duration of vigorous and light or moderate activities. Individuals were categorized as adherent to exercise recommendations if reporting vigorous activity of 60 or more minutes per week, moderate activity of 150 or more minutes per week, or combined vigorous and moderate activities of 150 or more minutes per week. Individuals were categorized as having some leisure-time physical activity if reporting less than that amount but engaging in any vigorous or moderate activities. Individuals were categorized as having no leisure-time activity if reporting no vigorous or moderate activities. Non–leisure-time activity was assessed by asking how respondents usually spent their day: sit during most of the day, stand during most of the day, or walk around most of the day. An overall sun-protection variable was formed from responses to five individual questions. Individuals were classified as protecting themselves from sun always or most of the time when they answered always or most of the time to any of the following sun-protecting behaviors: stay in the
shade, wear hat, wear long-sleeved shirt, use sunscreen with at least a 15 sun-protection factor. Individuals were classified as protecting themselves less often when they answered sometimes, rarely, or never to all questions. The Survey Data Analysis computer package was used to analyze the data because it takes into account the sampling pattern of the survey to calculate the standard errors, from which the 95% confidence intervals were derived (12). RESULTS Demographic Comparisons Estimated median intakes of servings of fruits and vegetables, and percentage of energy from fat and grams of fiber are presented by demographic characteristics in Tables 2-4. Men consumed more fruits and vegetables and fiber than women, but women’s percentage energy from fat was less.
March 2005 ● Journal of the AMERICAN DIETETIC ASSOCIATION
357
Table 5. Estimated percentage meeting recommendations for intake of fruits and vegetables, percentage of energy from fat, and fiber by demographic characteristics: NHISa 2000 Percentage Meeting Recommendation Fruits and Vegetables >5 Servings/day
Energy from Fat <30%
Fiber >20 g/day
Men (nⴝ12,704)
Women (nⴝ16,828)
Men (nⴝ12,350)
Women (nⴝ16,470)
Men (nⴝ12,379)
Women (nⴝ16,460)
59.2
37.0
22.7
35.1
38.5
16.2
59.7 58.5 59.3
35.8 38.9 36.1
20.2 20.6 32.0
32.3 31.0 46.5
44.9 36.7 27.2
18.9 14.9 13.2
Household income (as percent of poverty) ⬍1.25 1.25-3.49 ⬎3.49
61.2 57.0 59.5
34.8 35.2 39.4
20.4 19.4 25.3
29.7 32.0 39.4
44.2 39.8 37.4
19.0 16.6 17.2
Region Northeast Midwest South West
63.2 56.3 57.7 61.8
40.4 38.3 33.4 38.3
31.4 22.6 17.9 23.0
43.8 36.7 29.8 34.0
34.8 39.2 37.9 42.2
16.1 16.4 13.5 21.3
Urbanization Metropolitan area Nonmetropolitan area
59.5 57.9
37.1 36.5
24.4 15.8
37.2 27.6
38.2 39.4
16.5 15.1
Demographic characteristic All subjects ⱖ18 y Age (y) 18-39 40-59 ⱖ60
a
NHIS⫽National Health Interview Survey.
For servings of fruits and vegetables (Table 2), intakes were greater at younger ages among men but slightly lower among women. For both men and women, intakes of fruits and vegetables were highest among Latinos. Among men, intakes for blacks and whites were similar; among women, blacks had lower intakes than whites. For both men and women, fruit and vegetable intakes were higher for more highly educated groups. Among women, intake was somewhat higher for the highest income group, but among men, intake did not vary across income groups. Slightly higher intakes of fruits and vegetables were reported in the Northeast and the West. There was little difference in intake by urbanization. For both men and women, percentage of energy from fat was lower in the oldest age group compared with the younger age groups (Table 3). Percentage of energy from fat was lowest for those of other race/ethnicity and was highest for blacks. For both men and women, percentage of energy from fat was less for more highly educated groups and for the highest income group. Percentage of energy from fat was least in the Northeast, followed by the Midwest, West, and South, and was less in metropolitan regions compared with nonmetropolitan areas. For fiber (Table 4), estimated median intakes were greater at younger ages. For both men and women, fiber intakes were highest among Latinos, followed by whites and blacks. For both men and women, intake of fiber was highest among the most highly educated group, although high school graduates who did not attend college had
358
March 2005 Volume 105 Number 3
lower fiber intakes than did those without a high school degree. Among men, fiber intakes were higher for lower income groups, but among women, fiber intake did not vary across income groups. The highest fiber intakes were reported in the West. Fiber intake was somewhat higher among women living in metropolitan areas as compared with those living in nonmetropolitan areas. Prevalence of meeting certain levels of dietary intake for selected demographic groups is presented in Table 5. These results reflect the previously reported relationships among median intakes and demographic characteristics, but highlight particular demographic groups that could be most profitably targeted for intervention. Behavioral Characteristics Relationships between dietary intakes and other behavioral characteristics are presented in Tables 6-8. In general, reported dietary intakes were more favorable for groups with other favorable behaviors. For servings of fruits and vegetables (Table 6), median estimated intakes were greater for those who had never smoked than for current smokers, for nondrinkers than for heavy (more than 14 drinks/week) alcohol drinkers, for those engaged in sun-exposure protection, for normalweight than for obese individuals, for more physically active groups, and for daily/frequent supplement users than for nonusers. Percentage of energy from fat (Table 7) was less for
Table 6. Estimated median (and 95% CIa) intake of fruits and vegetables (servings/day) by behavioral characteristics: NHISb 2000 Fruits and Vegetables (servings) Behavioral characteristic
n
Men (nⴝ12,704)
Women (nⴝ16,828)
4™™™™™™™™™ median (95% CI) ™™™™™™™™™ 3 Smoking Current smoker Ex-smoker Never smoker
6,855 6,495 16,147
5.22 (5.15-5.30) 5.43 (5.37-5.49) 5.61 (5.56-5.66)
4.23 (4.18-4.28) 4.58 (4.53-4.63) 4.61 (4.58-4.65)
Alcohol consumption in past year (drinks/week) None ⬍1 1-7 7.1-14 ⬎14
11,559 8,260 6,767 1,609 903
5.47 (5.40-5.53) 5.44 (5.37-5.51) 5.51 (5.45-5.58) 5.42 (5.32-5.53) 5.28 (5.15-5.42)
4.53 (4.48-4.57) 4.53 (4.48-4.57) 4.58 (4.53-4.63) 4.29 (4.15-4.43) 4.31 (4.00-4.63)
Sun exposure protective behavior Always/most of the time Less often
16,145 11,444
5.50 (5.45-5.55) 5.44 (5.38-5.49)
4.63 (4.59-4.66) 4.40 (4.35-4.44)
Body mass index (based on self-reported height and weight) Underweight (⬍18.5) Normal (18.5-24.9) Overweight (25.0-29.9) Obese (ⱖ30)
641 11,580 10,125 7,186
5.74 (5.22-6.28) 5.54 (5.48-5.60) 5.46 (5.41-5.51) 5.33 (5.26-5.40)
4.58 (4.43-4.73) 4.59 (4.55-4.63) 4.52 (4.47-4.57) 4.41 (4.36-4.46)
Leisure-time physical activity None Some Adherent
12,183 4,708 12,160
5.25 (5.19-5.31) 5.32 (5.24-5.40) 5.66 (5.61-5.71)
4.34 (4.30-4.38) 4.53 (4.48-4.59) 4.73 (4.69-4.77)
Nonleisure daily activities Sit most of the day Stand most of the day Walk around most of the day
10,873 3,873 14,379
5.34 (5.28-5.40) 5.49 (5.38-5.59) 5.54 (5.49-5.59)
4.37 (4.33-4.41) 4.50 (4.41-4.58) 4.65 (4.62-4.68)
Vitamin/mineral supplement use in past year Daily use Frequent use Occasional use Not used
10,628 881 4,243 13,767
5.69 (5.63-5.75) 5.51 (5.33-5.70) 5.40 (5.31-5.49) 5.35 (5.29-5.40)
4.74 (4.70-4.78) 4.75 (4.62-4.90) 4.42 (4.36-4.48) 4.34 (4.30-4.38)
a
CI⫽confidence interval. NHIS⫽National Health Interview Survey.
b
former smokers and nonsmokers than for current smokers, for nondrinkers than for heavy alcohol drinkers, for those engaged in sun exposure protection, for the nonobese weight groups compared with the obese groups, for those reporting more leisure-time physical activity, and for daily supplement users than for nonusers. Percentage of energy from fat was less for men who reported sitting during their nonleisure daily activities compared with those who reported standing or walking around, and for women who reported sitting or walking around compared with those who reported standing. Fiber intake (Table 8) was higher for nonsmokers than for current or ex-smokers, for the nonobese weight groups compared with the obese groups, for those reporting more
leisure-time physical activity, for those reporting standing or walking around compared with those reporting sitting, and for daily supplement users compared with nonusers. DISCUSSION There is a well-recognized need for periodic detailed foodintake data about the US population. However, within the national nutrition monitoring system, the large time and cost burden of dietary recalls has restricted their use to a single survey (NHANES) that alone cannot answer all diet-related questions of interest. For example, NHANES does not include measures of sun-protection
March 2005 ● Journal of the AMERICAN DIETETIC ASSOCIATION
359
Table 7. Estimated mean (and 95% CIa) intake of percentage of energy from fat by behavioral characteristics: NHISb 2000 Energy from Fat (%) Behavioral characteristic
n
Men (nⴝ12,374)
Women (nⴝ16,470)
4™™™™™™™™™ mean (95% CI) ™™™™™™™™™3 Smoking Current smoker Ex-smoker Never smoker
6,695 6,348 15,768
35.5 (35.3-35.6) 33.3 (33.2-33.5) 33.5 (33.4-33.6)
34.0 (33.8-34.1) 31.8 (31.6-32.0) 31.7 (31.6-31.8)
Alcohol consumption in past year (drinks/week) None ⬍1 1-7 7.1-14 ⬎14
11,262 8,126 6,638 1,569 881
33.8 (33.6-33.9) 33.7 (33.6-33.9) 33.8 (33.7-34.0) 34.6 (34.3-34.9) 35.7 (35.3-36.0)
32.1 (32.0-32.2) 32.2 (32.0-32.3) 32.1 (31.9-32.3) 33.1 (32.6-33.5) 34.2 (33.3-35.0)
Sun exposure protective behavior Always/most of the time Less often
15,812 11,192
33.8 (33.7-33.9) 34.2 (34.1-34.3)
31.9 (31.8-32.0) 32.7 (32.5-32.8)
Body mass index (based on self-reported height and weight) Underweight (⬍18.5) Normal (18.5-24.9) Overweight (25.0-29.9) Obese (ⱖ30)
619 11,305 9,920 7,000
33.9 (33.1-34.6) 34.0 (33.8-34.1) 33.8 (33.7-34.0) 34.2 (34.1-34.4)
32.3 (31.8-32.9) 31.9 (31.8-32.0) 32.1 (31.9-32.2) 32.8 (32.6-32.9)
Leisure-time physical activity None Some Adherent
11,853 4,632 11,923
34.4 (34.2-34.5) 33.8 (33.6-34.0) 33.7 (33.6-33.8)
32.8 (32.6-32.9) 32.1 (31.9-32.3) 31.5 (31.4-31.7)
Nonleisure daily activities Sit most of the day Stand most of the day Walk around most of the day
10,622 3,780 14,065
33.5 (33.3-33.6) 34.4 (34.2-34.6) 34.2 (34.1-34.3)
32.1 (32.0-32.3) 32.6 (32.4-32.9) 32.1 (32.0-32.2)
Vitamin/mineral supplement use in past year Daily use Frequent use Occasional use Not used
10,413 878 4,150 13,395
33.2 (33.0-33.3) 33.4 (33.0-33.8) 33.9 (33.7-34.1) 34.5 (34.3-34.6)
31.4 (31.3-31.6) 32.0 (31.4-32.5) 32.4 (32.1-32.6) 32.8 (32.7-33.0)
a
CI⫽confidence interval. NHIS⫽National Health Interview Survey.
b
behavior, family history of cancer, or detailed cancerscreening behaviors, and thus interrelationships between these variables and diet cannot be assessed. Furthermore, the NHANES sample is limited in the number of individuals from various racial/ethnic groups. The current study shows that short dietary assessment instruments, when coupled with measures obtained from more detailed surveys, may give useful information about particular aspects of the diet. Many of the bivariate relationships found in NHIS between the estimated dietary factors and demographic and behavioral characteristics were consistent with findings from previous studies using more accurate instruments. Krebs-Smith and Kantor (13), estimating mean fruit and vegetable servings for
360
March 2005 Volume 105 Number 3
various groups from 24-hour recalls performed in the CSFII 1994-1996, reported relationships with sex, education, region, and urbanization similar to our results. Analyses of dietary intakes in ethnic groups from CSFII 1994-1996 were also consistent with these NHIS findings (14), as were bivariate relationships between these diet indicators and education and poverty status from NHANES 1999-2000 (unpublished data). Baseline measures in the Working Well Trial showed higher fruit and vegetable and fiber and lower fat intakes among more highly-educated individuals; higher intake of cooked dried beans was found in less-educated individuals (15). Analysis of CSFII 1985 24-hour recall data for women found that intakes of fruits and vegetables, fiber, and
Table 8. Estimated median (and 95% CIa) intake of fiber (g/day) by behavioral characteristics: NHISb 2000 Fiber (g) Behavioral characteristic
n
Men (nⴝ12,379)
Women (nⴝ16,460)
4™™™™™™™™™ median (95% CI) ™™™™™™™™™ 3 Smoking Current smoker Ex-smoker Never smoker
6,695 6,352 15,759
18.7 (18.4-19.0) 18.5 (18.3-18.7) 19.9 (19.7-20.1)
13.4 (13.3-13.6) 14.3 (14.2-14.5) 14.8 (14.6-14.9)
Alcohol consumption in past year (drinks/week) None ⬍1 1-7 7.1-14 ⬎14
11,274 8,113 6,632 1,568 881
19.2 (18.9-19.5) 19.1 (18.9-19.4) 19.3 (19.1-19.6) 19.2 (18.8-19.6) 18.7 (18.3-19.2)
14.5 (14.3-14.6) 14.4 (14.2-14.5) 14.5 (14.4-14.7) 13.5 (13.1-14.0) 13.7 (12.8-14.7)
Sun exposure protective behavior Always/most of the time Less often
15,811 11,187
19.3 (19.2-19.5) 19.2 (18.9-19.4)
14.7 (14.6-14.8) 14.0 (13.9-14.2)
Body mass index (based on self-reported height and weight) Underweight (⬍18.5) Normal (18.5-24.9) Overweight (25.0-29.9) Obese (ⱖ30)
622 11,315 9,906 6,996
20.3 (18.7-22.0) 19.7 (19.4-19.9) 19.1 (18.9-19.3) 18.7 (18.4-18.9)
14.6 (14.1-15.2) 14.7 (14.6-14.9) 14.3 (14.1-14.5) 13.9 (13.7-14.0)
Leisure-time physical activity None Some Adherent
11,850 4,627 11,925
18.5 (18.2-18.7) 18.7 (18.4-19.0) 19.9 (19.7-20.1)
13.8 (13.6-13.9) 14.4 (14.2-14.6) 15.1 (15.0-15.3)
Nonleisure daily activities Sit most of the day Stand most of the day Walk around most of the day
10,617 3,772 14,075
18.7 (18.5-18.9) 19.6 (19.2-20.0) 19.5 (19.3-19.6)
13.9 (13.8-14.0) 14.5 (14.2-14.8) 14.8 (14.6-14.9)
Vitamin/mineral supplement use in past year Daily use Frequent use Occasional use Not used
10,404 877 4,146 13,405
19.6 (19.4-19.9) 19.4 (18.7-20.0) 19.5 (19.1-19.9) 18.9 (18.7-19.1)
15.0 (14.8-15.1) 15.1 (14.6-15.6) 14.4 (14.1-14.6) 13.8 (13.7-14.0)
a
CI⫽confidence interval. NHIS⫽National Health Interview Survey.
b
vitamin C were more and intake of dietary cholesterol was less among smokers than among nonsmokers (16). Analyses of NHANES II 24-hour recall data indicated that smokers were less likely than nonsmokers to consume vegetables, fruits, high-fiber grains, and low-fat milk, and had lower intakes of vitamin C, folate, fiber, and vitamin A (17), consistent with our findings. In addition, qualitative associations between dietary variables estimated in this study and several demographic variables such as sex, race/ethnicity, tobacco use, and others are similar to those found in NHANES III (18). Maskarinec and colleagues found that among women, dietary patterns with large amounts of vegetables, fruits, or dried beans were negatively (or inversely) related to body mass
index (19). Eaton and colleagues found in the Pawtucket Heart Health Program that regular physical activity was positively related to more fiber and fruit and vegetable intake and less total and saturated fat (20). Gillman and colleagues, in the Harvard Pilgrim Health Care population, a managed care organization in New England, found that sedentary individuals consumed fewer fruits and vegetables and fiber and more saturated fat, trans fat, and dietary cholesterol (21). However, not all relationships noted in earlier surveys and studies were consistent with these NHIS findings. Results from NHANES III (1988-1994) indicate that adherence to recommendations concerning fruit and vegetable consumption increases with age (18). Similarly,
March 2005 ● Journal of the AMERICAN DIETETIC ASSOCIATION
361
Krebs-Smith and Kantor found in CSFII 1994-1996 that individuals age 60 years and older had higher fruit and vegetable intake than did younger adults (13); analyses of NHANES 1999-2000 data also found higher fruit and vegetable intake in older than in younger adults (unpublished data). Therefore, the somewhat inconsistent finding in NHIS may reflect the limitations in accuracy of this and other short methods. Estimates of intake reported here are not comparable to previous NHIS intake estimates (22-24) using different methods. The NHIS estimates provided here are more similar to those from multiple 24-hour recall data, adjusted for intraindividual variation, because they were developed and calibrated to that methodology. However, given the brevity and nature of the screener instrument, these NHIS estimates are not as accurate as those based on 24-hour recalls. Methods that include only frequency of intake information without portion-size information are less similar because this NHIS method explicitly incorporates the typically larger portion sizes ingested by males compared with females and by younger individuals compared with older individuals. Thus, some studies using short instruments with frequency questions only have reported higher intakes of fruits and vegetables among women than men (25). The ability of the Multifactor Screener to reflect true dietary intakes may differ for different subgroups of the population, for example, for certain ethnic groups or for less-educated individuals. As described earlier (2), data evaluating the performance of this screener derived from three population studies; however, respondents in these studies were primarily white, well-educated individuals. More research is needed to examine the screener’s performance among less-advantaged population groups, and, if needed, to further refine the instrument. Estimation of the distributions of intakes from the screener has a further limitation. The estimation of an adjustment factor to better estimate variances in the overall distribution of intakes was based mainly on data from one external validation dataset [Observing Protein and Energy Nutrition Study (4)]. Whereas analysis of the fruit and vegetable servings data in another external validation dataset [Eating at America’s Table Study (5)] indicated similar relationships, the assumption is that the adjustment factor (ie, the ratio of the standard deviation of the true intake to the standard deviation of the screener intake), is approximately the same in the NHIS population as it is in the Observing Protein and Energy Nutrition Study. Because of this further limitation, the estimates of the distributions of usual dietary intake in NHIS should be considered only as approximations. The overall consistency of the NHIS results to other published work can be interpreted as external validity of the methodology and dietary data in the Cancer Control Module. Although these NHIS estimates of intake are not as accurate as more detailed methods, they are useful to examine interrelationships between diet and other factors, thus allowing characterization of population subgroups to direct targeted interventions. A similar short dietary instrument is planned for use in the upcoming NHIS 2005. In addition, the Multifactor Screener used in the NHIS may prove useful for state or local surveys in which the use of a dietary recall is
362
March 2005 Volume 105 Number 3
not feasible. The national NHIS estimates presented here provide a useful benchmark for comparison of results from such studies. CONCLUSIONS ●
●
●
In the absence of more accurate dietary methods, the Multifactor Screener may provide useful information to characterize population median intakes of fruits and vegetables, fiber, and percentage of energy from fat. Interrelationships between diet, as measured by the Multifactor Screener, and other factors in NHIS can be usefully examined. For example, the interrelationships of fruit and vegetable consumption, cancer-screening behavior, and leisure-time physical activity could be examined among different smoking status groups. The NHIS diet estimates derived from the Multifactor Screener can be used as comparison data for other studies using the Multifactor Screener.
The authors thank Susan M. Krebs-Smith, MPH, PhD, RD, Chief, Risk Factor Monitoring and Methods Branch, Applied Research Program, Division of Cancer Control and Population Sciences, National Cancer Institute, for her helpful editorial comments. References 1. Thompson FE, Subar AF. Dietary Assessment Methodology. Nutrition in the Prevention and Treatment of Disease. Ed: Coulston AM, Rock CL, Monsen ER. San Diego, CA: Academic Press; 2001. 2. Thompson FE, Midthune D, Subar AF, Kahle LL, Schatzkin A, Kipnis V. Performance of a short tool to assess dietary intakes of fruits and vegetables, percent energy from fat, and fibre. Public Health Nutr. 2004;7:1097-1106. 3. National Center for Health Statistics. Data file documentation, National Health Interview Survey, 2000 (machine readable datafile and documentation). Hyattsville, MD: National Center for Health Statistics, 2001. 4. Subar AF, Kipnis V, Troiano RP, Midthune D, Schoeller DA, Bingham S, Sharbaugh CO, Trabulsi J, Runswick S, Ballard-Barbash R, Sunshine J, Schatzkin A. Using intake biomarkers to evaluate the extent of dietary misreporting in a large sample of adults: The Observing Protein and Energy Nutrition Study Study. Am J Epidemiol. 2003;158:1-13. 5. Subar AF, Thompson FE, Kipnis V, Midthune D, Hurwitz P, McNutt S, McIntosh A, Rosenfeld S. Comparative validation of the Block, Willett, and National Cancer Institute food frequency questionnaires: The Eating at America’s Table Study. Am J Epidemiol. 2001;154:1089-1099. 6. National Center for Health Statistics Web site. Available at: www.cdc.gov/nchs/nhis.htm#2000_NHIS. Accessed May 2004. 7. US Department of Agriculture. What we eat in America 1994-96, Continuing Survey of Food Intakes by Individuals (CSFII) 1994-1996. [Public Use CD-Rom]. 1998. 8. Food Guide Pyramid. A Guide to Daily Food Choices.
9.
10. 11. 12. 13. 14.
15.
16. 17.
Washington, DC: US Department of Agriculture, Human Nutrition Information Service; 1992. Home and Garden Bulletin No. 252. US National Cancer Institute, Division of Cancer Control and Population Sciences [Applied Research Program Web site]. Available at: www.appliedresearch. cancer.gov. Accessed May 2004. National Research Council. Nutrient Adequacy: Assessment Using Food Consumption Surveys. Washington, DC: National Academy Press; 1986. Institute of Medicine. Dietary Reference Intakes: Applications in Dietary Assessment. Washington, DC: National Academy Press; 2000. Research Triangle Institute. SUDAAN User’s Manual, release 8.0. Research Triangle Park, NC: Research Triangle Institute; 2001. Krebs-Smith SM, Kantor LS. Choose a variety of fruits and vegetables daily: Understanding the complexities. J Nutr. 2001;131(2S-1):S487-S501. Kumanyika SK, Krebs-Smith SM. Preventive nutrition issues in ethnic and socioeconomic groups in the United States. In: Bendich A, Deckelbaum RJ, eds. Primary and Secondary Preventive Nutrition. Totowa, NJ: Humana Press, Inc; 2000. Hunt M.K, Stoddard AM, Glanz K, Hebert JR, Probart C, Sorensen G, Thomson S, Hixson ML, Linnan L, Palombo R. Measure of food choice behavior related to intervention messages in worksite health promotion. J Nutr Ed. 1997;29:3-11. Larkin FA, Basiotis PP, Riddick HA, Sykes KE, Pao EM. Dietary patterns of women smokers and nonsmokers. J Am Diet Assoc. 1990;90:230-237. Subar AF, Harlan LC, Mattson ME. Food and nutrient intake differences between smokers and non-
18. 19. 20.
21.
22.
23. 24.
25.
smokers in the US. Am J Public Health. 1990;80: 1323-1329. Berrigan D, Dodd K, Troiano R, Krebs-Smith SM, Ballard Barbash R. Patterns of health behavior in US adults. Prev Med. 2003;36:615-623. Maskarinec G, Novotny R, Tasaki K. Dietary patterns are associated with body mass index in multiethnic women. J Nutr. 2000;130:3068-3072. Eaton CB, McPhillips JB, Gans KM, Garber CE, Assaf AR, Lasater TM, Carleton RA. Cross-sectional relationship between diet and physical activity in two southeastern New England communities. Am J Prev Med. 1995;11:238-244. Gillman MW, Pinto BM, Tennstedt S, Glanz K, Marcus B, Friedman RH. Relationships of physical activity with dietary behaviors among adults. Prev Med. 2001; 32:295-301. Block G, Subar AF. Estimates of nutrient intake from a food frequency questionnaire: The 1987 National Health Interview Survey. J Am Diet Assoc. 1992;92: 969-977. Breslow RA, Subar AF, Patterson BH, Block G. Trends in food intake: The 1987 and 1992 National Health Interview Surveys. Nutr Cancer. 1997;28:86-92. Norris J, Harnack L, Carmichael S, Pouane T, Wakimoto P, Block G. US trends in nutrient intake: The 1987 and 1992 National Health Interview Surveys. Am J Public Health. 1997;87:740-746. Li R, Serdula M, Bland S, Mokdad A, Bowman B, Nelson D. Trends in fruit and vegetable consumption among adults in 16 US states: Behavioral Risk Factor Surveillance System, 1990-1996. Am J Public Health. 2000;90:777-781.
March 2005 ● Journal of the AMERICAN DIETETIC ASSOCIATION
363
Continuing Education Questionnaire, page 487 Meets Learning Need Codes 3000, 3010, 4000, and 9070
Dietary Intake Estimates in the National Health Interview Survey, 2000: Methodology, Results, and Interpretation FRANCES E. THOMPSON, PhD, MPH; DOUGLAS MIDTHUNE, MS; AMY F. SUBAR, PhD, MPH, RD; TIMOTHY MCNEEL; DAVID BERRIGAN, PhD, MPH; VICTOR KIPNIS, PhD
ABSTRACT Objectives To describe the implementation of the Multifactor Screener in the 2000 National Health Interview Survey (NHIS); to provide intake estimates for fruits and vegetables, fiber, and percentage of energy from fat by various demographic and behavioral characteristics; and to discuss the strengths and weaknesses of the method. Design/Subjects The 2000 NHIS was conducted in 38,632 households in a clustered sample representative of the 48 contiguous states in the United States. The Cancer Control Module was administered to adults (18 years and older) and included 17 dietary intake questions. Analyses After excluding individuals with missing data or unlikely values on the diet questions, we computed individual intake of servings of fruits and vegetables, percentage of energy from fat, and grams of fiber. We estimated median intakes and distributions of intakes using adjusted variance estimates. We present bivariate relationships between diet and demographics and
F. E. Thompson is an epidemiologist, A. F. Subar is a research nutritionist, D. Berrigan is a cancer prevention fellow, and D. Midthune and V. Kipnis are mathematical statisticians, National Cancer Institute, Bethesda, MD. T. McNeel is with Information Management Services, Inc, Silver Spring, MD. Address correspondence to: Frances E. Thompson, MPH, PhD, Risk Factor Monitoring and Methods Branch, Applied Research Program, Division of Cancer Control and Population Sciences, National Cancer Institute, EPN 4016, 6130 Executive Blvd, MSC 7344, Bethesda, MD 20892-7344. E-mail: [email protected]. GOV Copyright © 2005 by the American Dietetic Association. 0002-8223/05/10503-0005$30.00/0 doi: 10.1016/j.jada.2004.12.032
352
Journal of the AMERICAN DIETETIC ASSOCIATION
diet and behavioral characteristics in almost 30,000 adults in the United States. Results In general, intakes of these dietary factors were closer to recommendations among well-educated individuals, those engaged in other healthful behaviors, and underweight and normal weight individuals. Latinos had higher intakes of fruits and vegetables (men: 6 servings; women: 4.8 servings) and fiber (men: 23 g; women: 17 g), and generally a lower percentage of energy intake from fat (men: 33.7%; women: 32.1%) than did non-Latino whites (men: 5.4 servings; women: 4.5 servings; men: 19 g; women: 14 g; men: 33.9%; women: 32.0%) and non-Latino blacks (men: 5.4 servings; women: 4.4 servings; men: 19 g; women: 13 g; men: 34.7%; women: 33.5%). The strengths and limitations of the short dietary assessment method are discussed. Conclusions The Multifactor Screener in NHIS, when used in conjunction with external reference data, provides reasonable estimates for three dietary factors and suggests relationships between intakes and other characteristics that are consistent with other data. Thus, these NHIS estimates could provide useful national benchmarks for local surveys using the same instrument. J Am Diet Assoc. 2005;105:352-363.
T
he interviewer-administered 24-hour dietary recall continues to be generally recognized as the most accurate and complete self-reported information about an individual’s diet for a given day (1). Thus, recalls have been used in national surveys to characterize the US population’s dietary intake and to track changes over time. However, because the 24-hour recall necessitates highly-trained interviewers, a heavy respondent burden, and complex coding, it is prohibitively expensive for many research applications. In some situations, self-administered food frequency questionnaires, which often may be optically scanned for inexpensive data entry, have been used. However,
© 2005 by the American Dietetic Association
Questions asked on Multifactor Screener: NHIS 2000 How many times per day, week, or month did you USUALLY eat (or drink): . . . cold cereals? What kind of milk did you usually use? . . . milk, either to drink or on cereal? Response categories were: whole milk, 2% fat, 1% fat, 1⁄2% milk, nonfat or skim milk, or other. . . . bacon or sausage, not including low-fat, light, or turkey varieties? . . . hot dogs made of beef or pork? . . . whole-grain bread including toast, rolls, and in sandwiches? Whole-grain breads include whole wheat, rye, oatmeal, and pumpernickel. . . . 100% fruit juice such as orange, grapefruit, apple, and grape juices? Do NOT count fruit drinks such as Kool-Aid,a lemonade, cranberry juice cocktail, Hi-C,b and Tang.a . . . fruit? COUNT fresh, frozen, or canned fruit. Do NOT count juices. . . . regular-fat salad dressing or mayonnaise, including on salad and sandwiches? . . . lettuce or green leafy salad, with or without other vegetables? . . . french fries, home fries, or hash brown potatoes? . . . other white potatoes? COUNT baked potatoes, boiled potatoes, mashed potatoes, and potato salad. . . . cooked dried beans, such as refried beans, baked beans, bean soup, and pork and beans? . . . not counting what you just indicated (about lettuce salads, white potatoes, cooked dried beans), and not counting rice . . . OTHER vegetables? . . . any kind of pasta? COUNT spaghetti, noodles, macaroni and cheese, pasta salad, and any other kind of pasta. . . . peanuts, walnuts, seeds, or other nuts, not including nut butters. . . . regular fat potato chips, tortilla chips, or corn chips? Do NOT include low-fat chips. (Additional help was given for individual questions as the need arose.) Response categories were recorded for 1) Number of times (0 to 94; 95⫽95⫹) and 2) time period (day, week, month, year). Figure. Questions asked on Multifactor Screener: National Health Interview Survey 2000 (3). aKraft Foods, Northfield, IL. bCoca-Cola Co, Atlanta, GA.
complete quantitative food frequency questionnaires themselves often require up to an hour to complete, and thus are unsuitable for many studies. Various shorter tools that measure a limited number of dietary factors rather than the entire diet have been developed (1). One of these—the Multifactor Screener, developed by Thompson and colleagues (2)—assesses intakes of fruits and vegetables, fat, and fiber, and was included in the 2000 National Health Interview Survey (NHIS). The purpose of this paper is to describe the implementation of the screener in the 2000 NHIS; to provide intake estimates of intakes of fruits and vegetables, fiber, and percentage of energy from fat by various demographic and behavioral characteristics; and to discuss the strengths and weaknesses of the method. METHODS Study Sample and Design The NHIS is conducted annually by the National Center for Health Statistics to ascertain a variety of self-reported health behaviors and conditions. Periodically, a Cancer Control Module is included to obtain information pertinent to cancer researchers and not asked in the other US national health survey, the National Health and Nutrition Examination Survey (NHANES). The 2000 Cancer Control Module supplement (3) consisted of questions about diet, physical activity, tobacco use, sun exposure, cancer screening, genetic testing, family history of cancer, drug use, and Latino acculturation. Because the 2000 Cancer Control Module was constrained to take no longer
than 20 minutes, only a limited number of questions about diet were included. The goal of the diet questions was to allow characterization of the diets of subgroups of the population in terms of servings of fruits and vegetables, grams of fiber, and percentage of energy from fat for examination of interrelationships among other cancerrelated behaviors and disease risk factors. The NHIS consists of a clustered, randomized sample of households from the 48 contiguous states. The 2000 survey was administered in the home by computerassisted personal interviewing. The 2000 Cancer Control Module supplement was administered to one adult sample person selected randomly from men and women age 18 and older living in the households participating in the NHIS. In 2000, information for the NHIS was collected from 38,632 households, 88.9% of those eligible for interview. From the participating households, 32,374 adults were interviewed, 82.6% of those eligible. Thus, the overall response rate for the sample adults was 72.1% (3). Multifactor Screener The Multifactor Screener (shown in the Figure) asks frequency of use information for 16 food groupings and one question about type of milk consumed; portion-size information is not asked. Development of the Multifactor Screener reflects a novel approach to the construction and analysis of short instruments. Using nationally representative food intake data, the foods selected for inclusion in the screener were those “explaining” the most variability in dietary intakes. Scoring algorithms to translate fre-
March 2005 ● Journal of the AMERICAN DIETETIC ASSOCIATION
353
quency of use responses into estimated daily intakes using portion-size information from an external source were also developed. We evaluated this screener’s performance with several external datasets (2), principally in the Observing Protein and Energy Nutrition Study conducted in 1999-2000 (4) and the Eating at America’s Table Study conducted in 1997-1998 (5). The Multifactor Screener in the NHIS differed slightly from that in the Observing Protein and Energy Nutrition Study in that an additional category of other milk to the query about type of milk usually consumed was present in NHIS. Data Processing Individuals who did not give substantive responses to at least 50% of the questions in selected fields of the Cancer Control Module were excluded from further analyses (n⫽1,152, or 3.6% of sample adults) (6). The data collected on the Cancer Control Module for each food were in the form of a rate (times per day, week, month, or year), coded as frequency and time unit (day, week, month, or year). We used the US Department of Agriculture’s Continuing Survey of Food Intakes of Individuals (CSFII) 1994-1996 data on reported intakes over two days of 24-hour recalls (7) to make judgments about reasonable frequencies of consumption that were reported on a perday basis. We accepted frequency values reported in NHIS that were reported on a per day basis up to the maximum daily (averaged over 2 days) values (rounded to the next whole number) reported among adults (age 18 and older) in the CSFII. The maximum acceptable values for the daily report of each food were: cold cereals, 10; whole milk, 5; 2% fat milk, 6; 1% fat milk, 6; fat-free milk, 5; other milk, 5; bacon or sausage, 3; hotdogs, 2; wholegrain bread, 5; 100% fruit juice, 4; fruit, 12; salad dressing, 3; salad, 5; fried potatoes, 3; other white potatoes, 3; dried beans, 3; other vegetables, 9; pasta, 3; nuts, 3; and chips, 3. In addition, we evaluated the acceptability of frequency reports when reported in weekly and monthly time periods. For example, a report of 25 times may be most logically associated with a month or year time period, but is not so logically associated with a week time period. Therefore, missing values were assigned to reports indicating consumption more than 14 times per week and more than 60 times per month. For results presented here, we excluded individuals only for the dietary factors for which they had missing data. The total number of excluded individuals was 2,842 (8.8%) for fruits and vegetables, 3,554 (11.0%) for percentage of energy from fat,
and 3,535 (10.9%) for fiber. These numbers include those excluded for not giving substantive responses to at least 50% of the questions; those with refused or don’t know responses on individual food items asked; and those with unreasonable values. Analytical Algorithms After individuals’ foods with extreme values were assigned missing values, we proceeded to estimate each individual’s dietary factor. First, we converted the reported frequency category to mean daily number of times consumed. Second, we applied external information about portion size from CSFII 1994-1996 data. Median portion size estimates for each food asked on the screener were computed for sex and 10-year age-specific subgroups. For the fruit and vegetable servings variable, portion size was in terms of servings as reflected in the Food Guide Pyramid (8). For fruits, a serving is defined as a whole fruit, such as a medium apple; ½ cup of cut-up fruit; or ¾ cup fruit juice. For vegetables, a serving is defined as 1 cup of raw leafy vegetables, like lettuce; ½ cup of other vegetables; or ¾ cup of vegetable juice. For percentage of energy from fat and fiber variables, portion sizes of the individual food groups were in terms of grams. [Portion size estimates are available on (9).] For the NHIS data, for each food the portion-size estimates for the appropriate sex and age group of each individual were multiplied by that individual’s reported daily frequency of intake. Finally, we used linear regression models to predict the dietary factors [servings of fruits and vegetables with and without french fries, percentage of energy from fat, and fiber (g)] for each individual based on his or her portion-size adjusted responses to the Multifactor Screener. The regression coefficients in the prediction formula were estimated from CSFII 19941996 data, after square-root transformation of fruits and vegetables and cube-root transformation of fiber, to better approximate normal distributions. The resulting “screener estimates of intake” estimate intake on these transformed scales. These procedures are described more fully in the paper by Thompson and colleagues (2) and regression coefficients are reported on (9). Estimation of Distributions Under various model assumptions, the distribution of the screener estimate of intake should have approximately the same mean and median as the distribution of “true” usual intake (as estimated from 24-hour recalls). The
Table 1. Variance-adjustment factorsa for the NHISb Multifactor Screener
Sex
Square-root fruit and vegetable
Square-root fruit and vegetable (excluding french fries)
% Energy from fat
Cube-root fiber
Male Female
1.3 1.1
1.3 1.2
1.5 1.3
1.2 1.2
a
True intake standard deviation⫼screener standard deviation. NHIS⫽National Health Interview Survey.
b
354
March 2005 Volume 105 Number 3
Table 2. Estimated median (and 95% CIa) intake of fruits and vegetables (servings/day) by demographic characteristics: NHISb 2000 Fruits and Vegetables (servings) Demographic characteristic
n
Men (nⴝ12,704)
All subjects ⱖ18 y Age (y) 18-39 40-59 ⱖ60
29,532
4™™™™™™™™™ median (95% CI) ™™™™™™™™™3 5.46 (5.43-5.50) 4.52 (4.50-4.55)
12,217 10,167 7,148
5.55 (5.49-5.61) 5.42 (5.36-5.48) 5.34 (5.28-5.41)
4.48 (4.44-4.52) 4.57 (4.53-4.62) 4.54 (4.50-4.58)
4,060 4,940 19,526 1,006
5.38 (5.27-5.49) 5.99 (5.87-6.11) 5.39 (5.35-5.43) 5.63 (5.45-5.80)
4.36 (4.29-4.43) 4.79 (4.73-4.86) 4.51 (4.48-4.55) 4.52 (4.39-4.66)
Education completed (among respondents aged 25 or older) Less than high school graduate High school graduate Some college College graduate
5,390 7,677 6,936 6,190
5.33 (5.24-5.43) 5.32 (5.25-5.38) 5.39 (5.32-5.46) 5.62 (5.56-5.67)
4.40 (4.34-4.45) 4.38 (4.33-4.42) 4.57 (4.52-4.62) 4.77 (4.71-4.82)
Household income (as percent of poverty) ⬍1.25 1.25-3.49 ⬎3.49
4,654 9,264 9,459
5.52 (5.41-5.64) 5.44 (5.37-5.51) 5.46 (5.41-5.51)
4.42 (4.36-4.49) 4.49 (4.44-4.54) 4.62 (4.57-4.66)
Region Northeast Midwest South West
5,520 6,812 10,728 6,472
5.60 (5.52-5.67) 5.38 (5.30-5.46) 5.41 (5.35-5.48) 5.53 (5.46-5.60)
4.66 (4.60-4.72) 4.56 (4.52-4.61) 4.39 (4.35-4.43) 4.59 (4.53-4.66)
Urbanization Metropolitan area Nonmetropolitan area
23,594 5,938
5.47 (5.43-5.51) 5.41 (5.33-5.50)
4.54 (4.51-4.57) 4.46 (4.41-4.52)
Race/ethnicity Black non-Latino Latino White non-Latino Other non-Latino
a
Women (nⴝ16,828)
CI⫽confidence interval. NHIS⫽National Health Interview Survey.
b
variance of the screener, however, is expected to be smaller than the variance of true intake because the screener prediction formula estimates the conditional expectation of true intake given the screener responses, and in general the variance of a conditional expectation of a variable X is smaller than the variance of X itself. As a result, the screener estimates of intake cannot be used to estimate quantiles (other than median) or prevalence estimates of true intake without adjustment. In this paper we suggest a simple method for adjusting the screener estimates to obtain more realistic estimates of the variance of true usual intake; this method may be used to estimate quantiles and prevalences more or less than given cutoffs (such as the proportion meeting dietary recommendations, provided here). Using repeat 24-hour recall and screener data from the Observing Protein and Energy Nutrition Study (4) and the Eating at America’s Table Study (5), we estimated the standard deviation of true usual intake and the standard
deviation of the screener estimate of intake for each dietary factor, by sex and on the transformed scale (square root for fruits and vegetables, cube root for fiber). (In the Eating at America’s Table Study we were only able to do this for fruits and vegetables because the screener administered in that study only assessed fruit and vegetable intake). The standard deviation of true usual intake was estimated from the repeat 24-hour dietary recalls, using the National Research Council method (10,11) to adjust for day-to-day variability in the 24-hour recall. Table 1 presents our estimates of the ratio of estimated standard deviation of true intake to standard deviation of screener estimate of intake for each dietary factor. We call these ratios variance adjustment factors because they are the factors by which the screener estimate must be multiplied to have its variance equal the estimated variance of true intake. We computed a variance-adjusted screener estimate of true intake in NHIS by multiplying the screener estimate for each individual by the appropriate
March 2005 ● Journal of the AMERICAN DIETETIC ASSOCIATION
355
Table 3. Estimated mean (and 95% CIa) intake of percentage of energy from fat by demographic characteristics: NHISb 2000 Percent Energy from Fat Demographic characteristic
n
Men (nⴝ12,374)
All subjects ⱖ18 y Age (y) 18-39 40-59 ⱖ60
28,844
4™™™™™™™™™ mean (95% CI) ™™™™™™™™™3 34.0 (33.9-34.1) 32.2 (32.1-32.3)
11,925 9,964 6,955
34.4 (34.3-34.5) 34.2 (34.0-34.3) 32.6 (32.5-32.7)
32.4 (32.3-32.6) 32.6 (32.5-32.8) 31.0 (30.8-31.1)
3,954 4,814 19,093 983
34.7 (34.5-35.0) 33.7 (33.5-33.9) 33.9 (33.8-34.0) 33.4 (33.0-33.8)
33.5 (33.2-33.7) 32.1 (31.9-32.3) 32.0 (31.9-32.1) 31.7 (31.3-32.1)
Education completed (among respondents aged ⱖ25) Less than high school graduate High school graduate Some college College graduate
5,242 7,485 6,793 6,082
34.5 (34.3-34.7) 34.5 (34.4-34.7) 34.2 (34.0-34.4) 32.7 (32.6-32.9)
32.7 (32.5-32.9) 32.7 (32.6-32.9) 32.3 (32.2-32.5) 31.0 (30.8-31.1)
Household income (as percent of poverty) ⬍1.25 1.25-3.49 ⬎3.49
4,552 9,089 9,306
34.4 (34.2-34.7) 34.4 (34.3-34.6) 33.6 (33.5-33.8)
33.1 (32.8-33.3) 32.4 (32.3-32.6) 31.7 (31.5-31.8)
Region Northeast Midwest South West
5,401 6,621 10,530 6,292
33.0 (32.8-33.2) 33.8 (33.6-34.0) 34.6 (34.5-34.8) 33.9 (33.7-34.1)
31.2 (31.0-31.4) 31.8 (31.6-32.0) 32.9 (32.7-33.0) 32.3 (32.0-32.5)
Urbanization Metropolitan area Nonmetropolitan area
23,066 5,778
33.8 (33.6-33.9) 34.8 (34.6-35.0)
31.9 (31.8-32.0) 33.0 (32.8-33.3)
Race/ethnicity Black non-Latino Latino White non-Latino Other non-Latino
a
Women (nⴝ16,470)
CI⫽confidence interval. NHIS⫽National Health Interview Survey.
b
variance adjustment factor and adding a constant so that the overall mean was unchanged: Variance-adjusted screener ⫽(variance adjustment factor)⫻(unadjusted screener ⫺meanunadjusted screener)⫹meanunadjusted screener For fruits and vegetables or fiber, these adjustments were performed on the transformed scale (square root for fruits and vegetables, cube root for fiber). The resulting variables were squared or cubed, respectively, to obtain estimates on the original scale. Assuming that the variance adjustment factors appropriate to NHIS are similar to those in the Observing Protein and Energy Nutrition Study (and Eating at America’s Table Study), the variance-adjusted screener estimate of intake should have a variance closer to the estimated variance of true intake that would have been obtained from repeat 24-hour recalls using the National Research Council method.
356
March 2005 Volume 105 Number 3
Data Analysis For the NHIS data, we used the unadjusted screener estimates of intake to estimate means and medians of intake by demographic and behavioral characteristics. Means and confidence intervals for fruits and vegetables and fiber were estimated on the transformed scale and subsequently back-transformed to the original units, and as such should be considered estimates of median rather than mean intake. (This is because the distribution of intake is symmetric on the transformed scale so that median⫽mean on that scale; under the back-transformation, the mean/median loses its character as mean but retains its character as median). We used the varianceadjusted screener estimates to estimate prevalence of attaining certain levels of dietary intake: at least 5 servings of fruits and vegetables and 20 g or more fiber daily and 30% or less of total energy as fat. Behavioral variables were constructed in the following way. Leisure-time physical activity questions included fre-
Table 4. Estimated median (and 95% CIa) intake of fiber (g/day) by demographic characteristics: NHISb 2000 Fiber (g) Demographic characteristic
n
Men (nⴝ12,379)
All subjects ⱖ18 y Age (y) 18-39 40-59 ⱖ60
28,839
4™™™™™™™™™ median (95% CI) ™™™™™™™™™3 19.2 (19.1-19.3) 14.4 (14.3-14.5)
11,921 9,968 6,950
20.3 (20.0-20.5) 18.8 (18.6-19.0) 17.7 (17.5-17.9)
14.8 (14.7-15.0) 14.1 (14.0-14.3) 14.1 (13.9-14.2)
3,964 4,816 19,081 978
18.5 (18.1-18.9) 22.5 (21.9-23.1) 18.9 (18.8-19.1) 18.0 (17.5-18.5)
13.4 (13.1-13.6) 16.6 (16.3-17.0) 14.3 (14.2-14.4) 13.8 (13.4-14.3)
Education completed (among respondents aged 25 or older) Less than high school graduate High school graduate Some college College graduate
5,234 7,494 6,789 6,076
18.7 (18.4-19.1) 18.5 (18.2-18.7) 18.8 (18.6-19.1) 19.3 (19.1-19.6)
14.3 (14.1-14.5) 13.6 (13.4-13.7) 14.3 (14.1-14.4) 15.3 (15.1-15.5)
Household income (as percent of poverty) ⬍1.25 1.25-3.49 ⬎3.49
4,547 9,090 9,296
20.0 (19.5-20.5) 19.5 (19.2-19.8) 19.1 (18.9-19.2)
14.6 (14.4-14.9) 14.4 (14.3-14.6) 14.6 (14.5-14.8)
Region Northeast Midwest South West
5,399 6,616 10,533 6,291
18.8 (18.6-19.1) 19.2 (18.9-19.5) 19.0 (18.7-19.2) 20.0 (19.7-20.3)
14.4 (14.1-14.6) 14.5 (14.3-14.7) 14.0 (13.8-14.1) 15.2 (14.9-15.4)
Urbanization Metropolitan area Nonmetropolitan area
23,062 5,777
19.2 (19.1-19.4) 19.2 (18.8-19.5)
14.5 (14.4-14.6) 14.1 (13.9-14.4)
Race/ethnicity Black non-Latino Latino White non-Latino Other non-Latino
a
Women (nⴝ16,460)
CI⫽confidence interval. NHIS⫽National Health Interview Survey.
b
quency and duration of vigorous and light or moderate activities. Individuals were categorized as adherent to exercise recommendations if reporting vigorous activity of 60 or more minutes per week, moderate activity of 150 or more minutes per week, or combined vigorous and moderate activities of 150 or more minutes per week. Individuals were categorized as having some leisure-time physical activity if reporting less than that amount but engaging in any vigorous or moderate activities. Individuals were categorized as having no leisure-time activity if reporting no vigorous or moderate activities. Non–leisure-time activity was assessed by asking how respondents usually spent their day: sit during most of the day, stand during most of the day, or walk around most of the day. An overall sun-protection variable was formed from responses to five individual questions. Individuals were classified as protecting themselves from sun always or most of the time when they answered always or most of the time to any of the following sun-protecting behaviors: stay in the
shade, wear hat, wear long-sleeved shirt, use sunscreen with at least a 15 sun-protection factor. Individuals were classified as protecting themselves less often when they answered sometimes, rarely, or never to all questions. The Survey Data Analysis computer package was used to analyze the data because it takes into account the sampling pattern of the survey to calculate the standard errors, from which the 95% confidence intervals were derived (12). RESULTS Demographic Comparisons Estimated median intakes of servings of fruits and vegetables, and percentage of energy from fat and grams of fiber are presented by demographic characteristics in Tables 2-4. Men consumed more fruits and vegetables and fiber than women, but women’s percentage energy from fat was less.
March 2005 ● Journal of the AMERICAN DIETETIC ASSOCIATION
357
Table 5. Estimated percentage meeting recommendations for intake of fruits and vegetables, percentage of energy from fat, and fiber by demographic characteristics: NHISa 2000 Percentage Meeting Recommendation Fruits and Vegetables >5 Servings/day
Energy from Fat <30%
Fiber >20 g/day
Men (nⴝ12,704)
Women (nⴝ16,828)
Men (nⴝ12,350)
Women (nⴝ16,470)
Men (nⴝ12,379)
Women (nⴝ16,460)
59.2
37.0
22.7
35.1
38.5
16.2
59.7 58.5 59.3
35.8 38.9 36.1
20.2 20.6 32.0
32.3 31.0 46.5
44.9 36.7 27.2
18.9 14.9 13.2
Household income (as percent of poverty) ⬍1.25 1.25-3.49 ⬎3.49
61.2 57.0 59.5
34.8 35.2 39.4
20.4 19.4 25.3
29.7 32.0 39.4
44.2 39.8 37.4
19.0 16.6 17.2
Region Northeast Midwest South West
63.2 56.3 57.7 61.8
40.4 38.3 33.4 38.3
31.4 22.6 17.9 23.0
43.8 36.7 29.8 34.0
34.8 39.2 37.9 42.2
16.1 16.4 13.5 21.3
Urbanization Metropolitan area Nonmetropolitan area
59.5 57.9
37.1 36.5
24.4 15.8
37.2 27.6
38.2 39.4
16.5 15.1
Demographic characteristic All subjects ⱖ18 y Age (y) 18-39 40-59 ⱖ60
a
NHIS⫽National Health Interview Survey.
For servings of fruits and vegetables (Table 2), intakes were greater at younger ages among men but slightly lower among women. For both men and women, intakes of fruits and vegetables were highest among Latinos. Among men, intakes for blacks and whites were similar; among women, blacks had lower intakes than whites. For both men and women, fruit and vegetable intakes were higher for more highly educated groups. Among women, intake was somewhat higher for the highest income group, but among men, intake did not vary across income groups. Slightly higher intakes of fruits and vegetables were reported in the Northeast and the West. There was little difference in intake by urbanization. For both men and women, percentage of energy from fat was lower in the oldest age group compared with the younger age groups (Table 3). Percentage of energy from fat was lowest for those of other race/ethnicity and was highest for blacks. For both men and women, percentage of energy from fat was less for more highly educated groups and for the highest income group. Percentage of energy from fat was least in the Northeast, followed by the Midwest, West, and South, and was less in metropolitan regions compared with nonmetropolitan areas. For fiber (Table 4), estimated median intakes were greater at younger ages. For both men and women, fiber intakes were highest among Latinos, followed by whites and blacks. For both men and women, intake of fiber was highest among the most highly educated group, although high school graduates who did not attend college had
358
March 2005 Volume 105 Number 3
lower fiber intakes than did those without a high school degree. Among men, fiber intakes were higher for lower income groups, but among women, fiber intake did not vary across income groups. The highest fiber intakes were reported in the West. Fiber intake was somewhat higher among women living in metropolitan areas as compared with those living in nonmetropolitan areas. Prevalence of meeting certain levels of dietary intake for selected demographic groups is presented in Table 5. These results reflect the previously reported relationships among median intakes and demographic characteristics, but highlight particular demographic groups that could be most profitably targeted for intervention. Behavioral Characteristics Relationships between dietary intakes and other behavioral characteristics are presented in Tables 6-8. In general, reported dietary intakes were more favorable for groups with other favorable behaviors. For servings of fruits and vegetables (Table 6), median estimated intakes were greater for those who had never smoked than for current smokers, for nondrinkers than for heavy (more than 14 drinks/week) alcohol drinkers, for those engaged in sun-exposure protection, for normalweight than for obese individuals, for more physically active groups, and for daily/frequent supplement users than for nonusers. Percentage of energy from fat (Table 7) was less for
Table 6. Estimated median (and 95% CIa) intake of fruits and vegetables (servings/day) by behavioral characteristics: NHISb 2000 Fruits and Vegetables (servings) Behavioral characteristic
n
Men (nⴝ12,704)
Women (nⴝ16,828)
4™™™™™™™™™ median (95% CI) ™™™™™™™™™ 3 Smoking Current smoker Ex-smoker Never smoker
6,855 6,495 16,147
5.22 (5.15-5.30) 5.43 (5.37-5.49) 5.61 (5.56-5.66)
4.23 (4.18-4.28) 4.58 (4.53-4.63) 4.61 (4.58-4.65)
Alcohol consumption in past year (drinks/week) None ⬍1 1-7 7.1-14 ⬎14
11,559 8,260 6,767 1,609 903
5.47 (5.40-5.53) 5.44 (5.37-5.51) 5.51 (5.45-5.58) 5.42 (5.32-5.53) 5.28 (5.15-5.42)
4.53 (4.48-4.57) 4.53 (4.48-4.57) 4.58 (4.53-4.63) 4.29 (4.15-4.43) 4.31 (4.00-4.63)
Sun exposure protective behavior Always/most of the time Less often
16,145 11,444
5.50 (5.45-5.55) 5.44 (5.38-5.49)
4.63 (4.59-4.66) 4.40 (4.35-4.44)
Body mass index (based on self-reported height and weight) Underweight (⬍18.5) Normal (18.5-24.9) Overweight (25.0-29.9) Obese (ⱖ30)
641 11,580 10,125 7,186
5.74 (5.22-6.28) 5.54 (5.48-5.60) 5.46 (5.41-5.51) 5.33 (5.26-5.40)
4.58 (4.43-4.73) 4.59 (4.55-4.63) 4.52 (4.47-4.57) 4.41 (4.36-4.46)
Leisure-time physical activity None Some Adherent
12,183 4,708 12,160
5.25 (5.19-5.31) 5.32 (5.24-5.40) 5.66 (5.61-5.71)
4.34 (4.30-4.38) 4.53 (4.48-4.59) 4.73 (4.69-4.77)
Nonleisure daily activities Sit most of the day Stand most of the day Walk around most of the day
10,873 3,873 14,379
5.34 (5.28-5.40) 5.49 (5.38-5.59) 5.54 (5.49-5.59)
4.37 (4.33-4.41) 4.50 (4.41-4.58) 4.65 (4.62-4.68)
Vitamin/mineral supplement use in past year Daily use Frequent use Occasional use Not used
10,628 881 4,243 13,767
5.69 (5.63-5.75) 5.51 (5.33-5.70) 5.40 (5.31-5.49) 5.35 (5.29-5.40)
4.74 (4.70-4.78) 4.75 (4.62-4.90) 4.42 (4.36-4.48) 4.34 (4.30-4.38)
a
CI⫽confidence interval. NHIS⫽National Health Interview Survey.
b
former smokers and nonsmokers than for current smokers, for nondrinkers than for heavy alcohol drinkers, for those engaged in sun exposure protection, for the nonobese weight groups compared with the obese groups, for those reporting more leisure-time physical activity, and for daily supplement users than for nonusers. Percentage of energy from fat was less for men who reported sitting during their nonleisure daily activities compared with those who reported standing or walking around, and for women who reported sitting or walking around compared with those who reported standing. Fiber intake (Table 8) was higher for nonsmokers than for current or ex-smokers, for the nonobese weight groups compared with the obese groups, for those reporting more
leisure-time physical activity, for those reporting standing or walking around compared with those reporting sitting, and for daily supplement users compared with nonusers. DISCUSSION There is a well-recognized need for periodic detailed foodintake data about the US population. However, within the national nutrition monitoring system, the large time and cost burden of dietary recalls has restricted their use to a single survey (NHANES) that alone cannot answer all diet-related questions of interest. For example, NHANES does not include measures of sun-protection
March 2005 ● Journal of the AMERICAN DIETETIC ASSOCIATION
359
Table 7. Estimated mean (and 95% CIa) intake of percentage of energy from fat by behavioral characteristics: NHISb 2000 Energy from Fat (%) Behavioral characteristic
n
Men (nⴝ12,374)
Women (nⴝ16,470)
4™™™™™™™™™ mean (95% CI) ™™™™™™™™™3 Smoking Current smoker Ex-smoker Never smoker
6,695 6,348 15,768
35.5 (35.3-35.6) 33.3 (33.2-33.5) 33.5 (33.4-33.6)
34.0 (33.8-34.1) 31.8 (31.6-32.0) 31.7 (31.6-31.8)
Alcohol consumption in past year (drinks/week) None ⬍1 1-7 7.1-14 ⬎14
11,262 8,126 6,638 1,569 881
33.8 (33.6-33.9) 33.7 (33.6-33.9) 33.8 (33.7-34.0) 34.6 (34.3-34.9) 35.7 (35.3-36.0)
32.1 (32.0-32.2) 32.2 (32.0-32.3) 32.1 (31.9-32.3) 33.1 (32.6-33.5) 34.2 (33.3-35.0)
Sun exposure protective behavior Always/most of the time Less often
15,812 11,192
33.8 (33.7-33.9) 34.2 (34.1-34.3)
31.9 (31.8-32.0) 32.7 (32.5-32.8)
Body mass index (based on self-reported height and weight) Underweight (⬍18.5) Normal (18.5-24.9) Overweight (25.0-29.9) Obese (ⱖ30)
619 11,305 9,920 7,000
33.9 (33.1-34.6) 34.0 (33.8-34.1) 33.8 (33.7-34.0) 34.2 (34.1-34.4)
32.3 (31.8-32.9) 31.9 (31.8-32.0) 32.1 (31.9-32.2) 32.8 (32.6-32.9)
Leisure-time physical activity None Some Adherent
11,853 4,632 11,923
34.4 (34.2-34.5) 33.8 (33.6-34.0) 33.7 (33.6-33.8)
32.8 (32.6-32.9) 32.1 (31.9-32.3) 31.5 (31.4-31.7)
Nonleisure daily activities Sit most of the day Stand most of the day Walk around most of the day
10,622 3,780 14,065
33.5 (33.3-33.6) 34.4 (34.2-34.6) 34.2 (34.1-34.3)
32.1 (32.0-32.3) 32.6 (32.4-32.9) 32.1 (32.0-32.2)
Vitamin/mineral supplement use in past year Daily use Frequent use Occasional use Not used
10,413 878 4,150 13,395
33.2 (33.0-33.3) 33.4 (33.0-33.8) 33.9 (33.7-34.1) 34.5 (34.3-34.6)
31.4 (31.3-31.6) 32.0 (31.4-32.5) 32.4 (32.1-32.6) 32.8 (32.7-33.0)
a
CI⫽confidence interval. NHIS⫽National Health Interview Survey.
b
behavior, family history of cancer, or detailed cancerscreening behaviors, and thus interrelationships between these variables and diet cannot be assessed. Furthermore, the NHANES sample is limited in the number of individuals from various racial/ethnic groups. The current study shows that short dietary assessment instruments, when coupled with measures obtained from more detailed surveys, may give useful information about particular aspects of the diet. Many of the bivariate relationships found in NHIS between the estimated dietary factors and demographic and behavioral characteristics were consistent with findings from previous studies using more accurate instruments. Krebs-Smith and Kantor (13), estimating mean fruit and vegetable servings for
360
March 2005 Volume 105 Number 3
various groups from 24-hour recalls performed in the CSFII 1994-1996, reported relationships with sex, education, region, and urbanization similar to our results. Analyses of dietary intakes in ethnic groups from CSFII 1994-1996 were also consistent with these NHIS findings (14), as were bivariate relationships between these diet indicators and education and poverty status from NHANES 1999-2000 (unpublished data). Baseline measures in the Working Well Trial showed higher fruit and vegetable and fiber and lower fat intakes among more highly-educated individuals; higher intake of cooked dried beans was found in less-educated individuals (15). Analysis of CSFII 1985 24-hour recall data for women found that intakes of fruits and vegetables, fiber, and
Table 8. Estimated median (and 95% CIa) intake of fiber (g/day) by behavioral characteristics: NHISb 2000 Fiber (g) Behavioral characteristic
n
Men (nⴝ12,379)
Women (nⴝ16,460)
4™™™™™™™™™ median (95% CI) ™™™™™™™™™ 3 Smoking Current smoker Ex-smoker Never smoker
6,695 6,352 15,759
18.7 (18.4-19.0) 18.5 (18.3-18.7) 19.9 (19.7-20.1)
13.4 (13.3-13.6) 14.3 (14.2-14.5) 14.8 (14.6-14.9)
Alcohol consumption in past year (drinks/week) None ⬍1 1-7 7.1-14 ⬎14
11,274 8,113 6,632 1,568 881
19.2 (18.9-19.5) 19.1 (18.9-19.4) 19.3 (19.1-19.6) 19.2 (18.8-19.6) 18.7 (18.3-19.2)
14.5 (14.3-14.6) 14.4 (14.2-14.5) 14.5 (14.4-14.7) 13.5 (13.1-14.0) 13.7 (12.8-14.7)
Sun exposure protective behavior Always/most of the time Less often
15,811 11,187
19.3 (19.2-19.5) 19.2 (18.9-19.4)
14.7 (14.6-14.8) 14.0 (13.9-14.2)
Body mass index (based on self-reported height and weight) Underweight (⬍18.5) Normal (18.5-24.9) Overweight (25.0-29.9) Obese (ⱖ30)
622 11,315 9,906 6,996
20.3 (18.7-22.0) 19.7 (19.4-19.9) 19.1 (18.9-19.3) 18.7 (18.4-18.9)
14.6 (14.1-15.2) 14.7 (14.6-14.9) 14.3 (14.1-14.5) 13.9 (13.7-14.0)
Leisure-time physical activity None Some Adherent
11,850 4,627 11,925
18.5 (18.2-18.7) 18.7 (18.4-19.0) 19.9 (19.7-20.1)
13.8 (13.6-13.9) 14.4 (14.2-14.6) 15.1 (15.0-15.3)
Nonleisure daily activities Sit most of the day Stand most of the day Walk around most of the day
10,617 3,772 14,075
18.7 (18.5-18.9) 19.6 (19.2-20.0) 19.5 (19.3-19.6)
13.9 (13.8-14.0) 14.5 (14.2-14.8) 14.8 (14.6-14.9)
Vitamin/mineral supplement use in past year Daily use Frequent use Occasional use Not used
10,404 877 4,146 13,405
19.6 (19.4-19.9) 19.4 (18.7-20.0) 19.5 (19.1-19.9) 18.9 (18.7-19.1)
15.0 (14.8-15.1) 15.1 (14.6-15.6) 14.4 (14.1-14.6) 13.8 (13.7-14.0)
a
CI⫽confidence interval. NHIS⫽National Health Interview Survey.
b
vitamin C were more and intake of dietary cholesterol was less among smokers than among nonsmokers (16). Analyses of NHANES II 24-hour recall data indicated that smokers were less likely than nonsmokers to consume vegetables, fruits, high-fiber grains, and low-fat milk, and had lower intakes of vitamin C, folate, fiber, and vitamin A (17), consistent with our findings. In addition, qualitative associations between dietary variables estimated in this study and several demographic variables such as sex, race/ethnicity, tobacco use, and others are similar to those found in NHANES III (18). Maskarinec and colleagues found that among women, dietary patterns with large amounts of vegetables, fruits, or dried beans were negatively (or inversely) related to body mass
index (19). Eaton and colleagues found in the Pawtucket Heart Health Program that regular physical activity was positively related to more fiber and fruit and vegetable intake and less total and saturated fat (20). Gillman and colleagues, in the Harvard Pilgrim Health Care population, a managed care organization in New England, found that sedentary individuals consumed fewer fruits and vegetables and fiber and more saturated fat, trans fat, and dietary cholesterol (21). However, not all relationships noted in earlier surveys and studies were consistent with these NHIS findings. Results from NHANES III (1988-1994) indicate that adherence to recommendations concerning fruit and vegetable consumption increases with age (18). Similarly,
March 2005 ● Journal of the AMERICAN DIETETIC ASSOCIATION
361
Krebs-Smith and Kantor found in CSFII 1994-1996 that individuals age 60 years and older had higher fruit and vegetable intake than did younger adults (13); analyses of NHANES 1999-2000 data also found higher fruit and vegetable intake in older than in younger adults (unpublished data). Therefore, the somewhat inconsistent finding in NHIS may reflect the limitations in accuracy of this and other short methods. Estimates of intake reported here are not comparable to previous NHIS intake estimates (22-24) using different methods. The NHIS estimates provided here are more similar to those from multiple 24-hour recall data, adjusted for intraindividual variation, because they were developed and calibrated to that methodology. However, given the brevity and nature of the screener instrument, these NHIS estimates are not as accurate as those based on 24-hour recalls. Methods that include only frequency of intake information without portion-size information are less similar because this NHIS method explicitly incorporates the typically larger portion sizes ingested by males compared with females and by younger individuals compared with older individuals. Thus, some studies using short instruments with frequency questions only have reported higher intakes of fruits and vegetables among women than men (25). The ability of the Multifactor Screener to reflect true dietary intakes may differ for different subgroups of the population, for example, for certain ethnic groups or for less-educated individuals. As described earlier (2), data evaluating the performance of this screener derived from three population studies; however, respondents in these studies were primarily white, well-educated individuals. More research is needed to examine the screener’s performance among less-advantaged population groups, and, if needed, to further refine the instrument. Estimation of the distributions of intakes from the screener has a further limitation. The estimation of an adjustment factor to better estimate variances in the overall distribution of intakes was based mainly on data from one external validation dataset [Observing Protein and Energy Nutrition Study (4)]. Whereas analysis of the fruit and vegetable servings data in another external validation dataset [Eating at America’s Table Study (5)] indicated similar relationships, the assumption is that the adjustment factor (ie, the ratio of the standard deviation of the true intake to the standard deviation of the screener intake), is approximately the same in the NHIS population as it is in the Observing Protein and Energy Nutrition Study. Because of this further limitation, the estimates of the distributions of usual dietary intake in NHIS should be considered only as approximations. The overall consistency of the NHIS results to other published work can be interpreted as external validity of the methodology and dietary data in the Cancer Control Module. Although these NHIS estimates of intake are not as accurate as more detailed methods, they are useful to examine interrelationships between diet and other factors, thus allowing characterization of population subgroups to direct targeted interventions. A similar short dietary instrument is planned for use in the upcoming NHIS 2005. In addition, the Multifactor Screener used in the NHIS may prove useful for state or local surveys in which the use of a dietary recall is
362
March 2005 Volume 105 Number 3
not feasible. The national NHIS estimates presented here provide a useful benchmark for comparison of results from such studies. CONCLUSIONS ●
●
●
In the absence of more accurate dietary methods, the Multifactor Screener may provide useful information to characterize population median intakes of fruits and vegetables, fiber, and percentage of energy from fat. Interrelationships between diet, as measured by the Multifactor Screener, and other factors in NHIS can be usefully examined. For example, the interrelationships of fruit and vegetable consumption, cancer-screening behavior, and leisure-time physical activity could be examined among different smoking status groups. The NHIS diet estimates derived from the Multifactor Screener can be used as comparison data for other studies using the Multifactor Screener.
The authors thank Susan M. Krebs-Smith, MPH, PhD, RD, Chief, Risk Factor Monitoring and Methods Branch, Applied Research Program, Division of Cancer Control and Population Sciences, National Cancer Institute, for her helpful editorial comments. References 1. Thompson FE, Subar AF. Dietary Assessment Methodology. Nutrition in the Prevention and Treatment of Disease. Ed: Coulston AM, Rock CL, Monsen ER. San Diego, CA: Academic Press; 2001. 2. Thompson FE, Midthune D, Subar AF, Kahle LL, Schatzkin A, Kipnis V. Performance of a short tool to assess dietary intakes of fruits and vegetables, percent energy from fat, and fibre. Public Health Nutr. 2004;7:1097-1106. 3. National Center for Health Statistics. Data file documentation, National Health Interview Survey, 2000 (machine readable datafile and documentation). Hyattsville, MD: National Center for Health Statistics, 2001. 4. Subar AF, Kipnis V, Troiano RP, Midthune D, Schoeller DA, Bingham S, Sharbaugh CO, Trabulsi J, Runswick S, Ballard-Barbash R, Sunshine J, Schatzkin A. Using intake biomarkers to evaluate the extent of dietary misreporting in a large sample of adults: The Observing Protein and Energy Nutrition Study Study. Am J Epidemiol. 2003;158:1-13. 5. Subar AF, Thompson FE, Kipnis V, Midthune D, Hurwitz P, McNutt S, McIntosh A, Rosenfeld S. Comparative validation of the Block, Willett, and National Cancer Institute food frequency questionnaires: The Eating at America’s Table Study. Am J Epidemiol. 2001;154:1089-1099. 6. National Center for Health Statistics Web site. Available at: www.cdc.gov/nchs/nhis.htm#2000_NHIS. Accessed May 2004. 7. US Department of Agriculture. What we eat in America 1994-96, Continuing Survey of Food Intakes by Individuals (CSFII) 1994-1996. [Public Use CD-Rom]. 1998. 8. Food Guide Pyramid. A Guide to Daily Food Choices.
9.
10. 11. 12. 13. 14.
15.
16. 17.
Washington, DC: US Department of Agriculture, Human Nutrition Information Service; 1992. Home and Garden Bulletin No. 252. US National Cancer Institute, Division of Cancer Control and Population Sciences [Applied Research Program Web site]. Available at: www.appliedresearch. cancer.gov. Accessed May 2004. National Research Council. Nutrient Adequacy: Assessment Using Food Consumption Surveys. Washington, DC: National Academy Press; 1986. Institute of Medicine. Dietary Reference Intakes: Applications in Dietary Assessment. Washington, DC: National Academy Press; 2000. Research Triangle Institute. SUDAAN User’s Manual, release 8.0. Research Triangle Park, NC: Research Triangle Institute; 2001. Krebs-Smith SM, Kantor LS. Choose a variety of fruits and vegetables daily: Understanding the complexities. J Nutr. 2001;131(2S-1):S487-S501. Kumanyika SK, Krebs-Smith SM. Preventive nutrition issues in ethnic and socioeconomic groups in the United States. In: Bendich A, Deckelbaum RJ, eds. Primary and Secondary Preventive Nutrition. Totowa, NJ: Humana Press, Inc; 2000. Hunt M.K, Stoddard AM, Glanz K, Hebert JR, Probart C, Sorensen G, Thomson S, Hixson ML, Linnan L, Palombo R. Measure of food choice behavior related to intervention messages in worksite health promotion. J Nutr Ed. 1997;29:3-11. Larkin FA, Basiotis PP, Riddick HA, Sykes KE, Pao EM. Dietary patterns of women smokers and nonsmokers. J Am Diet Assoc. 1990;90:230-237. Subar AF, Harlan LC, Mattson ME. Food and nutrient intake differences between smokers and non-
18. 19. 20.
21.
22.
23. 24.
25.
smokers in the US. Am J Public Health. 1990;80: 1323-1329. Berrigan D, Dodd K, Troiano R, Krebs-Smith SM, Ballard Barbash R. Patterns of health behavior in US adults. Prev Med. 2003;36:615-623. Maskarinec G, Novotny R, Tasaki K. Dietary patterns are associated with body mass index in multiethnic women. J Nutr. 2000;130:3068-3072. Eaton CB, McPhillips JB, Gans KM, Garber CE, Assaf AR, Lasater TM, Carleton RA. Cross-sectional relationship between diet and physical activity in two southeastern New England communities. Am J Prev Med. 1995;11:238-244. Gillman MW, Pinto BM, Tennstedt S, Glanz K, Marcus B, Friedman RH. Relationships of physical activity with dietary behaviors among adults. Prev Med. 2001; 32:295-301. Block G, Subar AF. Estimates of nutrient intake from a food frequency questionnaire: The 1987 National Health Interview Survey. J Am Diet Assoc. 1992;92: 969-977. Breslow RA, Subar AF, Patterson BH, Block G. Trends in food intake: The 1987 and 1992 National Health Interview Surveys. Nutr Cancer. 1997;28:86-92. Norris J, Harnack L, Carmichael S, Pouane T, Wakimoto P, Block G. US trends in nutrient intake: The 1987 and 1992 National Health Interview Surveys. Am J Public Health. 1997;87:740-746. Li R, Serdula M, Bland S, Mokdad A, Bowman B, Nelson D. Trends in fruit and vegetable consumption among adults in 16 US states: Behavioral Risk Factor Surveillance System, 1990-1996. Am J Public Health. 2000;90:777-781.
March 2005 ● Journal of the AMERICAN DIETETIC ASSOCIATION
363