Multifactor Screener in OPEN: Scoring Procedures & Results

On this page...

• How Analytical Scoring Procedures Were Developed
• Scoring Procedures
• Validation Results

How Analytical Scoring Procedures Were Developed

Scoring procedures were developed to convert a respondent's screener responses to estimates of individual dietary intake for percentage energy from fat, grams of fiber, and servings of fruits and vegetables, using USDA's 1994-96 Continuing Survey of Food Intakes of Individuals (CSFII 94-96)dietary recall data. The following equations were estimated in the CSFII 94-96:

For percentage energy from fat and fiber:

E(Dietary Factor) = b₀+ b₁N_FG1P₁ + b₂N_FG2P₂ + … + b₁₉N_FG19P₁₉

E(Dietary Factor) indicates the expected values for percentage energy from fat and for fiber, and assumes a normal distribution. In the CSFII 94-96 dataset percentage energy from fat was normally distributed. However, fiber was positively skewed and required a cube-root transformation to approximate normality. N_FGk is the usual number of times per day an individual consumed food group k; P_k is the median portion size of group k; and k indexes the 19 food groups. These 19 food groups were formed to reflect the same food groups on the screener. We calculated weighted least-squares estimates of the regression coefficients b_k, k = 0, …, 19 on CSFII 94-96 adults aged 18 and above, stratifying by gender and excluding extreme exposure values. We first included all 19 food groups in the regression model. After examining the results, we dropped food groups that failed to attain statistical significance at the α = 0.25 level to form more parsimonious final models. In both the percentage energy from fat model and the fiber model, the lettuce food group was dropped. Because of the complex survey design, the analysis was performed using SUDAAN (RTI Inc., Research Triangle Park, NC).

For Pyramid servings of fruits and vegetables (defined by USDA in the 1992 Dietary Guidelines Food Guide Pyramid):

E([Fruits and Veg]^1/2) = b₀ + b₁ ([N_FG1P₁ + N_FG2P₂ + … + N_FG7P₇]^1/2)

Pyramid servings of fruits and vegetables was square-root-transformed to approximate normality; N_FGk is the usual number of times per day an individual consumed food group k; P_k is the median portion size of group k; and k indexes the 7 fruit and vegetable food groups. We calculated weighted least-squares estimates of the regression coefficients b₀ and b₁ on the adults in the CSFII 94-96 sample, stratifying by gender and excluding extreme exposure values.

Scoring Procedures

1. Estimation of N_FGk: Express each reported frequency as a daily average. To do this, standardize the midpoint of each frequency category to the number of times per day.

View Frequency Table

2. Estimation of P_k: The median age- and gender-specific portion sizes for each food were estimated from CSFII 94-96. For percentage energy from fat and fiber variables, the units were in grams (Tables 1-2); for fruit and vegetable servings variables, the units were in Pyramid servings (Table 3).

A Pyramid serving is defined by the U.S. Department of Agriculture in the Dietary Guidelines Food Guide Pyramid (used from 1992-2004) as:

• vegetables: 1 cup raw leafy, ½ cup of other vegetables, or ¾ cup vegetable juice; and
• fruit: 1 whole fruit, ½ cup of cut-up fruit, or ¾ cup fruit juice.

Note: current dietary guidance uses cups rather than servings.

Table 1: Median Portion Size (Pk) in Grams per Mention by Age for Percentage Energy from Fat and Fiber Analyses: Men

View Table 1

Table 2: Median Portion Size (Pk) in Grams per Mention by Age for Percentage Energy from Fat and Fiber Analyses: Women

View Table 2

Table 3a. Median Portion Size (Pk) in Pyramid Servings * per Mention by Gender and Age for Fruits and Vegetables Analyses

View Table 3a

Table 3b. Median Portion Size (Pk) in Cup Servings ** per Mention by Gender and Age for Fruits and Vegetables Analyses

View Table 3b

3. For percentage energy from fat and fiber (grams), estimation of b_k, k = 0, ..., 19: the values for each parameter, for each gender, are in the following table:

Table 4. Estimated Regression Coefficients for Foods as Predictors of Percentage Energy from Fat and Grams of Fiber, by Gender

View Table 4

4. For Pyramid servings of total fruits and vegetables, estimation of b₀ and b₁:

The model is: E([Fruits and Veg]^1/2) = b₀ + b₁ ([N_FG1P₁ + N_FG2P₂ + … + N_FG7P₇]^1/2)

For Pyramid servings of fruits and vegetables without French fries, estimation of b₀ and b₁:

The model is: E([F&V not FF]^1/2) = b₀ + b₁ ([N_FG1P₁ + N_FG2P₂ + … + N_FG7P₇]^1/2)

For Pyramid servings of fruits and vegetables, including and excluding French fries, for each gender, the estimates of the parameters are:

Table 5. Estimated Regression Coefficients for Sum of Foods Predicting Pyramid Servings of Total Fruits and Vegetables and Fruits and Vegetables Excluding French Fries, by Gender

View Table 5

For cups of fruits and vegetables (2005 MyPyramid definition), including and excluding French fries, for each gender, the estimates of the parameters are:

Table 6. Estimated Regression Coefficients for Sum of Foods Predicting Cups of Total Fruits and Vegetables and Fruits and Vegetables Excluding French Fries, by Gender

View Table 6

Download the SAS program to compute the variables

Validation Results

Staff in the Epidemiology and Genomics Research Program's Risk Factor Assessment Branch have assessed the validity of the Multifactor Screener in several studies: NCI's OPEN Study, the Eating at America's Table Study (EATS), and the joint NIH-AARP Diet and Health Study. In all studies, multiple 24-hour recalls in conjunction with a measurement error model were used to assess validity. In general, the validation results reflect the Multifactor Screener's hierarchical design -- fruit and vegetable intake was estimated best by the screener, followed by percentage energy from fat, and lastly grams of fiber.

In the OPEN Study, estimates of median intake were:

• Pyramid Servings of Fruits and Vegetables:
  • Men: recalls - 6.3; screener - 5.3;
  • Women: recalls - 5.4; screener - 4.7.
• Percentage Energy from Fat:
  • Men: recalls - 31.8; screener - 32.0;
  • Women: recalls - 32.0; screener - 30.5.
• Fiber:
  • Men: recalls - 21.2; screener - 18.3;
  • Women: recalls - 16.5; screener - 14.1.

These validation results suggest that dietary exposure estimates computed from the Multifactor Screener may be useful to compare subgroup means, especially for populations consuming mainstream diets. The estimates may be less useful for populations with more ethnic diets, including Asian and possibly Latino populations.

At the individual level, correlations between the screener and estimated true intake ranged from 0.54 (fiber for men) to 0.76 (Pyramid servings of fruits and vegetables for women); about 25 to 50 percent of the variability in the true intake will be captured by the screener questions. Thus, although significant error may be associated with these estimates of diet, we believe the exposure estimates still substantially reflect what individuals are actually consuming.

Validation results are reported in detail in:

• 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, percentage energy from fat and fibre. Public Health Nutr. 2004 Dec;7(8):1097-105.