Abstract

Dual X-ray absorptiometry (DXA) has a range of clinical applications, from assessing associations between adipose or lean body mass and the risk of disease to measuring the effects of dietary interventions on adipose deposition and oxidation and/or muscle accumulation. Many lifestyle-related studies, however, are short- to medium-term interventions, and inter- or intradevice variation between DXA scanners can facilitate type I and type II errors during data analysis. Studies demonstrate that variation in body composition measurements exist not only between DXA instruments using fan-beam and pencil-beam technologies but also between DXA instruments produced by different manufacturers. Moreover, studies show inter- and intrainstrument variation between identical DXA instruments. Such inter- and intrascan variability between instruments can be compounded by the particular patient population being investigated. The objective of this review is to discuss inter- and intradevice variation of DXA instruments and to outline quality control procedures that should be implemented prior to initiating short-term single or multicenter clinical trials that use DXA to investigate the effects of an intervention on loss or accretion of lean or fat mass.

Introduction

Originally developed for the measurement of bone mineral density, dual X-ray absorptiometry (DXA) has become recognized for its ability to accurately and precisely measure total body composition. In addition, the ongoing evolution of DXA technology has led to the development of fan-beam technology, which, compared with older pencil-beam scanners, has dramatically reduced the time required to assess total mass, lean body mass, fat mass, and percent body fat to less than 10 minutes. Moreover, given the relatively low cost of ownership of DXA technology compared with that of other advanced technologies, such as magnetic resonance imaging (MRI) and computed tomography (CT), the use of DXA has transcended across various clinical disciplines, from assessing associations between adipose and/or lean body mass with the risk of disease, to measuring the effects of dietary interventions on adipose deposition/oxidation and/or muscle accumulation.

Weight-maintenance strategies have been shown to be more effective when diet and exercise paradigms persist over the long term.1 Similarly, large weight-loss interventions have been shown to be most effective when diet periods are at least 4 to 24 months in duration.2,5 In addition, increasing the duration of a dietary intervention increases the likelihood of observing changes in body composition. However, given that many lifestyle-related interventions, including nutrition and exercise, are less than 4 months in duration, inter- and intradevice variation between DXA scanners could facilitate type I and type II errors when they are used to assess changes in body composition. Studies demonstrate that variation in body composition measurements exists not only between DXA instruments using fan- and pencil-bean technologies but also between DXA instruments using similar technologies but produced by different manufacturers. In addition, studies show inter- and intradevice instrument variation between identical DXA instruments. Thus, proper quality-control procedures must be implanted to reduce the likelihood of erroneous DXA-derived body composition data and to ensure that changes observed in body composition are secondary to the imposed intervention. In addition, the inter- and intrascan variability between instruments can be compounded by the particular patient population being investigated.

The objective of this review is to discuss inter- and intravariation of DXA-derived body composition measurements obtained with fan-beam DXA technology. This review will outline quality-control procedures that should be implemented prior to initiating short-term single or multicenter clinical trials that use DXA as a tool for body composition analysis to investigate the effects of an intervention on lean/fat mass loss or accretion.

Interdevice Measurement Error with DXA Devices

Typically, DXA instruments are produced by either GE Healthcare (formerly GE Medical Systems), a branch of General Electric (GE) (Wauwatosa, Wisc., USA), or Hologic (Bedford, Mass., USA). Both GE Healthcare and Hologic manufacture a range of DXA scanners with varying capabilities, precision, and accuracy. GE Healthcare manufactures the frequently used GE Lunar machines, formerly manufactured by Lunar Corporation (Madison, Wisc., USA).

Precision and accuracy become especially problematic for large multicenter clinical trials in which body composition is to be measured by different models of DXA instruments. Therefore, various studies have focused on evaluating measurement error between DXA instruments produced by different manufacturers.

As with any technology, DXA scanners continue to evolve. Table 1 provides examples of studies that have evaluated interdevice variation in total body mass, lean mass, fat mass, and percent body fat between DXA models manufactured by GE Healthcare and Hologic. Overall, the aforementioned body composition endpoints are highly correlated between different DXA models manufactured by different and corresponding manufacturers.6,11 Nonetheless, evaluation of absolute measures of lean body mass and fat mass reveals considerable variation between DXA models. For example, two fan-beam DXA scanners, the Hologic QDR4500W and the Hologic Discovery Wi, revealed net differences in absolute measures of lean body mass and fat mass of 320–450 g and 500–690 g, respectively.7 When a 28-kg total body phantom was scanned on seven Hologic QDR 4500 DXA scanners across seven centers, coefficients of variation of 3.9%, 2.5%, 2.0%, and 5.6% were observed for total area, total mass, lean mass, and fat mass, respectively.10 Similarly, comparative body composition scans conducted on a Lunar Prodigy and a Hologic Delphi W revealed significant differences between measures of lean mass, fat mass, and percent body fat,6 while Ioannidou et al.8 demonstrated that the Lunar Prodigy (fan-beam technology) underestimated appendicular lean soft tissue compared with pencil-beam scanners such as the Lunar DPX and Lunar DPX-L. Regression analysis between the Lunar Prodigy and the Lunar DPX showed standard error of estimates of 298 g, 911 g, and 959 g for total mass, lean mass, and body fat, respectively.9 The implications of observing standard error of estimates of 298 g, 911 g, or 959 g for DXA measurements may not be outwardly apparent. However, for nutrition or exercise interventions that are short in duration or utilize a bioactive compound that facilitates subtle effects on body composition, expected fat mass loss is often equal to 1 kg or less, which falls within the interdevice error range. This could mask the effect of an imposed treatment.

Table 1

Variation in total body mass, lean mass, fat mass, and percent body fat between DXA models manufactured by GE Healthcare (formerly Lunar Corporation) and Hologic

Reference Devices compared or evaluated Subjects Intrainstrument variation Interinstrument variation 
Aasen et al. (2006)6 Lunar Prodigy (fan-beam angle, 30°)
Hologic Delphi W (fan-beam angle, 30°)
Lunar Expert (fan-beam angle, 12°) 
8 men, 13 women
Age: 30–84 yrs (61.5 ± 12.5 yrs)
BMI: 21.8–39.8 kg/m2 (28.8 ± 5.0 kg/m2)
Scanned on each DXA device within 60 min
WBP: 18 scans 
WBP CV (Prodigy, Delphi W, Expert, respectively)
Weight: 0.02%, 0.9%, 0.07%
FM: 0.7%, 1.3%, 1.3%
LM: 0.7%, 5.5%, 2.0% 
Humans: no significant difference in body composition parameters; significant correlations between DXAs (see text).
Phantom: no difference in BM, LM, FM, or %FM between the Lunar Prodigy and Lunar Expert; LM, FM, and %FM differed (P < 0.005) between Prodigy and Delphi W 
Covey et al. (2010)7 Hologic QDR4500W (fan-beam)
Hologic Discovery Wi (fan-beam) 
39 women
Age: 25.7–68.5 (50.6 ±  9.6 yrs)
BMI: 19.3–41.4 yrs (26.8 ± 5.5 kg/m2)
%FM: 16–45% (33 ± 7%)
Scanned 2× on each instrument over 5 wks 
Intradevice net difference (QDR4500W, Discovery Wi, respectively)
Truncal region:
Nonosseous LM: 0.06 ± 0.62 kg, 0.01 ± 0.42 kg
FM: −0.01 ± 0.34 kg, 0.01 ± 0.42 kg
Appendicular region:
Nonosseous LM: −0.01 kg ± 0.42, −0.01 ± 0.39 kg
FM: 0.05 ± 0.36 kg, 0.02 ± 0.19 kg 
Net difference (kg) (QDR 4500W – Discovery Wi)
Truncal region:
Nonosseous LM: 0.32 ± 0.82 kg
FM: −0.69 ± 0.60 kg
Appendicular region:
Nonosseous LM: 0.45 ± 0.39 kg
FM: −0.50 ± 0.53 kg 
Ioannidou et al. (2003)8 Lunar DPX (pencil-beam)
Lunar DPX-L (pencil-beam)
Lunar Prodigy (fan-beam)
Hologic Delphi A (fan-beam) 
11 men, 24 women
Healthy
Comparison of ALST 
N/A Measurements across all instruments were highly correlated.
Lunar Prodigy underestimated ALST compared with the Lunar DPX and the Lunar DPX-L 
Nord et al. (2000)9 Lunar Prodigy (fan-beam)
Lunar DPX-IQ (pencil-beam) 
46 men and women scanned 3× on the Prodigy and 1× on the DPX-IQ
20 men and women scanned 1× on each device 
Prodigy CV:
Total mass: 0.3%
LM: 0.9%
FM: 1.3% 
R2 (SEE),
Prodigy vs DPX:
Total mass: 0.999 (298 g)
LM: 0.989 (911 g)
FM: 0.991(959 g) 
Louis et al. (2006)10 Hologic QDR 4500 DXA (fan-beam) 28-kg modular whole-body phantom
Comparison across 7 Hologic QDR 4500 DXA scanners
Phantom scanned 18× over 5 months as a measure of intradevice variation 
CV:
Total area: 4.7%
Total mass: 0.3%
LM: 0.9%
FM: 1.2% 
CV:
Total area: 3.9%
Total mass: 2.5%
LM: 2.0%
FM: 5.6%
Variability is a concern when using the same device between centers 
Oldroyd et al. (2003)11 Lunar Prodigy (fan-beam)
Lunar DPX-L (pencil-beam) 
Lunar VCP
LTBP
Phantoms were scanned 10× on each device and on the same day.
In vivo
28 men, 44 women
Age: 16.1–65 yrs (16.1 ± 65 yrs)
BMI: 12–34.7 kg/m2 (21.6 ± 4.4 kg/m2
VCP %FM (CV) (Prodigy, DPX-L, respectively):
At 44.4% FM VCP: 44.4 (1.8%), 45.2 (1.7%)
At 36.8% FM VCP: 37.9 (1.2%), 38.4 (1.3%)
At 28.1% FM VCP: 30.1 (2.2%), 31.3 (1.8%)
At 21.2% FM VCP: 23.7 (1.2%), 24.6 (1.7%)
At 16.0% FM VCP: 16.5 (2.9%), 17.1 (1.7%)
LTBP (CV) (Prodigy, DPX-L, respectively):
TBtissue: 70.74 kg (0.2%), 69.23 kg (0.2%)
LM: 50.92 kg (0.7%), 50.33 kg (0.8%)
FM: 19.83 kg (1.7%), 18.90 kg (2.7%)
%FM: 27.3% (1.7%), 26.5% (2.3%) 
VCP:
VCP at %FM of 28.1%, 21.1%, and 16.0% was significantly different between the Prodigy and the DPXL
LTBP:
TBtissue, FM, and %FM were significantly different between the Prodigy and the DPX-L
In vivo:
Prodigy scans produced higher measures for TBtissue, BM, LM, FM, and %FM 
Guo et al. (2004)14 Lunar DPX (pencil-beam)
Lunar Prodigy (fan-beam) 
Phantoms scanned 10× at each scan speed
10% FM phantom
50% FM phantom
Scan speed: slow, med, fast
  1. Lunar DPX vs Lunar Prodigy

  2. Lunar Prodigy A vs Lunar Prodigy B

 
CV (DPX, Prodigy A, Prodigy B, respectively)
10% FM Phantom:
Slow: 2.13%, 3.49%, 1.79%
Med: 5.18%, 4.49%, 3.49%
Fast: 3.57%, 1.57%, 2.79%
50% FM Phantom:
Slow: 0.46%, 0.39%, 0.41%
Med: 0.57%, 0.58%, 0.75%
Fast: 0.69%, 0.42%, 0.88% 
10% FM Phantom (DPX, Prodigy A, Prodigy B, respectively)
LM:
Slow: 4,888 g, 5,047 g, 4,984 g
Med: 4,846 g, 5,122 g, 4,995 g
Fast: 4,820 g, 5,190 g, 5,116 g
FM:
Slow: 532 g, 587 g, 615 g
Med: 534 g, 596 g, 615 g
Fast: 572 g, 605 g, 654 g
%FM:
Slow: 9.8%, 10.4%, 11.0%
Med: 9.9%, 10.4%, 11.0%
Fast: 10.6%, 10.4%, 11.3%
50% FM Phantom (DPX, Prodigy A, Prodigy B, respectively):
LM:
Slow: 2,690 g, 2,899 g, 2,815 g
Med: 2,717 g, 2,963 g, 2,859 g
Fast: 2,713 g, 3,025 g, 2,861 g
FM:
Slow: 2,827 g, 2,868 g, 2,842 g
Med: 2,851 g, 2,923 g, 2,842 g
Fast: 2,841 g, 3,036 g, 2,977 g
%FM:
Slow: 51.3%, 49.7%, 50.2%
Med: 51.2%, 49.7%, 49.9%
Fast: 51.2%, 50.1%, 51.0% 
Hind et al. (2011)15 Lunar iDXA 18 men, 34 women
Low BMI (<18.5 g/m2) n = 2; normal BMI (18.5–24.9 kg/m2), n = 25; overweight (25–29.9 kg/m2) n = 13; obese (30–34.9 kg/m2) n = 10; and severely obese (=35.0 kg/m2), n = 2 
CV:
FM: 0.82
%FM: 0.86
%AF: 2.32
%GF: 0.96
LM: 0.51 
N/A 
Reference Devices compared or evaluated Subjects Intrainstrument variation Interinstrument variation 
Aasen et al. (2006)6 Lunar Prodigy (fan-beam angle, 30°)
Hologic Delphi W (fan-beam angle, 30°)
Lunar Expert (fan-beam angle, 12°) 
8 men, 13 women
Age: 30–84 yrs (61.5 ± 12.5 yrs)
BMI: 21.8–39.8 kg/m2 (28.8 ± 5.0 kg/m2)
Scanned on each DXA device within 60 min
WBP: 18 scans 
WBP CV (Prodigy, Delphi W, Expert, respectively)
Weight: 0.02%, 0.9%, 0.07%
FM: 0.7%, 1.3%, 1.3%
LM: 0.7%, 5.5%, 2.0% 
Humans: no significant difference in body composition parameters; significant correlations between DXAs (see text).
Phantom: no difference in BM, LM, FM, or %FM between the Lunar Prodigy and Lunar Expert; LM, FM, and %FM differed (P < 0.005) between Prodigy and Delphi W 
Covey et al. (2010)7 Hologic QDR4500W (fan-beam)
Hologic Discovery Wi (fan-beam) 
39 women
Age: 25.7–68.5 (50.6 ±  9.6 yrs)
BMI: 19.3–41.4 yrs (26.8 ± 5.5 kg/m2)
%FM: 16–45% (33 ± 7%)
Scanned 2× on each instrument over 5 wks 
Intradevice net difference (QDR4500W, Discovery Wi, respectively)
Truncal region:
Nonosseous LM: 0.06 ± 0.62 kg, 0.01 ± 0.42 kg
FM: −0.01 ± 0.34 kg, 0.01 ± 0.42 kg
Appendicular region:
Nonosseous LM: −0.01 kg ± 0.42, −0.01 ± 0.39 kg
FM: 0.05 ± 0.36 kg, 0.02 ± 0.19 kg 
Net difference (kg) (QDR 4500W – Discovery Wi)
Truncal region:
Nonosseous LM: 0.32 ± 0.82 kg
FM: −0.69 ± 0.60 kg
Appendicular region:
Nonosseous LM: 0.45 ± 0.39 kg
FM: −0.50 ± 0.53 kg 
Ioannidou et al. (2003)8 Lunar DPX (pencil-beam)
Lunar DPX-L (pencil-beam)
Lunar Prodigy (fan-beam)
Hologic Delphi A (fan-beam) 
11 men, 24 women
Healthy
Comparison of ALST 
N/A Measurements across all instruments were highly correlated.
Lunar Prodigy underestimated ALST compared with the Lunar DPX and the Lunar DPX-L 
Nord et al. (2000)9 Lunar Prodigy (fan-beam)
Lunar DPX-IQ (pencil-beam) 
46 men and women scanned 3× on the Prodigy and 1× on the DPX-IQ
20 men and women scanned 1× on each device 
Prodigy CV:
Total mass: 0.3%
LM: 0.9%
FM: 1.3% 
R2 (SEE),
Prodigy vs DPX:
Total mass: 0.999 (298 g)
LM: 0.989 (911 g)
FM: 0.991(959 g) 
Louis et al. (2006)10 Hologic QDR 4500 DXA (fan-beam) 28-kg modular whole-body phantom
Comparison across 7 Hologic QDR 4500 DXA scanners
Phantom scanned 18× over 5 months as a measure of intradevice variation 
CV:
Total area: 4.7%
Total mass: 0.3%
LM: 0.9%
FM: 1.2% 
CV:
Total area: 3.9%
Total mass: 2.5%
LM: 2.0%
FM: 5.6%
Variability is a concern when using the same device between centers 
Oldroyd et al. (2003)11 Lunar Prodigy (fan-beam)
Lunar DPX-L (pencil-beam) 
Lunar VCP
LTBP
Phantoms were scanned 10× on each device and on the same day.
In vivo
28 men, 44 women
Age: 16.1–65 yrs (16.1 ± 65 yrs)
BMI: 12–34.7 kg/m2 (21.6 ± 4.4 kg/m2
VCP %FM (CV) (Prodigy, DPX-L, respectively):
At 44.4% FM VCP: 44.4 (1.8%), 45.2 (1.7%)
At 36.8% FM VCP: 37.9 (1.2%), 38.4 (1.3%)
At 28.1% FM VCP: 30.1 (2.2%), 31.3 (1.8%)
At 21.2% FM VCP: 23.7 (1.2%), 24.6 (1.7%)
At 16.0% FM VCP: 16.5 (2.9%), 17.1 (1.7%)
LTBP (CV) (Prodigy, DPX-L, respectively):
TBtissue: 70.74 kg (0.2%), 69.23 kg (0.2%)
LM: 50.92 kg (0.7%), 50.33 kg (0.8%)
FM: 19.83 kg (1.7%), 18.90 kg (2.7%)
%FM: 27.3% (1.7%), 26.5% (2.3%) 
VCP:
VCP at %FM of 28.1%, 21.1%, and 16.0% was significantly different between the Prodigy and the DPXL
LTBP:
TBtissue, FM, and %FM were significantly different between the Prodigy and the DPX-L
In vivo:
Prodigy scans produced higher measures for TBtissue, BM, LM, FM, and %FM 
Guo et al. (2004)14 Lunar DPX (pencil-beam)
Lunar Prodigy (fan-beam) 
Phantoms scanned 10× at each scan speed
10% FM phantom
50% FM phantom
Scan speed: slow, med, fast
  1. Lunar DPX vs Lunar Prodigy

  2. Lunar Prodigy A vs Lunar Prodigy B

 
CV (DPX, Prodigy A, Prodigy B, respectively)
10% FM Phantom:
Slow: 2.13%, 3.49%, 1.79%
Med: 5.18%, 4.49%, 3.49%
Fast: 3.57%, 1.57%, 2.79%
50% FM Phantom:
Slow: 0.46%, 0.39%, 0.41%
Med: 0.57%, 0.58%, 0.75%
Fast: 0.69%, 0.42%, 0.88% 
10% FM Phantom (DPX, Prodigy A, Prodigy B, respectively)
LM:
Slow: 4,888 g, 5,047 g, 4,984 g
Med: 4,846 g, 5,122 g, 4,995 g
Fast: 4,820 g, 5,190 g, 5,116 g
FM:
Slow: 532 g, 587 g, 615 g
Med: 534 g, 596 g, 615 g
Fast: 572 g, 605 g, 654 g
%FM:
Slow: 9.8%, 10.4%, 11.0%
Med: 9.9%, 10.4%, 11.0%
Fast: 10.6%, 10.4%, 11.3%
50% FM Phantom (DPX, Prodigy A, Prodigy B, respectively):
LM:
Slow: 2,690 g, 2,899 g, 2,815 g
Med: 2,717 g, 2,963 g, 2,859 g
Fast: 2,713 g, 3,025 g, 2,861 g
FM:
Slow: 2,827 g, 2,868 g, 2,842 g
Med: 2,851 g, 2,923 g, 2,842 g
Fast: 2,841 g, 3,036 g, 2,977 g
%FM:
Slow: 51.3%, 49.7%, 50.2%
Med: 51.2%, 49.7%, 49.9%
Fast: 51.2%, 50.1%, 51.0% 
Hind et al. (2011)15 Lunar iDXA 18 men, 34 women
Low BMI (<18.5 g/m2) n = 2; normal BMI (18.5–24.9 kg/m2), n = 25; overweight (25–29.9 kg/m2) n = 13; obese (30–34.9 kg/m2) n = 10; and severely obese (=35.0 kg/m2), n = 2 
CV:
FM: 0.82
%FM: 0.86
%AF: 2.32
%GF: 0.96
LM: 0.51 
N/A 

Abbreviations: %AF, percent android fat; %FM, percent fat mass; %GF, percent gynoid fat; ALST, appendicular lean soft tissue; BM: body mass; BMI, body mass index; CV, coefficient of variation; DXA, dual X-ray absorptiometry; FM, fat mass; LM, lean mass; LTBP, Leeds total body phantom; med, medium; N/A, not applicable; SEE, standard error of the estimate; TBtissue, total body tissue; VCP, variable composition phantom; WBP, whole-body phantom.

In vitro comparisons between the Lunar Prodigy and Lunar DPX-L scanners on the Leeds Total Body Phantom resulted in significant variation in total body tissue, lean mass, and fat mass. In the same study, in vivo measures in 72 human subjects showed that, compared with the Lunar DPX-L, the Lunar Prodigy produced higher measures of total body tissue, body mass, lean mass, fat mass, and percent fat mass.11

Altogether, the aforementioned comparisons reveal considerable variation in body composition measurements between different models of DXA scanners, which could incorrectly identify or could mask the effects of an intervention on changes in lean or fat mass.

Comparison of body composition data between different or corresponding models of DXA devices

Given the observed variability in body composition measurements, quality control and standardized procedures should be implemented prior to initiating multicenter studies that use DXA for assessing lean mass, body fat, and percent body fat. Although measurements between different and corresponding DXA models are highly correlated, variability – upward of 900 g – could skew results and introduce errors when DXA-derived values for body composition are used in regression analysis to develop indices that estimate risk of disease. As a means of limiting measurement error in multicenter trials, total body phantoms of known body composition can be used to develop regression equations that normalize DXA-derived values for total mass, lean mass, body fat, and percent body fat between instruments. Body composition phantoms can consist of natural (meat, bone, and fat) or synthetic constituents that mimic attenuation patterns of tissues. Synthetic components include high-density polyethylene or acrylic to mimic fat tissue, polyvinylchloride to mimic lean tissue, and aluminum to mimic bone tissue.12 As an example, Aasen et al.6 developed coefficients for converting body composition measurements obtained with the Lunar Expert and Hologic Delphi W to values that correspond to those obtained with the Lunar Prodigy (Table 2). However, when developing coefficients for aligning interdevice DXA measurements, bias that is contingent on differences in body composition analysis should be considered. Using the Lunar variable composition phantom (VCP) at 16.0%, 21.2%, 28.1%, 36.8%, and 44.4% body fat, body compositional analysis revealed significant differences in results for percent body fat between the Lunar Prodigy and the Lunar DPX, respectively.11 Thus, developing conversion coefficients using a phantom of varying body composition and/or human subjects will further minimize interdevice error. The following two sections discuss the effects of gender and scan speed on DXA measurement variability, both of which can also be incorporated into regression equations if effects of these factors are demonstrated between instruments.

Table 2

Examples of coefficients for converting body composition measurements obtained on the Lunar Expert and the Hologic Delphi W to values that correspond to those obtained on the Lunar Prodigy

DXA measurement Conversion factora 
Total body mass 
Lunar Expert 1.003 
Hologic Delphi W 1.011 
Lean mass 
Lunar Expert 1.018 
Hologic Delphi W 0.967 
Fat mass 
Lunar Expert 0.954 
Hologic Delphi W 1.079 
DXA measurement Conversion factora 
Total body mass 
Lunar Expert 1.003 
Hologic Delphi W 1.011 
Lean mass 
Lunar Expert 1.018 
Hologic Delphi W 0.967 
Fat mass 
Lunar Expert 0.954 
Hologic Delphi W 1.079 
a

According to Aasen et al. (2006)6

Effect of gender on interdevice measurement error

When formulating regression equations that permit interdevice comparisons of DXA-derived values for body mass, lean mass, fat mass, and percent body fat, gender should also be assessed as a source of interdevice measurement error. Using 39 men and 39 women with body mass indices (BMIs) ranging 17–45 kg/m2, Soriano et al.13 compared body fat percent measurements across pencil-beam (Lunar DPX and Lunar DPX-L) and fan-beam (Lunar Prodigy and Hologic Delphi A) DXA devices. When transition equations were developed for converting percent body fat values obtained on the Hologic Delphi A, Lunar DPX, and Lunar DPX-L to values that correspond to those obtained on the Lunar Prodigy, slope coefficients and constants were significantly different for males and females.13 Thus, results demonstrate that, although data obtained from the different DXA models were highly correlated, body composition values varied significantly between devices, and most interdevice error was gender dependent. The Soriano et al.13 study identified gender as a source of measurement error when comparing body composition data obtained on different models of DXA devices. Given that measurement error tends to increase as fat deposition increases, the effect of gender on interdevice measurement error may occur only when the level of adipose tissue reaches a certain threshold within a certain body region. The effects of excess adiposity alongside fundamental assumptions of DXA body composition analyses are discussed further in subsequent sections of the present manuscript.

Effect of scan speed on interdevice measurement error

Scan speed should also be considered when using DXA for assessing body composition. Typically, DXA software will automatically adjust the scan speed, depending on the weight or body mass index (BMI) of the subject undergoing DXA analysis. However, using a phantom consisting of 10% and 50% body fat, Guo et al.14 compared interdevice measurement error between two identical Lunar Prodigy DXAs (A and B) and a Lunar DPX-L. Differences in tissue composition for the high-fat phantom between fast and medium modes as well as between medium and slow modes were significantly greater for the Lunar Prodigy (B) (Lunar Prodigy (B)fast/medium: 1.1%; Lunar Prodigy (B)medium/slow: 0.3%) compared with the Lunar DPX-L (Lunar DPX-L fast/medium: 0.0%; Lunar DPX-L medium/slow: 0.1%).14 After adjusting for scan speed, values for lean and fat mass were significantly different between all three DXA instruments for the 10%- and 50%-fat phantoms.14 However, results demonstrated that scan speed was the primary source of interdevice variation when estimating body fat percentage. Thus, within the confines of clinical studies, scan speed should be recognized as a potential source of interdevice measurement error when using different DXA scanners for body composition analysis. If an effect of scan speed is noted, scan speed can also be factored into regression equations that transition body composition measurements between instruments.

Intrainstrument Measurement Error with DXA Devices

In addition to interdevice variation, intrainstrument variation can facilitate false-positive results when DXA is used to assess shifts in body composition. Given that most nutritional or exercise intervention studies using DXA to investigate lean/fat mass accretion or loss are single-center trials, baseline and postintervention measures of body composition are evaluated using the same single DXA unit. As demonstrated in Table 1, intrainstrument coefficients of variation for fan-beam and pencil-beam DXA models are well within scientific acceptability at <6%.6,7,9,11,14,15 Depending on the length of a dietary or exercise intervention, however, expected changes in lean and/or fat mass could be within the limits of coefficients of variation and thus mitigate the ability to determine whether the intervention facilitated a change in body composition. For example, for a subject who initially has 50 kg of lean mass, DXA analysis could determine that a dietary intervention increased lean mass by 1,500 g over a 6-month period. If predetermined intradevice coefficients of variation are 4% for lean body mass, it is difficult to determine whether the 1,500 g of lean mass accretion is secondary to the intervention or is attributable to the intrainstrument measurement error. In addition, if intradevice precision error is contingent on levels of adiposity or lean body mass, measurement error will vary, depending on the body composition of the subjects being examined.16

Determination of intradevice measurement error

Most methodologies currently used to assess whether differences in consecutive DXA measures are clinically relevant or are secondary to intradevice precision error were designed for changes in bone mineral density and the diagnosis of osteoporosis17,19; thus, they are not applicable to DXA-derived body composition analysis. However, Bland and Altman's 95% limits of agreement,20 a method typically used to verify the agreement between a standard and novel methodology (Figure 1), could be used to determine the smallest detectable difference (SDD) between any endpoint obtained by DXA analysis, including lean mass, fat mass, and percent body fat measures. Briefly, SDD is the magnitude of 95% confidence intervals that coincide with the difference between two measurements21,22: 

1
formula
 
2
formula
where d represents the mean difference between two measurements and 1.96 establishes the 95% confidence interval of a normal distribution. The standard error of the mean (SEM) is calculated using equation 1 and represents the standard error of the measurement. The factor graphic reflects the error within each of the two measurements.22 SDD, as calculated by equation 1, is reported in absolute units. Likewise, SDD can also be represented as a percentage, using the coefficient of variation: 
formula
where %SDD and %CVdiff represent the percent smallest detectable difference and the percent total coefficient of variation of the difference, respectively.23 Once SDD and %SDD are calculated, it is assumed that differences beyond graphic and d ± 1.96 × %CVdiff are likely not artifacts of intradevice precision errors and are secondary to the intervention that was imposed during the study period.

Figure 1

Bland-Altman plot demonstrating limits of agreement and the smallest detectable difference.

Figure 1

Bland-Altman plot demonstrating limits of agreement and the smallest detectable difference.

Data can be evaluated graphically using Bland-Altman plots, where differences between two measurements are plotted against the mean (Figure 1). Within the context of DXA analysis, Bland-Altman plots are often used to evaluate the agreement between different DXA models. However, within the realm of the same DXA model, Bland and Altman's limits of agreement and SDD provide researchers with two important pieces of information. First, prior to initiating a clinical trial, the calculation of SDD provides a starting point for determining a standardized effect size to be used for calculating sample size for an upcoming clinical study that assesses whether an intervention facilitates changes in body composition. Second, upon completion of a clinical trial, researchers can determine the likelihood that observed changes in body composition are attributable to intradevice precision error.

It should be noted that SEM (absolute units) or %CVdiff (relative) can yield divisive results for SDD. A study by Wosje et al.23 compared DXA-derived intradevice precision error for lean and fat mass between lean and obese children. When SDD was calculated using SEM, SDD values were 3.3 and 2.8 times higher for obese children and lean children, respectively, and demonstrated that, as weight increases, whole-body standard deviations increase as well. Conversely, %SDD values derived from %CVdiff were 1.5 times lower for obese children. Given that measurement error was greater in obese children, Wosje et al.23 indicate that smaller %SDD values for body composition were secondary to a larger denominator when calculating the coefficient of variation (SDwithin-person/meandiff). The aforementioned results suggest that SDD calculated from SEM is a more reliable indicator of SDD because, compared with %CVdiff, SEM better reflects variation in the data within the sample population.

Effect of excess body fat on intradevice measurement error

Studies demonstrate that intradevice precision error is increased in obese subjects. As discussed above, differences in lean and fat mass from duplicate DXA scans resulted in larger standard deviations in obese children than in lean controls.23 Using a fan-beam Hologic QDR 4500A, Valentine et al.24 examined how regional adiposity modulates intradevice measurement error. Subjects underwent an initial DXA scan followed by a second scan with packets of lard placed on the abdominal and thigh regions of men and on the abdominal, thigh, and chest regions of women.24 Results demonstrated that measurement error, as indicated by regression analysis on Bland-Altman plots, was more pronounced when lard was placed in the abdominal regions. Moreover, 59% of exogenous lard was detected as abdominal adipose, and the remaining 41% was detected as lean mass. Similar results were observed when lard was applied to the thigh regions, but to a lesser degree. The aforementioned results suggest that DXA overestimates lean body mass and underestimates fat mass in obese subjects.24 Valentine et al.24 suggest that miscalculations in body composition result from assumptions ingrained within the algorithms that decipher lean, fat, and mineral mass. This will be discussed further in a subsequent section of this review. Similar results were observed when values for lean and fat mass in obese, lean, and anorexic premenopausal women were compared between DXA and CT scans.25 Given that CT is considered a gold standard methodology for accurately identifying and quantifying tissue, CT scanning was considered a reference method for determining body composition data. Although values for lean and fat mass were highly correlated between DXA and CT scans, DXA underestimated thigh and abdominal fat mass and overestimated thigh lean mass.25 Moreover, Bland-Altman plots demonstrate that the degree of measurement error increased as body weight increased.25 Overall, studies demonstrate that the degree of adiposity contributes to DXA-derived measurement error.

Compared with lean individuals, some measurement error in obese subjects could be secondary to differences in contact points with the DXA table. Lohman et al.26 suggest that manual adjustment of regions of interest can significantly increase the reliability of body composition measures for extremities, especially in the upper body.

To minimize measurement error that is secondary to adiposity, appropriate quality-control procedures should be implemented prior to initiating a study that uses DXA to assess body composition. For example, knowing which population an intervention is targeting, lean versus obese, would facilitate the appropriate calculations, such as SDD, for considering measurement error. The use of phantoms of known body composition would allow researchers to assess intradevice precision error over a range of body fat percentages. If phantoms are not available, SDD could also be calculated by scanning individuals across a range of body compositions or within a specific range for an upcoming clinical trial. The aforementioned analysis would be useful when deciphering an appropriate effect size for future clinical trials as well as for measurement drift over time.

Effect of scan speed on intradevice measurement error

Similar to interdevice measurement error, scan speed should be considered when conducting consecutive DXA scans on the same instrument. From the Guo et al.14 study discussed above, scans conducted on the lean phantom (10% body fat) with the Lunar DPX-L and Lunar Prodigy (B) revealed significantly higher measures of percent body fat (DPX-Lfast: 10.6%; Lunar Prodigy (B)fast: 11.3%) compared with slow and medium-speed scans (DPX-Lmedium: 9.9%; DPX-Lslow: 9.8%; Lunar Prodigy (B)medium: 11.0%; Lunar Prodigy (B)slow: 11.0%). However, scans conducted on the lean phantom using the Lunar Prodigy (A) measured percent body fat as 10.4% across all scan speeds. Conversely, Lunar DPX-L scans conducted on the adipose-rich phantom (50% body fat) demonstrated no difference in values for percent body fat with varying scan speeds (DPX-Lfast: 51.2%; DPX-Lmedium: 51.2%; DPX-Lslow: 51.3%), while analysis on the Lunar Prodigy (A) showed that fast mode (Lunar Prodigy (A)fast: 50.1%) significantly overestimated percent body fat compared with slow and medium speeds (Lunar Prodigy (A)medium: 49.7%; Lunar Prodigy (A)slow: 49.7%).14 The above-mentioned results reinforce scan speed as a potential source of intradevice measurement error and suggest that variability of scan speed within instruments should be assessed. The observation that scan speed modulates intradevice variation also suggests that scan speed should remain constant, especially within subjects, for clinical trials in which DXA is used to investigate the effects of an intervention on body composition.

Comparison between DXA and Other Methodologies for Assessing Body Composition

DXA absorptiometry versus CT

CT is considered the gold standard for assessing body composition. However, given its high cost of ownership and the high level of expertise required for its operation, CT is often an inaccessible methodology for measuring shifts in lean and fat mass following a dietary or exercise intervention. Nonetheless, studies demonstrate that body composition measurements obtained by CT and DXA are highly correlated. For example, for body fat, strong correlations between DXA-derived trunk fat and CT-derived abdominal fat area were observed for anorexic (r = 0.86), lean (r = 0.94), and obese (r = 0.94) premenopausal women.25 In the same study, DXA-derived leg fat and CT-derived thigh adipose area were also highly correlated at r = 0.92, 0.93, and 0.94, respectively.25 Similarly, Glickman et al.27 demonstrated strong correlations between CT-derived total abdominal fat mass and DXA-derived total abdominal tissue mass (r = 0.86) as well as between CT-derived abdominal fat mass and DXA-derived abdominal fat mass (L1–L4 region of interest) (r = 0.97) among men and women with percent body fat ranging from 8.0% to 58%. In addition, limits of agreement between CT and DXA measurements of total abdominal tissue mass (−1.56–2.54 kg) and fat mass (−0.40–1.94 kg) were fairly narrow and suggest congruity between instruments.27 Finally, comparisons between DXA-derived and CT-derived total body fat mass have also been shown to be highly correlated (r = 0.99) between men and women with BMIs ranging from 22.9 to 38.7 and from 28.8 to 37, respectively.28 Overall, studies demonstrate excellent correlations between CT-derived and DXA-derived measures of total and abdominal adipose tissue.

Given that visceral adiposity is a risk factor for lifestyle-related diseases such as type II diabetes, cardiovascular disease, and the metabolic syndrome, shifts in intra-abdominal fat is often the primary endpoint for clinical studies that engage dietary or exercise interventions. However, despite the strong correlations outlined above, agreement between measures of adiposity obtained by DXA and CT weaken when analysis is focused on visceral adiposity. Results from the Bredella et al.25 study discussed above demonstrated lower correlations between CT-derived visceral fat area and DXA-derived trunk fat mass (anorexic, r = 0.62; lean, r = 0.51; obese, r = 0.70) compared with the above-mentioned correlations observed for abdominal tissue (r = 0.86) and total abdominal fat mass (r = 0.97). Similar results were observed by Snijder et al.,29 where correlations between CT-derived and DXA-derived abdominal fat ranged between 0.87 and 0.97, while correlations between CT-derived visceral fat and DXA-derived trunk and subregion (manually derived region of interest above the iliac crest) adipose tissue ranged from 0.65 to 0.79 and from 0.61 to 0.83, respectively.29 It should be noted, however, that the Snijder et al.29 study used pencil-beam technology, which is rarely used in more recent clinical trials. Although measures of trunk fat mass from DXA are moderate predictors of visceral adipose, extrapolating fat measurements obtained by DXA to intra-abdominal adipose could become problematic and result in erroneous conclusions when visceral adiposity is the primary endpoint for a clinical trial.

Similar to observations discussed above, absolute values for DXA-derived fat and lean mass differ significantly from values obtained by CT. Kullberg et al.28 demonstrated that, compared with CT and MRI, DXA underestimated total fat mass by 5.23 ± 1.71 kg and 4.67 ± 2.38 kg, respectively. Conversely, the difference in fat mass between CT-derived and MRI-derived values was nonsignificant, at 0.56 ± 1.08 kg. Furthermore, within the Snijder et al.29 study, a 10% underestimation in fat mass was observed using DXA compared with CT. When Levine et al.30 evaluated the capacity of DXA to quantify appendicular muscle mass, results demonstrated strong correlations between CT-derived leg fat volume and DXA-derived leg fat (r = 0.97) as well as between CT-derived leg muscle and DXA leg fat free mass (r = 0.97). However, absolute measures of fat mass demonstrate that, compared with CT-derived thigh fat volume (3,764 ± 2,184 cm3), DXA underestimated thigh fat mass by approximately 10% (3,394 ± 1,957 g). In addition, DXA analysis overestimated thigh muscle mass by approximately 12%, or 705 ± 306 g, compared with CT.30

Inconsistencies in absolute measures of lean mass are even more pronounced when changes in lean and fat mass over a period of time are compared between DXA and CT. When body composition via DXA and CT was assessed before and after an approximately 6-kg weight loss, Tylavsky et al.31 demonstrated that DXA underestimated changes in thigh lean soft tissue (−24.9 ± 34.3 g) and overestimated changes in fat mass (−112.1 ± 51.5 g), respectively, compared with CT (lean soft tissue  =  −44.7 ± 50.1 g; fat mass  =  −45.7 ± 52.6 g). In addition, only moderate correlations were demonstrated between CT-derived and DXA-derived changes in thigh soft tissue mass (r = 0.55) as well as between changes in CT-derived and DXA-derived fat mass (r = 0.67).31 Investigators from the Tylavsky et al.31 study suggest that lower correlation coefficients in measures of change reflect compounding random errors. Similar results were demonstrated by Delmonico et al.,32 who assessed changes in thigh skeletal muscle by DXA and CT following a 10-week resistance exercise intervention. DXA and CT analyses revealed that resistance exercise facilitated significant increases in thigh skeletal muscle mass at 2.9% and 3.9%, respectively. However, measurement error, calculated as the coefficient of variation for DXA-derived fat-free thigh mass, was 3.4%, which is greater than the observed change in thigh skeletal muscle. Conversely, CT-derived coefficient of variation for fat-free thigh mass was only 0.6% and suggests that DXA is an inappropriate methodology for deciphering small changes in skeletal muscle mass following an exercise intervention.32 However, to detect small changes in lean or fat mass, quality-control procedures, such as the calculation of SDD, would help detect whether observed differences in body composition are secondary to treatment or possibly due to intradevice measurement error.

DXA versus bioelectrical impedance

Although DXA has been shown to be more sensitive than other techniques, bioelectrical impedance (BIA) and anthropometry are still used in large epidemiological studies and/or research that is monetarily constrained. With the exception of anthropometry, BIA represents the most cost-effective method for assessing body composition. With BIA, lean and fat mass are determined by applying a slight electrical current to a study participant and measuring its resistance through his or her body. Typically, body fat facilitates high resistance, while increased water content, as found in lean tissues, facilitates lower resistance.33 Final values for lean and fat mass, as well as for visceral adiposity, are based on linear regression equations. Recently, Bosy-Westphal et al.34 compared body composition data from DXA with data obtained by four different BIA instruments, the Tanita BC 532, the Soehnle Body Balance, the Omron BF400, and the Omron BF500. The Tanita BC 532, Soehnle Body Balance, and Omron BF400 used foot-to-foot technology. In contrast, the Omron BF500 used eight electrodes (two electrodes in each foot pad and two additional electrodes in each hand sensor). Results demonstrated strong correlations for percent fat mass between DXA and BIA at r = 0.94, 0.82, 0.92, and 0.96 for the Omron BF400, the Soehnle Body Balance, the Tanita BC532, and the Omron BF500, respectively. However, in addition to significant underestimations in percent fat mass with BIA compared with DXA, Bland-Altman analyses revealed wide limits of agreement between DXA and BIA for percent fat at −14.54% to +8.58%, −9.33% to +6.59%, −7.1% to +6.14%, and −6.59% to +4.61%, for the Soehnle Body Balance, the Tanita BC532, the Omron B400, and the Omron B500, respectively.34 It should also be mentioned that, in addition to having the lowest limits of agreement, the Omron BF500 was the only BIA instrument that did not produce systematic bias when compared with DXA, which could be a result of the use of eight versus four electrodes.34

Similar results were shown by Jensky-Squires et al.,35 who found that two BIA devices, the Omron HBF360 and the InBody 320, produced significantly different values for percent body fat than DXA. Neovius et al.36 demonstrated that changes in body fat, as assessed by DXA and eight-electrode BIA before and after a 6-month exercise intervention, varied significantly between the two methodologies, especially in patients with increased percent body fat. Finally, bioelectrical spectroscopy, a method of BIA that uses alternating electrical frequencies, underestimated percent fat mass by 4.6% in hockey players and by 1.1% in soccer players when compared with DXA analysis.37 Discrepancies between groups were suggested to be secondary to the principles of BIA and increased muscle mass in the arms of hockey players compared with soccer players.37 Overall, data suggest that, given the wide variability between body composition measures obtained on DXA and various BIA instruments, DXA and BIA are not comparable methodologies for assessing body composition during short-term clinical trials.

DXA versus anthropometry

Of all the methodologies used to determine body composition, anthropometry is the most rudimentary. Regression equations that use height, weight, skin folds, waist circumference, BMI, and waist-to-height ratios are used to quantify total fat and lean mass as well as visceral adiposity. Surprisingly, certain anthropometric measurements have been shown to be reasonable predictors of specific measures of body composition. For example, Snijder et al.29 demonstrated moderate correlations between CT-derived visceral fat and weight (r = 0.500–0.734), BMI (r = 0.562–0.721), abdominal circumference (r = 0.606–0.751), and sagittal diameter (r = 0.768–0.838). The aforementioned correlations between CT-derived visceral fat and anthropometrics were similar to the previously discussed correlation coefficients between CT-derived visceral fat and DXA.29 Moreover, waist circumference and sagittal diameter were strongly correlated with CT-derived total abdominal fat (waist circumference, r = 0.87; sagittal diameter, r = 0.93) and visceral abdominal fat (waist circumference, r = 0.84; sagittal diameter, r = 0.93), findings that were similar to strong correlations between CT-derived total and visceral fat and DXA-derived trunk and abdominal fat at r = 0.86–0.97.38 Conversely, waist-to-hip ratios were the worst predictors of total and visceral adiposity.38 Finally, Browning et al.,39 demonstrated that DXA, waist circumference, BMI, and BIA were more strongly correlated with total abdominal adipose tissue than with visceral adipose tissue, as measured by MRI. Given that anthropometry does not use imagery techniques to quantify target tissues, its suitability for measuring changes in lean or fat mass in short-term clinical trials is questionable. Although correlations with visceral adiposity and anthropometric measures have been demonstrated, anthropometry is likely more suitable for large-scale epidemiological or observational studies.39

DXA Software as a Source of Systematic Error when Measuring Body Composition

As discussed previously, although certain measurements obtained by CT and DXA are highly correlated, comparability between absolute values for lean and fat mass can vary between the two methodologies. Divisive conclusions regarding the comparability between CT- and DXA-derived values for fat and lean mass can, at least partially, be attributed to fundamental assumptions that the DXA software utilizes to assess data corresponding to the attenuation of X-ray energies through soft and bone tissue. First, fat content is estimated from the attenuation of fat found in bone-free soft tissue.40 Furthermore, the calibration of DXA facilitates the ability for software to decipher lean and fat mass within bone-free soft tissues. The aforementioned composition of soft tissue (lean and fat mass) is extrapolated to the composition of soft tissue that overlays bone.41 The algorithms that software packages use for extrapolating data are proprietary, making it difficult to compare assumptions made by manufacturers.41 Second, given that 40–45% of the pixels used to quantify lean and fat mass contain bone, they are excluded when determining the composition of soft tissue. That is, it is assumed that the areas of the body that are used to obtain body composition data (bone-free soft tissue) are associated with the composition of the area not analyzed (containing bone).40 In fact, given that the composition of soft tissue that overlays bone is an extrapolation, Snijder et al.29 suggest that a manually analyzed region of interest above the iliac crest, which is free of bone, is ideal for the analysis of abdominal fat since it does not require the use of algorithms to extrapolate measures of lean and fat mass. The final assumption is that body thickness (anterior to posterior) does not affect DXA measurements.40 However, in recent years, manufacturers of DXA, such as GE Healthcare, have attempted to reduce measurement errors associated with body thickness by providing a “high power” option for measuring subjects with =25 cm thickness. Given the assumptions made by DXA software for body composition analysis, it is reasonable to assume that software upgrades provided by DXA manufacturers will attempt to refine data extrapolation as well as to increase precision and accuracy.

Variability between different software versions has been verified. Using a fan-beam Hologic QDR 4500A, Cordero-MacIntyre et al.42 compared body composition data from weight-stable and obese postmenopausal women, using two succeeding versions of Hologic DXA software, V8.1a and V8.21, respectively. Although measures of lean and fat mass were correlated between both software versions for all body composition data (r ≥ 0.995), systematic errors between V8.1a and V8.21 were demonstrated.42 At baseline, trunk fat mass, trunk lean mass, total fat mass, and total lean mass were 7.3%, 8.4%, 4.2%, and 5.2% greater with V8.21 than with V8.1a, respectively. Moreover, baseline and 3-month values for body mass were closer to scale-derived body weights using V8.21 (baseline: −2.40%; 3-month: −2.89%) compared with V8.1a (baseline: −7.24%; 3-month: −7.65%).42 Enhanced accuracy for data analyzed using V8.21 stems from an increased ability to adjust for beam magnification, a source of systematic error within the realm DXA analysis.42

Using overweight and obese women with BMIs of ≥25 kg/m2, Genton et al.43 compared measures of lean and fat mass using Hologic software version V8.26 in normal and high-power mode. Body composition analysis on the Hologic system were also compared with values for lean and fat mass obtained on a GE Lunar Prodigy fan-beam DXA using software version V6.5.43 Values for body weight obtained on Hologic (normal and high-power modes) and Prodigy software were highly correlated with scale weights (r ≥ 0.995).43 However, in normal mode, the Hologic V8.26, underestimated body weight by 1.7 kg and 4.0 kg across all BMIs and BMIs ≥40 kg/m2, respectively. Conversely, the Hologic V8.26, at the high-power mode, overestimated body weight by approximately 2.0 kg. The Prodigy V6.5 gave the most accurate comparison to scale weight, at +0.5 ± 0.8 kg across all ranges of BMIs. In addition, compared with the normal mode, the Hologic V8.26 high-power mode significantly overestimated the lean body mass of subjects with BMIs ≥40 kg/m2 (54.2 ± 3.8 kg versus 59.5 ± 4.5 kg, respectively). Genton et al.43 indicated that the Hologic V8.26 normal-power mode should not be used for assessing the body composition of women weighing ≥90 kg. Compared with the percent body fat estimated by the Hologic V8.26 in normal mode (41.3 ± 5.0%) and high power mode (40.5 ± 4.8%), the Prodigy V6.5 software indicated increased percent body fat (44.7 ± 5.9%) for subjects across the entire BMI range.43 In addition, compared with the Hologic normal (total fat mass, 45.5 ± 3.4 kg; percent fat, 44.4 ± 4.1%) and high-power modes (total fat mass, 47.4 ± 4.1 kg; percent fat, 43.6 ± 4.1%), total fat mass and percent fat mass were also increased using the Prodigy V6.5 at 53.1 ± 4.3 kg and 49.6 ± 4.2%, respectively.43 Altogether, the above-mentioned results demonstrate that values for body composition can vary significantly, depending on which version of software is used to integrate raw data obtained from DXA. Given that DXA software is continually upgraded, with each subsequent version better able to adjust for the fundamental assumptions that facilitate systematic errors associated with DXA analysis, software updates for DXA are encouraged, especially when clinical studies are required to detect small changes in body composition.

Conclusion

Overall, the present review provides a comprehensive overview of factors to be considered when using DXA technology to evaluate body composition in the context of short- to medium-term clinical trials that assess nutrition and/or exercise interventions. Quality-control procedures that help adjust for inter- and intradevice measurement errors are crucial to ensure that changes in lean and fat mass are correctly identified. Comparisons of values for lean and fat mass obtained using DXA and CT suggest that, despite strong correlations, the fundamental assumptions of DXA technology can cause significant systematic errors that overestimate and underestimate measures of body composition. However, depending on the study protocol and pending the implementation of other quality-control measures, precision – rather than accuracy – may be more important for a clinical study's primary and secondary endpoints. Finally, compared with BIA and anthropometry, DXA is a superior methodology for determining values for lean and fat mass. Overall, DXA is an advanced, cost-effective technology that is useful for acquiring body composition data in short- to medium-term clinical trials as long as its limitations are considered during the planning stages of an intervention.

CM and AK were equally responsible in conceptualizing and writing the present manuscript.

Declaration of interest

The authors have no relevant interests to declare.

References

1
French
SA
Jeffery
RW
Murray
D
.
Is dieting good for you? Prevalence, duration and associated weight and behaviour changes for specific weight loss strategies over four years in US adults
.
Int J Obes Relat Metab Disord
 .
1999
;
23
:
320
327
.
2
Appel
LJ
Clark
JM
Yeh
HC
, et al.
Comparative effectiveness of weight-loss interventions in clinical practice
.
N Engl J Med
 .
2011
;
365
:
1959
1968
.
3
Schwartz
AV
Johnson
KC
Kahn
SE
, et al.
Effect of 1 year of an intentional weight loss intervention on bone mineral density in type 2 diabetes: results from the look AHEAD randomized trial
.
J Bone Miner Res
 .
2012
;
27
:
619
627
.
4
Caterson
ID
Finer
N
Coutinho
W
, et al.
Maintained intentional weight loss reduces cardiovascular outcomes: results from the Sibutramine Cardiovascular OUTcomes (SCOUT) trial
.
Diabetes Obes Metab
 .
2012
;
14
:
523
530
.
5
Wilcox
S
Sharpe
PA
Parra-Medina
D
, et al.
A randomized trial of a diet and exercise intervention for overweight and obese women from economically disadvantaged neighborhoods: Sisters Taking Action for Real Success (STARS)
.
Contemp Clin Trials
 .
2011
;
32
:
931
945
.
6
Aasen
G
Fagertun
H
Halse
J
.
Body composition analysis by dual X-ray absorptiometry: in vivo and in vitro comparison of three different fan-beam instruments
.
Scand J Clin Lab Invest
 .
2006
;
66
:
659
666
.
7
Covey
MK
Berry
JK
Hacker
ED
.
Regional body composition: cross-calibration of DXA scanners – QDR4500W and Discovery Wi
.
Obesity (Silver Spring)
 .
2010
;
18
:
632
637
.
8
Ioannidou
E
Padilla
J
Wang
J
, et al.
Pencil-beam versus fan-beam dual-energy X-ray absorptiometry comparisons across four systems: appendicular lean soft tissue
.
Acta Diabetol
 .
2003
;
40
(
Suppl 1
):
S83
S85
.
9
Nord
RH
Homuth
JR
Hanson
JA
, et al.
Evaluation of a new DXA fan-beam instrument for measuring body composition
.
Ann N Y Acad Sci
 .
2000
;
904
:
118
125
.
10
Louis
O
Verlinde
S
Thomas
M
, et al.
Between-centre variability versus variability over time in DXA whole body measurements evaluated using a whole body phantom
.
Eur J Radiol
 .
2006
;
58
:
431
434
.
11
Oldroyd
B
Smith
AH
Truscott
JG
.
Cross-calibration of GE/Lunar pencil and fan-beam dual energy densitometers – bone mineral density and body composition studies
.
Eur J Clin Nutr
 .
2003
;
57
:
977
987
.
12
Diessel
E
Fuerst
T
Njeh
CF
, et al.
Evaluation of a new body composition phantom for quality control and cross-calibration of DXA devices
.
J Appl Physiol
 .
2000
;
89
:
599
605
.
13
Soriano
JMP
Ioannidou
E
Wang
J
, et al.
Pencil-beam vs fan-beam dual-energy X-ray absorptiometry comparisons across four systems
.
J Clin Densitom
 .
2004
;
7
:
281
289
.
14
Guo
Y
Franks
PW
Brookshire
T
, et al.
The intra- and inter-instrument reliability of DXA based on ex vivo soft tissue measurements
.
Obes Res
 .
2004
;
12
:
1925
1929
.
15
Hind
K
Oldroyd
B
Truscott
JG
.
In vivo precision of the GE Lunar iDXA densitometer for the measurement of total body composition and fat distribution in adults
.
Eur J Clin Nutr
 .
2011
;
65
:
140
142
.
16
Gluer
CC
.
Monitoring skeletal changes by radiological techniques
.
J Bone Miner Res
 .
1999
;
14
:
1952
1962
.
17
Baim
S
Wilson
CR
Lewiecki
EM
, et al.
Precision assessment and radiation safety for dual-energy X-ray absorptiometry: position paper of the International Society for Clinical Densitometry
.
J Clin Densitom
 .
2005
;
8
:
371
378
.
18
Nelson
L
Gulenchyn
KY
Atthey
M
, et al.
Is a fixed value for the least significant change appropriate?
J Clin Densitom
 .
2010
;
13
:
18
23
.
19
Cummings
SR
Black
D
.
Should perimenopausal women be screened for osteoporosis?
Ann Intern Med
 .
1986
;
104
:
817
823
.
20
Bland
JM
Altman
DG
.
Statistical methods for assessing agreement between two methods of clinical measurement
.
Lancet
 .
1986
;
1
:
307
310
.
21
El Maghraoui
A
Achemlal
L
Bezza
A
.
Monitoring of dual-energy X-ray absorptiometry measurement in clinical practice
.
J Clin Densitom
 .
2006
;
9
:
281
286
.
22
de Vet
HC
Terwee
CB
Knol
DL
Bouter
LM
.
When to use agreement versus reliability measures
.
J Clin Epidemiol
 . 2006;
59
:
1033
1039
.
23
Wosje
KS
Knipstein
BL
Kalkwarf
HJ
.
Measurement error of DXA: interpretation of fat and lean mass changes in obese and non-obese children
.
J Clin Densitom
 .
2006
;
9
:
335
340
.
24
Valentine
RJ
Misic
MM
Kessinger
RB
, et al.
Location of body fat and body size impacts DXA soft tissue measures: a simulation study
.
Eur J Clin Nutr
 .
2008
;
62
:
553
559
.
25
Bredella
MA
Ghomi
RH
Thomas
BJ
, et al.
Comparison of DXA and CT in the assessment of body composition in premenopausal women with obesity and anorexia nervosa
.
Obesity (Silver Spring)
 .
2010
;
18
:
2227
2233
.
26
Lohman
M
Tallroth
K
Kettunen
JA
, et al.
Reproducibility of dual-energy x-ray absorptiometry total and regional body composition measurements using different scanning positions and definitions of regions
.
Metabolism
 .
2009
;
58
:
1663
1668
.
27
Glickman
SG
Marn
CS
Supiano
MA
, et al.
Validity and reliability of dual-energy X-ray absorptiometry for the assessment of abdominal adiposity
.
J Appl Physiol
 .
2004
;
97
:
509
514
.
28
Kullberg
J
Brandberg
J
Angelhed
JE
, et al.
Whole-body adipose tissue analysis: comparison of MRI, CT and dual energy X-ray absorptiometry
.
Br J Radiol
 .
2009
;
82
:
123
130
.
29
Snijder
MB
Visser
M
Dekker
JM
, et al.
The prediction of visceral fat by dual-energy X-ray absorptiometry in the elderly: a comparison with computed tomography and anthropometry
.
Int J Obes Relat Metab Disord
 .
2002
;
26
:
984
993
.
30
Levine
JA
Abboud
L
Barry
M
, et al.
Measuring leg muscle and fat mass in humans: comparison of CT and dual-energy X-ray absorptiometry
.
J Appl Physiol
 .
2000
;
88
:
452
456
.
31
Tylavsky
FA
Lohman
TG
Dockrell
M
, et al.
Comparison of the effectiveness of 2 dual-energy X-ray absorptiometers with that of total body water and computed tomography in assessing changes in body composition during weight change
.
Am J Clin Nutr
 .
2003
;
77
:
356
363
.
32
Delmonico
MJ
Kostek
MC
Johns
J
, et al.
Can dual energy X-ray absorptiometry provide a valid assessment of changes in thigh muscle mass with strength training in older adults?
Eur J Clin Nutr
 .
2008
;
62
:
1372
1378
.
33
Chumlea
C
Sun
SS
.
Bioelectrical impedance analysis
. In:
Heymsfield
SB
Lohman
TG
Zimain
W
, et al.,
eds. Human Body Composition
 ,
2nd ed
.
Champaign, IL
:
Human Kinetics
;
2005
:
79
88
.
34
Bosy-Westphal
A
Later
W
Hitze
B
, et al.
Accuracy of bioelectrical impedance consumer devices for measurement of body composition in comparison to whole body magnetic resonance imaging and dual X-ray absorptiometry
.
Obes Facts
 .
2008
;
1
:
319
324
.
35
Jensky-Squires
NE
Dieli-Conwright
CM
Rossuello
A
, et al.
Validity and reliability of body composition analysers in children and adults
.
Br J Nutr
 .
2008
;
100
:
859
865
.
36
Neovius
M
Udden
J
Hemmingsson
E
.
Assessment of change in body fat percentage with DXA and eight-electrode BIA in centrally obese women
.
Med Sci Sports Exerc
 .
2007
;
39
:
2199
2203
.
37
Svantesson
U
Zander
M
Klingberg
S
, et al.
Body composition in male elite athletes, comparison of bioelectrical impedance spectroscopy with dual energy X-ray absorptiometry
.
J Negat Results Biomed
 .
2008
;
7
:
1
.
38
Clasey
JL
Bouchard
C
Teates
CD
, et al.
The use of anthropometric and dual-energy X-ray absorptiometry (DXA) measures to estimate total abdominal and abdominal visceral fat in men and women
.
Obes Res
 .
1999
;
7
:
256
264
.
39
Browning
LM
Mugridge
O
Dixon
AK
, et al.
Measuring abdominal adipose tissue: comparison of simpler methods with MRI
.
Obes Facts
 .
2011
;
4
:
9
15
.
40
Lohman
TG
Chen
Z
.
Dual X-ray absorptiometry
. In:
Heymsfield
SB
Lohman
TG
Zimain
W
, et al., eds.
Human Body Composition
 ,
2nd ed
.
Champaign, IL
:
Human Kinetics
;
2005
:
63
77
.
41
Plank
LD
.
Dual-energy X-ray absorptiometry and body composition
.
Curr Opin Clin Nutr Metab Care
 .
2005
;
8
:
305
309
.
42
Cordero-MacIntyre
ZR
Peters
W
Libanati
CR
, et al.
Reproducibility of DXA in obese women
.
J Clin Densitom
 .
2002
;
5
:
35
44
.
43
Genton
L
Karsegard
VL
Zawadynski
S
, et al.
Comparison of body weight and composition measured by two different dual energy X-ray absorptiometry devices and three acquisition modes in obese women
.
Clin Nutr
 .
2006
;
25
:
428
437
.