A food frequency questionnaire (FFQ) (sometimes referred to as the NHANES food propensity questionnaire or “FPQ”) was used in NHANES 2003-2004 to collect information on the frequency of food consumption during the past 12 months.
Two public data release files were prepared. The FFQ Daily Frequency File (SAS name FFQDC_C) described here provides daily food frequency covariates. The second file, the FFQ Questionnaire File (SAS name: FFQRAW_C) is described separately. The FFQ data augment the other NHANES 2003-2004 dietary assessment components which include two 24-hour dietary recall interviews and interview information on dietary supplement use, food security, and dietary behavior.
FFQ Development And Testing
The NHANES FFQ questionnaire was developed by the National Institutes of Health, National Cancer Institute (NCI). The basis for the NHANES FFQ is the NCI Diet History Questionnaire (DHQ), a 124-item food frequency instrument that is widely used in nutritional epidemiology research (Subar 2001a). Unlike the DHQ, portion size information was not collected with the NHANES FFQ. The NHANES FFQ data were not intended to be used to derive estimates of absolute intake for either nutrients or foods. Diet*Calc software was developed by the National Cancer Institute (Diet*Calc 2005). Diet*Calc was used to produce daily food frequency estimates from the FFQ data.
All English or Spanish-speaking examinees 2+ years of age who completed at least one 24-hr dietary recall interview were eligible for the FFQ component.
Printed FFQ questionnaires were mailed to survey participants’ homes. Proxy respondents (usually a parent) completed the questionnaire for children less than 6 years of age and proxies assisted children 6-11 years of age and persons who could not complete the questionnaire on their own. Subjects 12 years of age and over self-reported. A postage-paid envelope was provided to respondents. Respondents who returned their FFQ form received $30 remuneration. The FFQ forms were scanned at a central office and the scanned data were added to the NHANES database.
The Diet*Calc software was used with the scanned FFQ data to produce daily frequencies for multiple foods and food groups. The FFQDC_C file contains more than 790,000 records. A majority of survey participants have more than 100 Diet*Calc records.
Quality control programs and manual verification checks were used to verify the completeness and accuracy of the scanned FFQ data files. All of the scanned FFQ records were included in the questionnaire file (FFQRAW_C). The questionnaire file documentation describes the methodology which was used to determine the completeness of the FFQ data.
None of the completed FFQ questionnaires or data were edited.
Data users should review the NHANES Analytic Guidelines carefully before analyzing the FFQ data. Additional guidance will be provided at a later date when NHANES website tutorials on NHANES dietary data analysis are available.
Two look-up tables are provided for use with the FFQ. The look-up tables (file names: FOODLOOK and VARLOOK) provide descriptive text labels for the food and variable identification number variables in this file.
A SAS program is provided on the NHANES website to show how to combine the descriptive text and frequency data.
The daily frequencies in this file are based on algorithms within Diet*Calc. These algorithms assign a daily frequency as follows:
For beverages:
Never = 0
1 time per month or less = 0.03
2-3 times per month = 0.08
1-2 times per week = 0.21
3-4 times per week = 0.5
5-6 times per week = 0.79
1 time per day = 1
2-3 times per day = 2.5
4-5 times per day = 4.5
6 or more times per day = 7
For foods:
Never= 0
1-6 time per year = 0.01
7-11 times per year = 0.028
1 time per month = 0.033
2-3 times per month = 0.08
1 time per week = 0.14
2 times per week = 0.29
3-4 times per week = 0.5
5-6 times per week = 0.79
1 time per day = 1
2 or more times per day = 2
For coffee, tea, and additions to coffee and tea:
None= 0
Less than 1 cup per month = 0.02
1-3 cups per month = 0.07
1 cup per week = 0.14
2-4 cups per week = 0.43
5-6 cups per week = 0.79
1 cup per day = 1
2-3 cups per day = 2.5
4-5 cups per day = 4.5
6 or more cups per day = 7
Proportion Question Algorithms: The proportion question(s) which accompany a particular frequency question will also affect the daily frequencies. The proportion (P) will act as a multiplier for the specific type of food mentioned in the proportion question, and the remaining proportion (1-P) is used as the multiplier for the food not specifically mentioned in the proportion question. For instance, the proportion question for question 6 would apply the answer supplied as a proportion of diet or sugar-free fruit drinks, while the remaining proportion would be applied to regular fruit drinks. The multipliers are obtained using the following algorithm:
Almost never or never = 0
About ¼ of the time = 0.25
About ½ of the time = 0.5
About ¾ of the time = 0.75
Almost always or always = 1
However, the proportion questions for questions 9, 84 use a different algorithm than the one above. For questions 9 and 84, each proportion can apply simultaneously for those foods. Using question 9 as an example, here is how the algorithm would work:
First proportion (how often soda was diet or sugar-free) = P1
Second proportion (how often soda was caffeine-free) = P2
Multiplier used for:
Diet, caffeine-free soda = P1 * P2
Diet, caffeinated soda = P1 * (1 – P2)
Regular, caffeine-free soda = (1 – P1) * P2
Regular, caffeinated soda = (1 – P1) * (1 – P2)
Seasonal Question Algorithms: for seasonal questions (9, 10, 13, 21, 23, 24, 25, 26, 35, 42, 43, 97), the following algorithm was used to assign missing values:
| Embedded questions (i.e. Q9a and Q9b) | No | Yes | Missing |
|---|---|---|---|
| a and b missing | 0 | 2 | 2 |
| a and b not missing | 0 | 0 | 0 |
| a or b missing | 1 | 1 | 1 |
Several references describing the FFQ development and validation, the NCI statistical model, and appropriate uses of FFQ information are included in the References.
Special FFQ sample weights were produced for the FFQ subsample. Analysts should use the FFQ sample weights (WTS_FFQ) for all FFQ data analyses. The FFQ sample weights were derived from the NHANES 2003-2004 examined sample weights and were designed to account for FFQ non-response. Respondents with fewer than ten missing frequency values (i.e. FFQ_MISS < 10) have an FFQ sample weight.
The NHANES Analytic Guidelines describe the recommended procedures for analyzing the NHANES data properly.
Data Uses
NCI, in collaboration with USDA/CNPP, has developed models that can be used to estimate population usual intake distributions and examine relationships between dietary intake and health outcomes. For both of these uses, the models depend primarily on data from 24-hour recalls, but additional covariates may be employed to improve the estimates. These FFQ data were collected specifically to be used to help estimate usual intakes of episodically consumed foods. The models are based on the idea that usual food intake can be expressed as the probability of consuming the food on a given day multiplied by the usual amount consumed on consumption days. While 24-hour recall data provide detailed information about the amount consumed on consumption days, they provide relatively little information about the probability to consume. Therefore the FFQ-based average daily frequency values from file FFQDC_C may provide useful covariates regarding probability to consume and hence, lead to improved predictions of usual intake. Here, “improved predictions” means that more of the variation in usual intake can be explained by observed data. Partitioning the variance of usual intake into explained and unexplained components is a key step in evaluating relationships between usual dietary intake and health outcomes using the statistical method known as regression calibration. However, when estimating a population distribution of usual intake, it is not required to partition the variance of usual intake, but rather to estimate it in total. Therefore, using FFQ-based covariates does not necessarily improve estimation for population usual intake distributions, and therefore may not be required for that purpose.
Furthermore, it is not recommended that these FFQ data be used alone to estimate absolute intakes of foods or nutrients. Such data have been shown to exhibit significant measurement error, making them an inappropriate choice in this dataset that includes 24-hour recall data. For that reason, nutrient values are not provided.