## Abstract

Dual-energy X-ray absorptiometry (DXA) is rapidly gaining acceptance as a reference method for analyzing body composition. An important and unresolved concern is whether and to what extent variation in soft tissue hydration causes errors in DXA fat estimates. The present study aim was to develop and validate a DXA physical hydration model and then to apply this model by simulating errors arising from hypothetical overhydration states. The DXA physical hydration model was developed by first linking biological substance elemental content with photon attenuation. The validated physical model was next extended to describe photon attenuation changes anticipated when predefined amounts of two known composition components are mixed, as would occur when overhydration develops. Two overhydration models were developed in the last phase of study, formulated on validated physical models, and error was simulated for fluid surfeit states. Results indicate that systematic errors in DXA percent fat arise with added fluids when fractional masses are varied as a percentage of combined fluid + soft tissue mass. Three independent determinants of error magnitude were established: elemental content of overhydration fluid, fraction of combined fluid + soft tissue as overhydration fluid, and initial soft tissue composition. Small but systematic and predictable errors in DXA soft tissue composition analysis thus can arise with fluid balance changes.

- body composition
- physical model
- fluid compartments

evaluating body composition is fundamental to the study of biological processes in animals and humans (7). Approaches developed over the past several decades now allow noninvasive assessment of over 30 distinct components in vivo (45).

Although there are many available body composition methods, only a few are sufficiently accurate for quantifying components in the research laboratory. A relatively new method, and one that also has clinical applicability, is dual-energy X-ray absorptiometry (DXA) (13, 16, 28). Whereas some methods are costly (4), require highly trained staff for their operation and implementation (23, 36), depend in part on subject participation (11), or expose human subjects to moderate radiation levels (4, 24, 36, 39), DXA systems are affordable, practical, require no active subject involvement, and impose minimal risk (42). Moreover, unlike most other body composition methods that are designed to quantitate a single whole body component (44), DXA permits quantification of multiple whole body and regional components, including bone mineral, fat, and lean soft tissue (10, 18, 33). As a result, DXA is gaining international acceptance as a body composition reference method (37).

Although the DXA method has many important features that contribute to its growing application in animal and human biological research, an important, incompletely resolved question is the influence of hydration on DXA soft tissue component estimates. Reports indicate that DXA makes no assumptions related to tissue hydration, and eight earlier reports in humans indicate that acutely altering fluid balance or distribution has no measurable influence on DXA or related dual-photon absorptiometry fat estimates (1, 8, 12, 20, 25, 26, 29, 47). However, other earlier reports suggest that physical principles and models on which DXA relies may be influenced by tissue hydration (8, 20, 33). The magnitude and clinical significance of these potential hydration effects have not been studied systematically. Because many experimental animal models and human subjects in whom DXA is applied have altered fluid balance, the unresolved question of DXA hydration effects is of major significance in the field of body composition research.

The aim of the present study was twofold: to develop and validate a DXA hydration physical model, and to use the developed model to quantitatively evaluate the various influences of fluid balance change on DXA soft tissue composition estimates.

## METHODS

### Study Design

The study aims were accomplished in a series of linked experiments. The goal was in vitro simulation of errors arising in DXA fat estimates as a result of changes in soft tissue hydration. Earlier in vivo human studies were incapable of producing systematic changes, particularly of large magnitude, in the fluid concentration of lean soft tissues. Simulation experiments, with the assumption that they represent in vivo effects, overcome the barriers posed by human experimentation.

Accordingly, the validity of the simulation model was established as follows. A DXA physical model was developed that links tissue elemental content with photon attenuation. The first experiment was designed to evaluate the validity of this physical model by use of a commercially available DXA scanner.

In the next stage, the physical model was extended to describe photon attenuation changes anticipated when predefined amounts of two components of known composition are thoroughly mixed, as would occur with overhydration of normal soft tissue. The validity of these models was tested in the second experiment by comparing the mixture’s expected photon attenuation with that measured by the DXA scanner. Once validated, the concepts advanced in the DXA physical model section then formed the basis of subsequent hydration error simulations.

Three stages then followed in developing hydration-error estimates. In the first stage, we derived fat and lean attenuation values for our DXA system by scanning beef phantoms of known composition. These “constants” are employed in DXA systems to predict soft tissue fat fraction from measured attenuation properties. The fat and lean attenuation values are required in the error simulation model. In the next stage, the attenuation values for two representative hydration fluids that differ in elemental composition were experimentally established with the DXA scanner. In the third and final stage of analysis, two new “overhydration” models were developed that allowed simulation of errors in DXA fat estimates that arise with systematic variation in three independent variables: the elemental content of excess fluid, the fractional amount of excess fluid, and the fat-to-lean composition of soft tissue.

### Physical Model

The DXA method assumes that nonosseous tissue consists of two distinct components, fat and lean soft tissue (33). The lean soft tissue component is the difference between body weight and the sum of fat and bone mineral ash. Fat and lean components are quantified over regions devoid of bone. The measured attenuation of DXA’s two main energy peaks is used to estimate each pixel’s fraction of fat and lean according to the following series of physical models (33).

A monoenergetic photon beam with incident intensity I_{0}passing across soft tissues is attenuated, and diminished beam intensity (I) is recorded in the detector (41). Fractional lowering of beam intensity is proportional to the substance’s linear attenuation coefficient (μ) and path length (*L*)
Equation 1 Integration of this equation results in the classical attenuation formula
Equation 2 As the linear attenuation coefficient is density (ρ) dependent, a convenient practice when working with tissues that differ in physical density is to calculate the mass attenuation coefficient (μ_{m}) as μ/ρ (34). Attenuation of monoenergetic photons using a substance’s mass attenuation coefficient can be calculated as
Equation 3 where*f i* is the mass fraction of the*i*th component as heterogeneous absorber. This equation indicates that photon attenuation in a heterogeneous absorber such as human soft tissues is a function of incident photon intensity and each component’s fractional mass, mass attenuation coefficient, and mass per unit area (*M*). As pixel area in DXA systems is constant and known, the mass per unit area represents total volume element or voxel mass.

*Equation 3
* can be rearranged, and attenuation can be expressed as the measurable transmitted-to-incident photon ratio
Equation 4For a mixture of *n* components, the mass attenuation coefficient for heterogeneous mixtures can be calculated from fractional mass and mass attenuation coefficient of components present as
Equation 5 An element’s mass attenuation coefficient is constant and known at any photon energy from classical experimental studies (21, 22, 35, 46).

Two photon energies are used with DXA systems, and these polyenergetic beams are generated either by pulsing the X-ray beam or by rare-earth filtration (14). The attenuation of each beam (i.e., I/I_{0}) by soft tissues can be measured as the ratio (R) of low to high energy calculated as
Equation 6
where*L* and*H* are low and high effective energies, respectively. Each element has a characteristic R value at specific energies, with increasing atomic number associated with higher R values (Table 1) (21, 22, 35, 46).

Elemental and complex absorber R values for DXA systems producing energies *L* and*H* can be calculated using*Eq. 6
* and known mass attenuation coefficients. For a two-component soft tissue mixture,*Eq. 6
* can be expressed as
Equation 7aA simplified R value formula can be derived, as in earlier reports (27, 28, 34), which assumes that μ_{m} values for the two soft tissue components at the higher energy are approximately equal (e.g., at 70 keV, μ_{m} values for protein, glycogen, water, extracellular fluid, and intracellular fluid are 0.183, 0.183, 0.194, 0.195, and 0.196, respectively)
Equation 7bIn the first experiment, we compared theoretical to measured R values for five simple compounds, including water, ethanol, glycine, alanine, and sucrose. Each compound was scanned with a DXA system three times, and the measured R value results were averaged. These results were then compared with theoretical R values calculated by using*Eqs. 6
* and *
7b
* with energy-specific R value data provided in Refs. 21 and 22. The equations gave equivalent results, and only data for *Eq. 7b
* are provided inresults. A similar protocol was then completed for four chemically more complex substances, including lean beef, lard, saline, and Ringer lactate solution. The composition of saline and Ringer lactate was established from manufacturers’ specifications. The elemental contents (H, C, N, O, P, Ca, Na, K, and Cl) of lean beef and lard were measured in triplicate on duplicate aliquots.

#### Error simulation.

Measured R values can be used to estimate the fractional (*f*) mass of each component in a two-component mixture. Because*f*
_{1} +*f*
_{2} = 1, then
Equation 8 and
Equation 9 where R is total R value consisting of two components with known R values R_{1} and R_{2}. *Equations8
* and *
9
* are often used in DXA systems for quantifying two-component mixtures (e.g., fat + lean) from measured R. These formulas will also be used in developing the overhydration models. In preparing a model that predicts the effects of hydration changes on fat estimates, we assume that*Eqs. 8
* and *
9
* are employed by the DXA system to estimate fat and lean fractions from measured soft tissue R.

The most common clinical situation is overhydration secondary to edema, ascites, and other forms of fluid accumulation, which can account for up to 20% of body weight. Dehydration is also common in clinically evaluated patients, but relative fluid loss is less than with overhydration. Because the model concepts are similar with all hydration changes, in the present report we develop only overhydration models. The outline of the model development that follows is presented in Fig. 1.

Assume the subject has normally hydrated soft tissue with known fat (*f*
_{FAT}) and lean (i.e., 1 −*f*
_{FAT}) fractions. The soft tissue R value can then be calculated using*Eq. 7b
* with known R_{FAT} and R_{LEAN} values. Overhydration then develops, leading to a three-component mixture of fat, lean, and added fluid (e.g., normal saline or water). The R value of the added fluid is known from measurement or is based on chemical analysis, and therefore the composite R value of the three-compartment mixture can be calculated using *Eq. 6
*. Alternatively, the overhydrated soft tissue can be considered as two new two-component models, soft tissue + added fluid and overhydrated lean (i.e., added fluid + lean) and fat (Fig. 1).

Under normal circumstances, the subject is scanned, and lean soft tissue and fat fractions are established using the standard system calibration. This, however, potentially results in an estimation error when additional fluid is present. We consider the magnitude of this error as the difference between actual fat fraction and that calculated using the standard DXA calibration equation, expressed in percent fat units. Predicted fat fractions that are less than actual values will have a negative sign and vice versa. All terms required for this analysis have algebraic solutions, with the assumption that initial fat fraction, the fraction of soft tissue as added fluid, and the R values for fat, lean, and fluid are known. The main clinical condition of interest is when variable amounts of fluid expand normal soft tissue (*model 1* in Fig. 1). Another possibility is when fluid expansion occurs and remains stable, producing an overhydrated lean compartment, while the proportion of soft tissue fat varies with energy balance.

The hydration model assumes that mixtures of two components follow predictable rules with respect to R value changes. That is, when fluid of known R value is added to soft tissue, the result is a predictable new R value reflecting the overhydrated state.

In the hydration experiment, we first established the validity of*Eq. 7b
* in predicting the R values for various mixtures of two components, each with a known R value. Mixtures of two of the substances evaluated in the first experiment (wt/wt) were made as follows: ethanol-water, 25:75, 50:50, 75:25; normal (i.e., 0.9%) saline-water, 25:75, 50:50; Ringer lactate-water, 25:75, 50:50, 75:25; lean beef-normal saline, 95:5, 90:10, 80:20; and lean beef-Ringer, 95:5, 90:10, 80:20.

We then experimentally established the R_{FAT} and R_{LEAN} values for our DXA system. This allowed development of fat fraction prediction equations based on*Eq. 8
*. Five beef phantoms of varying fat content were scanned, and R values were measured. The fat fraction vs. R value regression line was developed and then solved for fat fractions of 1 and 0 for estimating R_{FAT} and R_{LEAN}, respectively. Beef fat content was measured by lipid solvent extraction.

In the last stage of analysis, the soft tissue component R values and fat prediction equation were used to estimate errors arising with hypothetical levels of soft tissue overhydration. Water and normal saline were used as the excess fluids, and their R values were measured in triplicate as noted below. Hypothetical mixtures of beef and water or normal saline were then created, and R values were calculated using*Eq. 8
*, as described above.

### Sample Preparation

Water was triple distilled and ethanol was US Pharmacopeia Grade (Pharmco, Brookfield, CT). Glycine and alanine were >99% pure by thin-layer chromatography (Sigma Chemical, St. Louis, MO). Sucrose was obtained from the hospital pharmacy.

Lean beef and lard were obtained from a local market. Normal saline had, per 100 ml, 0.9 g NaCl as defined by the manufacturer (McGaw, Kendall, Ontario, Canada). Ringer lactate solution, according to the manufacturer (McGaw, Kendall, Ontario, Canada), had the following composition per 100 ml: sodium chloride, 0.60 g; sodium lactate, 0.31 g; potassium chloride, 0.030 g; and calcium chloride, 0.020 g.

### Dual-Energy Measurements

A Lunar DPX-L (Lunar, Madison, WI) pencil beam system with version 1.3z software was used for all studies. The DPX-L system has a 76.0 kVp X-ray source, and cerium k-edge filtration is used to generate two photon peaks with main energies at 40 keV and 70 keV (27). All substances were scanned in a 4-liter plastic container by use of the fast scan mode. The height of material in the container was kept constant for all scans at 10 cm. R values were obtained using a region of interest that excluded container edges and base. All scans were run in triplicate, and the R value results were averaged.

An assumption of the present study is that the DXA system used to evaluate various biological substances is representative of DXA systems as a whole. Other systems vary in mode of photon generation, use of various filters, and peak photon energies, although underlying physical concepts are similar for all DXA methodologies.

### Chemical Methods

Chemical composition of lean beef and lard was analyzed on two sample aliquots in triplicate. Fat was measured by the method of Folch et al. (6). Hydrogen, carbon, and nitrogen were determined by combustion at 1,050°C in a constant O_{2}stream with infrared absorption for CO_{2} and H_{2}O and thermal conductivity for N_{2} (Leco CHN 1000, St. Joseph, MI) (2, 5, 32). Oxygen was measured by pyrolysis in a N_{2} stream at 1,200°C, with detection of resulting CO and CO_{2}by separate infrared detectors (Leco RO-478, St. Joseph, MI) (30). Sodium, magnesium, phosphorus, potassium, and calcium were detected by inductively coupled plasma emission spectroscopy, with a detection limit of 0.01 parts/million (ppm) for Na, 0.85 μg/ml for Mg, 0.1 ppm for P and K, and 3.14 μg/l for Ca (43). Sulfur was determined with a quantitation limit of 0.08 mg by combustion in atmospheric O_{2} at 1,350°C and analysis of resulting SO_{2} by infrared detector. Chlorine was detected with a quantification limit of 10 ppm by ion chromatography, peak area ratio, and conductimetric detection (38).

### Statistical Methods

The developed hydration model is based on the assumption that R values follow physical rules related to photon attenuation and biological substance component proportions. The conceptual basis of these models was validated by comparison of theoretically derived with actually measured DXA R values. Simple linear regression analysis and means ± SD were used as the basis of these analyses. Our analysis was designed to explore qualitative R value associations, and exact theoretical-to-measured R value agreement was not needed for establishing hydration model validity. Additional tests (e.g., Bland Altman analyses) exploring the statistical significance of R value differences were therefore not carried out.

Simple linear regression analysis was used to develop a DXA fat fraction prediction formula based on R values as the independent variable. Two of the samples were the lean beef and lard. Lean beef and lard were then thoroughly mixed to form three additional mixtures of varying fat fraction. Fat content was then measured by the method of Folch et al. (6), as noted in *Chemical Methods*.

All analyses were carried out using the statistical program SAS Release 6.10 (Statistical Analysis System, SAS Institute, Cary, NC).

## RESULTS

### Physical Model

#### Theoretical vs. measured R values.

The theoretical R values and their measured counterparts are summarized in Table 2. The group mean theoretical (theor) and measured (meas) R values were similar (1.314 ± 0.055 vs. 1.322 ± 0.061), and the two R value estimates were highly correlated (Fig. 2; theor R = 0.896 × meas R + 0.129,*r*
^{2} = 0.956, SEE = −0.013, *P* < 0.0001). These observations support the validity of the overall theoretical DXA model presented in methods.

### Hydration Effects

#### Two-component mixture analysis.

The R value results for two-component mixtures are presented in Table3. Four sets of R values are provided along with each mixture’s elemental content. The R_{1} column indicates the theoretical R value calculated using *Eq.6
*, which is based on elemental composition.

This computation is identical to the one used earlier for estimating theoretical R values of the nine biological substances presented in Table 2. The R_{2} column is the R value calculated using *Eq. 7a
*, with the assumption of a two-component mixture with known theoretical R values for each component, as presented in Table 2. The R_{3} column gives the R value calculated using *Eq. 7a
* with measured R values for each of the two components as presented in Table 2. The actual measured R value for the composite mixtures is given in the fourth column.

As for the primary substances evaluated in Table 2, the theoretical R values based on elemental composition for the two-component mixtures were in close agreement with the corresponding measured R values (1.347 ± 0.026 vs. 1.365 ± 0.029; theor R = 0.841 × meas R + 0.198, *r*
^{2} = 0.910, SEE = −0.008, *P* < 0.0001).

The two calculated R values based on *Eq.7a
*, one with theoretical component R values and the other with measured component R values, were in good agreement with actual measured R values [1.346 ± 0.026 and 1.362 ± 0.028 vs. 1.365 ± 0.029, respectively; (Fig.3) theor 2 component R = 0.840 × meas R + 0.200, *r*
^{2} = 0.910, SEE = −0.008, *P* < 0.0001 and theor 2 component R = 0.976 × meas R − 0.031,*r*
^{2} = 0.994, SEE = −0.002, *P* < 0.0001].

Our analysis up to this point indicates a close concordance between*1*) R values calculated with physical constants and measured R values and*2*) R values calculated with the assumption of a mixture of two biological substances, each with a known R value, and measured R values. The next stage of analysis involved error simulation with changes in soft tissue hydration.

#### DXA fat fraction prediction equation.

The fat fraction coefficients of variation (CV) within and between aliquots for the five beef-lard mixtures were 1.52 and 1.54%, respectively. The between-measurement DXA R value CV for the five phantoms was 0.2%. Measured R values for the five beef-lard mixtures were highly correlated with fraction of soft tissue as fat:*f*
_{fat} = 9.418 − 6.815 × R,*r*
^{2} = 0.998, SEE = 0.09, and *P* < 0.001 (Fig.4). Solving this equation for fat and lean R values (i.e.,*f*
_{fat} = 1.0 and 0) gives R_{FAT} and R_{LEAN} of 1.235 and 1.382, respectively.

#### Error simulations.

The respective measured R values for H_{2}O and 0.9% NaCl were 1.363 ± 0.001 and 1.399 ± 0.001 (Table 2), respectively. Four levels of hydration were considered as a fraction of combined fluid plus soft tissue mass, 0.01, 0.05, 0.10, and 0.20. A base soft tissue composition was selected arbitrarily as 25% fat and 75% lean. If we assume R_{FAT} and R_{LEAN} values of 1.235 and 1.382, respectively, this soft tissue mixture has an R value of 1.345 based on*Eq. 7a
*.

With *model 1*, as normal saline is added, the actual fat fraction progressively declines and the soft tissue R value increases, as predicted with *Eq.7a
*. The error arising when the standard DXA fat fraction model (i.e.,*f*
_{fat} = 9.418 − 6.815 × measured R) is applied to the overhydrated soft tissue is shown in Fig. 5. The error is relatively small, predicted 0.11% less than actual fat fraction with soft tissue fluid fraction of 0.01, and rises to −2.3% with fluid fraction of 0.20. An error of smaller magnitude and opposite in sign is produced by adding water to the soft tissue mixture (added fluid fraction of 0.01 = 0.13% and 0.20 = 2.6%). Error is at a minimum (i.e., 0%) when no fluid is added and increases to a maximum (e.g., for 0.9% NaCl = −11.6%) when added fluid fraction = 1.

With *model 2*, we begin again with a base soft tissue composition of 25 and 75% fat and lean fractions, respectively. The soft tissue is then overhydrated to a level of 0.10 with 0.9% NaCl. The soft tissue R value increases from 1.345 to 1.351 with overhydration, and the lean R value increases from 1.382 to 1.384. The overhydrated lean R value of 1.384 for 0.10 fluid fraction is now maintained constant, and the fat fraction is then varied as might be found in representative men (0.05, 0.15) and women (0.25, 0.35). The simulated error increases from a minimum of −0.97% at fat fraction = 0.35 to a maximum of −1.42% at fat fraction = 0.05 (Fig. 6). A trend opposite in sign is observed when water is the overhydration fluid. Error in this model is at a maximum (e.g., for 0.9% NaCl = −1.5%) when fat fraction is zero and is at a minimum (i.e., 0%) when fat fraction is 1.

## DISCUSSION

Unlike previous investigations (1, 8, 12, 20, 25, 26, 29, 47), the present study results clearly demonstrate a bias error in DXA soft tissue composition estimates as a function of tissue hydration. Specifically, the present study results indicate the existence of a small magnitude bias error in fat fraction with changes in soft tissue hydration. Our findings are supported at two levels, one theoretical and the other experimental.

### Theoretical Error Basis

A theoretical basis strongly supports the hypothesis that DXA fat estimation errors occur secondary to soft tissue hydration changes. The underlying concept of DXA soft tissue composition analysis is that fat and lean have different but stable elemental proportions (14). Fat, which is mainly triglyceride, consists of stable proportions of the three relatively low-R elements H, C, and O (3, 15). Lean soft tissue also has H, C, and O, with additional amounts of higher-R elements such as Na, K, Cl, Ca, S, Mg, and Fe (3, 17, 19, 40). Although not a distinct chemical moiety as is triglyceride, lean composition and elemental proportions are by necessity relatively stable as part of overall homeostasis maintenance. Accordingly, DXA soft tissue analysis is founded on a two-component model in which fat and lean are assumed to have constant and different R values. A linear association between soft tissue fat fraction and R value can be demonstrated, leading to the potential for composition prediction equation development.

The theoretical links developed in the present report unequivocally show that, with changes in soft tissue hydration, there occurs alteration in lean tissue elemental proportions and R value. Addition of a high-R fluid such as normal saline would by necessity increase lean R in relation to the relative amount added. Lower-R hydration fluids, such as water or ethanol, would cause corresponding reductions in lean R values. Any change in the assumed constant lean R value would lead to soft tissue composition estimation errors.

### Experimental Error Basis

The experimental focus of the present investigation was a systematic attempt to support each step of the theoretical model development. We first showed that our DXA system provides measured R values in very close agreement with theoretically derived R values for substances ranging from simple molecules to complex animal tissues. In the next step, we demonstrated that various two-component mixtures produce R values also very similar to predicted R values based solely on elemental content of each component or the component’s actual measured R value. These first two experiments provide strong support for developing models that predict changes in soft tissue and lean R values with addition of various amounts of known R value fluids.

In the hydration study phase, we went on to confirm the previously reported (33) strong association between measured R value and fraction of soft tissue as fat, an experiment that allowed us to develop R value constants for fat and lean components. Last, these fat and lean R values for our DXA system were used to develop two different soft tissue hydration models that allowed error simulation.

### DXA Hydration Error

Results clearly show that DXA fat estimation errors occur as a function of added fluid R value, fraction of added fluid, and soft tissue composition. Overall, DXA fat estimation errors increased with larger deviations between assumed lean R and added fluid R. In effect, if the accumulating fluid has the same R value as lean and there is no R value difference, no fat estimation errors occur no matter how much extra fluid is added. If the hydration fluid and lean R values do differ, then greater relative fluid accumulation is associated with larger fat estimation errors. Finally, if we assume that overhydration results in a higher and stable new lean R, fat estimation error diminishes as the fat fraction increases.

Although the present study demonstrates a potential fat estimation bias error with soft tissue hydration changes, the magnitude of these errors is small when considered in the context of the physiological range of accumulated fluid compatible with life. Hydration, including fluid and electrolyte balance, is maintained remarkably stable in health (9). Simulated experiments suggest DXA fat errors of <1% with hydration changes of 1–5% (Figs. 5 and 6). Such changes are below the usual detection threshold of DXA systems unless subject samples are large and measurements are repeated and averaged to minimize error. The between-measurement technical error for most DXA systems is ∼1% fat units. The possibility does exist, however, for fat estimation errors in the range of several percent when soft tissue overhydration is severe, perhaps in the range of 20–25% of total soft tissue mass. Profound overhydration in this range is not common clinically, although regional DXA fat estimates might be affected by large local accumulations such as occur with pedal edema or ascites. On the other hand, some small between-subject variability in DXA fat estimation accuracy can be anticipated as hydration varies with fluid intake and other physiological processes throughout the day and with varying conditions such as prior exercise and season of the year.

Although our in vitro modeling experiments afford distinct advantages over more-complex-to-interpret in vivo studies, there are some limitations of the present study approach that should be recognized. First, added fluids such as water and normal saline may reflect conditions that are not physiological. Second, our selected models assumed stable volumes with changing fluid and fat fractions. Other model variations are possible, and these should be considered in future studies. Last, our hypothetical soft tissue system was devoid of bone minerals and leaves unanswered the magnitude and importance of errors arising in the broader context of in vivo pixel analysis.

### Previous Study Linkage

The present study results should be considered in the context of earlier DXA hydration investigations. As a representative example, Horber et al. (20) investigated the effects of water ingestion (0.8–2.4 liters) and hemodialysis in six healthy subjects and seven patients with chronic renal failure, respectively. Changes in fluid balance did not significantly affect either DXA fat or bone mineral mass estimates. Chronic renal failure per se results in fluid accumulation and presumably also alters the lean soft tissue R value. Errors in fat estimation, although small in magnitude, may already thus be present even before initiation of hemodialysis. Ingested water and fluid removed with hemodialysis would have different R values and would, accordingly, be expected to have different effects on DXA fat estimation errors. With respect to water ingestion, in healthy subjects, water is cleared rapidly by renal mechanisms, and it appears within a short time interval in the bladder. The pubic bone is anterior to the bladder, and DXA cannot directly evaluate soft tissue R values in pixels that also contain bone. Water ingestion as part of a hydration experiment may therefore be associated with errors caused by DXA technical limitations other than those related to fluid effects. Finally, the small magnitude of fluid change and the sample of less than ten subjects in each experiment make it unlikely that Horber et al. would have observed a statistically significant DXA fat estimation error with hydration change. Similar DXA experiments with relatively small fluid balance changes were carried out by Formica et al. (8), Going et al. (12), Abrahamsen et al. (1), and Woodrow et al. (47). Errors in fat estimation in these experiments predicted by our simulation model would be extremely small in magnitude and at the margin of detection with the evaluated samples.

A second and relevant example is provided by the study of Lands et al. (26). DXA (70 keV/140 keV system) was used to evaluate body composition in six healthy men, after which normal saline was administered by intravenous infusion. On repeat scan, the men had gained a mean of 2.2 kg by scale weight and 2.4 kg by DXA weight. Of the 2.4 kg DXA weight change, 2.9 kg was lean soft tissue, which indicates negative fat balance (∼0.5 kg). Our hydration model for 40 keV/70 keV similarly predicts overestimation of lean gain and corresponding underestimation of fat mass change with high-R fluid administration. For example, if normal saline were scanned in our system, the resulting fat percentage would be −11.6%. This occurs because the R value for normal saline (1.399) is higher than that for the assumed lean tissue R (e.g., 1.372 as established in the present study). Again, however, we emphasize that such small changes as evaluated by Lands et al. (26) may be within the between-measurement error of most DXA systems, and large subject groups would be needed to detect the bias error with confidence.

### Hydration Error in Other Methods

Other body composition methods are also based on two-component models and are prone to fat estimation errors as well. For example, the classic two-compartment total body water method assumes a constant fat-free body mass hydration of 0.732 (31). Error for the total body water method can also be simulated if we assume for theoretical purposes that fat-free body mass and lean soft tissue mass are equivalent compartments. Using 0.25 fat and 0.75 lean soft tissue fractions, water fraction of total soft tissue is 0.732 × 0.75 = 0.549. If total soft tissue mass is now increased by 10% with water addition, actual fat fraction decreases to 0.225 and total water fraction increases to 0.594. Assuming a lean soft tissue hydration level of 0.732, one arrives at a fat fraction of 0.188, an error of −0.036 or −3.6% fat. The corresponding DXA error, as shown in Fig. 6, was 1.16%, which is smaller in magnitude than the corresponding two-compartment total body water method error.

### Potential Error Correction

Overall, the simulated fat estimation error is small in magnitude in the context of normal, or even pathological, variation in hydration. Nevertheless, an important question is whether or not hydration errors are correctable. One approach available on DXA systems that provide R value information is to measure the actual R value of evacuated fluid (e.g., with paracentesis) and then correct the measured whole body soft tissue R value accordingly. Large fluid evacuations such as with ascites removal might be studied in this manner.

### Conclusion

The present investigation provides strong evidence that the widely used body composition method, DXA, is prone to fat estimation errors related to variation in soft tissue hydration. The magnitude of this potential error source is a function of two main independent variables, the fractional amount and the elemental content of the lost or gained fluid. Under normal or even most clinical conditions, the anticipated magnitude of this error is small and should not pose any substantial limitations to the accuracy of the DXA technique.

## Acknowledgments

The authors gratefully acknowledge Dr. Carol Boozer and the Galbraith Laboratories in Knoxville, TN, for assistance in analyzing chemical composition of biological materials and Dr. Oksana Duda and Martha Paszek for technical contributions in carrying out the DXA studies.

## Footnotes

Address for reprint requests: A. Pietrobelli, Weight Control Unit, Obesity Research Center, 1090 Amsterdam Ave, Floor 14, New York, NY 10025.

This study was supported by National Institute of Diabetes and Digestive and Kidney Diseases Grant RO1-DK-42618.

- Copyright © 1998 the American Physiological Society