Users of the 2005-2006 Dual-Energy X-ray Absorptiometry data (variable name prefix DXX_D) are strongly encouraged to read the documentation before accessing the data file.
Because missing or invalid data have been multiply imputed, the DXX_D data release file contains 5 records for each survey participant, 8-69 years of age, who was interviewed and examined. Only 1 record should be used in calculating sample sizes. However, all 5 records must be used in analyses in order to obtain more accurate variance estimates. The records for some survey participants, such as pregnant females, are blank; pregnant females were not eligible for the DXA scan.
Dual-energy x-ray absorptiometry (DXA) has become one of the most widely accepted methods of measuring body composition due in part to its speed, ease of use, and low radiation exposure (Genant, 1996, Njeh, 1999, Heymsfield, 1989, Tothill, 1996). Whole body DXA scans were administered in the NHANES mobile examination center (MEC) in 1999-2006. The NHANES DXA examination provides nationally representative data on body composition (bone and soft tissue), overall and for age, gender, and racial/ethnic groups, to study the association between body composition and other health conditions and risk factors, such as cardiovascular disease, diabetes, hypertension, and physical activity and dietary intake patterns.
The DXA scans provide bone and soft tissue measurements for the total body, for both arms and both legs, the trunk, and the head. Bone measurements also were obtained for the pelvis, left and right ribs, thoracic spine, and lumbar spine. Values for the total body and regions include:
- Total mass (gm)
- Bone mineral content (BMC) (gm)
- Bone area (cm2)
- Bone mineral density (BMD) (gm/cm2)
- Fat mass (gm)
- Lean mass excluding BMC (gm)
- Lean mass including BMC (gm)
- Percent body fat (%)
DXA scans were administered to eligible survey participants 8-69 years of age. Pregnant females were ineligible for the DXA examination. Participants who were excluded from the DXA examination for reasons other than pregnancy were considered to be eligible nonrespondents. Reasons for exclusion from the DXA examination were as follows:
- Pregnancy (positive urine pregnancy test and/or self-report at the time of the DXA examination). Females between the ages of 12–59 years and menstruating 8–11 year olds were not permitted to take the DXA examination without a negative MEC pregnancy test result. In addition, females aged 12–59 years were excluded from the examination if they said they were pregnant at the time of the exam, even if the pregnancy test was negative.
- Self-reported history of radiographic contrast material (barium) use in past 7 days.
- Self-reported weight over 300 pounds or height over 6’5” (DXA table limitations).
The variable DXAEXSTS indicates examination status. Equipment failure was the main reason for a completed, but invalid scan. The “Not scanned, other reason” code includes no time to complete the examination, pregnancy test not completed, and participant refusal, as well as exclusion for reasons other than pregnancy.
DXAEXSTS – examination status variable
1 = Scan completed
2 = Scan completed, but invalid
3 = Not scanned, pregnant
4 = Not scanned, weight > 300 lbs
5 = Not scanned, height > 6'5"
6 = Not scanned, other reason
Protocol and Procedure
Whole body DXA scans were taken with a Hologic QDR-4500A fan-beam densitometer (Hologic, Inc., Bedford, Massachusetts). Hologic software version 8.26:a3* was used to administer all scans through mid-2005. In 2005, the acquisition software was updated to Hologic Discovery v12.4. The densitometer scanned participants with an x-ray source using fan-beam scan geometry in three passes (1 minute per pass). The participants were positioned supine on the tabletop with their feet in a neutral position and hands flat by their side. A Velcro strap was used to keep the feet stationary and together. The DXA technique acquires two low-dose x-ray images at different average energies. The ratio of the attenuation of these two average energies, called an R-factor, is used to distinguish both bone from soft tissue, and the percent fat in soft tissue when bone isn’t present. The radiation exposure from DXA whole body scans is extremely low at less than 10 uSv.
The DXA examinations were administered by certified radiology technologists. Further details of the DXA examination protocol are documented in the Body Composition Procedures Manual located on the NHANES website.
Quality Assurance & Quality Control
A high level of quality control was maintained throughout the DXA data collection and scan analysis, including a rigorous phantom scanning schedule.
Monitoring of Field Staff and Densitometers
Staff from the National Center for Health Statistics (NCHS) and the NHANES data collection contractor monitored technologist acquisition performance through in-person observations in the field. Retraining sessions were conducted with the technologists annually and as needed to reinforce correct techniques and appropriate protocol. In addition, technologist performance codes were recorded by the NHANES quality control center at the University of California, San Francisco (UCSF), Department of Radiology as part of the participants’ scan review. The codes documented when the technologist had deviated from acquisition procedures and scan quality could have been improved. The performance codes were tracked for each technologist individually and a summary reported to NCHS on a quarterly basis. Constant communication was maintained throughout the year among the UCSF, the NCHS, and the data collection contractor regarding any issues that arose.
Hologic service engineers performed all routine densitometer maintenance and repairs. Copies of all reports completed by the manufacturer’s service engineers were sent to the UCSF when the scanners were serviced or repaired so any changes in measurement as a result of the work could be assessed. While some minor mechanical repairs were made during 2005-2006 survey operations, replacement or realignment of the detectors, apertures, or other major hardware was not required for any of the three densitometers.
Each participant and phantom scan was reviewed and analyzed by the UCSF using standard radiologic techniques and study-specific protocols developed for the NHANES. Hologic Discovery software, version 12.4, was used to analyze the scans. The Discovery analysis software incorporates the Auto WB application, which was developed to improve bone detection in children participating in the NHANES and other studies of children (Kelly, 2002, Fan, 2004). The Discovery analysis algorithms automatically detect and measure very low-density bone in children weighing 40 kg or less.
Expert review was conducted by the UCSF on 100% of analyzed participant scans to verify the accuracy and consistency of the results.
Invalidity codes were applied by the UCSF to indicate the reasons regions of the body could not be analyzed accurately. The invalidity codes are provided in the data file (see Analytic Notes for a description of the invalidity codes).
Quality Control Scans
The quality control phantoms were scanned according to a predetermined schedule. The Hologic Anthropomorphic Spine Phantom associated with each MEC was scanned daily as required by the manufacturer to ensure accurate calibration of the densitometer. Other MEC-specific phantoms, such as the Hologic Whole Body Slim-line Phantom and Hologic Tissue Step Phantom, were scanned 1 to 3 times weekly. Another set of phantoms, the Hologic Spine (HSP-Q96), Hologic Block, and Hologic Whole Body Phantoms, circulated among the MECs and were scanned at the start of operations at each survey site.
Air scans, phantom-less scans using the whole body scan mode, were used to describe and monitor the systems’ radiographic uniformity across the entire scan field. Poor uniformity could be caused by poor aperture alignment, incorrect gantry rotation, non-uniform gain in detectors, etc., that result in localized inaccuracies in the attenuation values.
The complete phantom scanning schedule is described in the Body Composition Procedures Manual located on the NHANES website.
Cross-calibration and Longitudinal monitoring
In multi-site studies such as the NHANES, verification that all DXA systems are performing within the expected limits is critical since data collected at the multiple sites are pooled for analysis. A cross-calibration study was conducted prior to the start of NHANES 1999 to identify the relationships among the densitometers in the three MECS. Since all three densitometers in the NHANES were the identical make and model, cross-calibration was simplified. However, in 1999, no standard existed for phantom cross-calibration for whole body BMD and soft tissue and new procedures were developed for the survey. At the time, the NHANES cross-calibration study was unique in that it included three scanners and in-vivo subjects and in-vitro phantoms.
In 2005-2006, longitudinal monitoring was conducted through the daily spine phantom scans as required by the manufacturer, 3 times weekly whole body slim-line phantom scans, and weekly air scans in order to correct any scanner-related changes in participant data. The circulating HSP-Q96, block, and whole body phantoms, which were scanned at the start of operations at each site, provided additional data for use in longitudinal monitoring and cross calibration. The cross-comparability of the data from each MEC was critical so the data could be pooled for analysis.
The UCSF used the Cumulative Statistics method (CUSUM) and the MEC-specific phantom data to determine breaks in the calibration of the densitometers over the course of the survey (Lu, 1996). Multiplicative correction factors were used to correct the phantom data back to the baseline calibration. The type, frequency, and magnitude of calibration problems detected in the NHANES data were similar to those in other studies using stationary densitometers that were being monitored by UCSF.
After applying the correction factors developed by UCSF from the cross-calibration and longitudinal phantom data to the NHANES participant data, the adjusted participant data were compared to unadjusted data. The magnitude of the changes and reduction in standard errors between the adjusted and unadjusted data were found to be small and correction of the participant data not required.
A number of issues were addressed through the quality control program. Direct feedback given to the technologists regarding acquisition problems affecting the quality of the scans and yearly refresher training resulted in improved technologist performance. The rigorous schedule of quality control scans provided continuous monitoring of machine performance. The expert review procedures assured that scan analysis was accurate and consistent. The air scan quality assurance tool used to evaluate whole body performance was first used in the NHANES and was subsequently adopted by Hologic as a mandatory scan mode for all whole body scanners.
Data Processing and Editing
Several steps were taken to produce the DXX_D data files.
5% Adjustment of Lean Mass and Fat Mass
The NHANES lean soft tissue mass and fat mass for the total body and regions were adjusted based on the results of an analysis of QDR- 4500A DXA data from seven research laboratories indicating that the QDR-4500A algorithm underestimated fat mass and overestimated lean mass (Schoeller, 2005). The analysis utilized six data sets provided by study investigators and one published data set. The analytic data included fat mass and lean mass measured on Hologic QDR-4500A densitometers and criteria measurements of body composition from total body water by dilution, underwater weighing, and four-compartment analysis. The QDR-4500A was determined to overestimate lean mass (p < 0.05) in the cohort of 1,198 subjects. A statistically significant difference was observed in all 7 data sets with a mean ± SE of 5 ± 1%. Based on the results of the analysis, the NHANES DXA lean mass was decreased by 5% and an equivalent kilogram weight added to the fat mass so the total mass did not change.
The percentage of eligible survey participants in 2005-2006 with 100% valid data (all analyzed regions were valid) is shown by age group in Table 1. The percentage of participants with valid data decreases with increasing age. The decrease in valid data with age was due primarily to an increase in the number of participants with implants such as pacemakers, stents, and hip replacements and higher rates of obesity resulting in invalid truncal data from “obesity noise.” The percentage of participants with 100% valid data also decreases with increasing BMI (Table 2).
Because valid data decreased with increasing age and increasing BMI and because individuals with body weight greater than 300 pounds were not scanned (exclusion criterion for the DXA examination), invalid and missing data could not be treated as a random subset of the data file. To resolve the problem of bias due to non-random invalid and missing data, multiple imputation of the DXX_D data was performed. With the exception of pregnant women (who were ineligible for the DXA exam) and participants with amputations other than fingers or toes, all participants with invalid or missing data were included in the multiple imputation process.
SAS-callable imputation and variance estimation software developed by the Survey Methodology Program at the University of Michigan’s Institute of Survey Research (ISR), IVEware, was used to impute the NHANES DXA data (Raghunathan, 2002). The IVEware module IMPUTE performs multiple imputations of missing values using the sequential regression imputation method (Raghunathan, 2001). A detailed description of the imputation procedures is provided in the Documentation for Multiple Imputation of National Health and Nutrition Examination Survey 1999-2004 Dual Energy X-Ray Absorptiometry Data on the NHANES and in Schenker, 2011.
Five complete records containing valid and/or imputed values were created for each survey participant to allow the assessment of variability due to imputation. The DXX data file contains all 5 records. The variable “_multi_“ has values 1-5 which can be used to identify the records. For participants with multiply imputed data, each of the 5 records contains a different set of imputed values. Participants who have 100% valid data have 5 identical records, since no data were imputed.
Use of the imputed data sets will provide complete DXA data for all participants and ensure a more accurate standard error of the estimate.
Imputation Indicator Variables
The data file contains imputation indicator variables as listed below; the values for each variable are 0 = data not imputed, 1 = data imputed, and 2 = highly variable imputed data:
DXITOT = overall indicator; 1 or more regions were imputed
DXIHE = head
DXILA = left arm
DXILL = left leg
DXIRA = right arm
DXIRL = right leg
DXILR = left rib
DXIRR = right rib
DXITS = thoracic spine
DXILS = lumbar spine,
DXIPE = pelvis
DXITR = trunk
A subset of participants with highly variable imputed data, fat mass in particular, has blank records in the 2005-2006 DXX file. The data for these participants can be found in the DXX_D_S data file. Participants with highly variable imputed data (all imputation indicator variables = 2) had no valid DXA data and were missing measured weight and waist circumference, which were critical predictor variables in the imputation model. The data in DXX_D_ S should be reviewed carefully before inclusion in any analysis.
The DXX_D data file contains 5 records for each survey participant.
The multiple records must be taken into account when calculating sample sizes. The following SAS example can be used to select a single record in order to calculate sample sizes:
merge dexa.dxx_d (where =(_mult_ = 1)) work.demo;
The frequency counts in the codebook are the total number of observations from all 5 records. The counts must be divided by 5 to calculate the actual number of participants with the code or value. Frequency counts are not provided for the DXX_D_S data file.
Analysts should read the Documentation for Multiple Imputation of National Health and Nutrition Examination Survey 1999-2004 Dual Energy X-Ray Absorptiometry Data on the NHANES website. The same model and procedures used in multiply imputing the 1999-2004 DXA data were used in imputing the 2005-2006 data. Additional information on the multiple imputation of the 1999-2004 DXA data can be found in Schenker, 2011. The documentation provides sample code for analysis of the multiply imputed data using SAS-callable SUDAAN. Other statistical packages, including Stata, R and SAS Survey, can be used in the analysis of multiply imputed complex survey data.
The NHANES examination sample weights should be used for all DXX_D analyses. Please refer to the NHANES Analytic Guidelines and the on-line NHANES Tutorial for further details on the use of sample weights and other analytic issues.
Relationship among examination status codes and imputation indicator codes
||Other Imputation Indicator Codes
||All data were valid and none were imputed.
||All codes = 0.
||Data for at least 1 region(s) were invalid and imputed.
||Code(s) for the imputed region(s) = 1.
||All data were invalid and all were imputed.
||All codes = 1.
||Participant was pregnant and excluded from the DXA exam. All data are missing and none were imputed. There are 371pregnant females in the DXX_D data file.
|4, 5, or 6
||Participant was excluded from the exam for a reason other than pregnancy. All data were imputed.
||All codes = 1.
|4 or 6
||The participant was excluded from the exam. All data were imputed, but were considered to be highly variable and placed in DXX_D_S. There are 47 participants with highly variable data in the DXX_D_S file.
||All codes = 2.
||The participant was excluded from the exam, but the data could not be imputed for reasons such as amputation. All data are missing. There are 6 such participants in the DXX_D data file.
Invalidity codes were applicable to completed scans only (DXAEXSTS=1). Valid regions were coded 0. Codes 1-7 indicate the reasons regions could not be analyzed accurately. If a participant was not scanned, all invalidity codes will be missing.
DXAHEBV = head bone
DXAHETV = head tissue
DXALABV = left arm bone
DXALATV = left arm tissue
DXALLBV = left leg bone
DXALLTV = left leg tissue
DXARABV = right arm bone
DXARATV = right arm tissue
DXARLBV = right leg bone
DXARLTV = right leg tissue
DXATRBV = trunk bone, includes thoracic and lumbar spine, left and right ribs, and pelvis
DXATRTV = trunk tissue
Values for invalidity codes
0 = Valid data
1 = Jewelry and other objects not removed
2 = Non-removable objects (includes prostheses, implants, casts)
3 = Excessive x-ray "noise" due to obesity, i.e., the DXA beam could not penetrate the layers of abdominal fat to provide an analyzable scan image (applied to the trunk region only)
4 = Arm/leg overlap
5 = Body parts out of scan region
6 = Positioning problem (head, arms/hands or feet turned)
7 = Other (includes participant motion, unknown artifacts, deformities)
Table 1. Percentages of interviewed and examined participants 8-69 years of age with valid DXA data by age group, NHANES 2005-2006.
|Gender-age group (Years)
||Interviewed and examined (N) *
||Eligible for DXA (N) †
||Eligible for DXA (%) †
||100% valid DXA data (N) ‡
||100% valid DXA data (%) ‡
* The number interviewed and examined is the total number of participants in the data file with a SEQN variable. This number includes pregnant females (n = 344).
† The total number eligible for DXA includes participants with both valid and imputed data (n = 7,621), participants with highly variable data in DXX_D_S (n = 252), and participants for whom data could not be imputed (n = 25). This number does not include pregnant females.
‡ Of those eligible for DXA who successfully completed a scan.
Table 2. Percentages of participants 20 years and older with valid DXA data by body mass index (BMI)* category, NHANES 2005-2006.
||Eligible for DXA †
||100% Valid Data (N) †
||100% Valid Data (%) ‡
| < 18
* Measured weight in kilograms divided by measured height in meters squared.
† Does not include pregnant females
‡ Of those eligible for DXA.