Medicine and science require precise, reliable techniques to assess total body water (TBW) and fluid compartment volumes. TBW has been measured reliably by using the isotopes of tritium, hydrogen, and deuterium (i.e.,3 H2O, 2H2O, H2 18O;ref. 19). Historically, extracellular fluid volume (ECV) was measured with thiocyanate and isotopes of sodium and chloride (i.e.,24 Na, 22Na, 36Cl), but in recent years bromide has been the most widely used tracer in research studies(12,15,16). A corrected bromide space can be calculated from the serum bromide concentration after administration of a known quantity of bromide; this is an excellent approximation of ECV(10).
During the 1980s, single frequency bioelectrical impedance analysis became a widely-used technique for measuring TBW(12,17). Bioelectrical impedance analysis is based on the assumption that electricity (i.e., a low level alternating current, single 50 kHz frequency) is conducted poorly by fat and bone but is conducted well by tissues that contain predominantly water and electrolytes. Single-frequency bioelectrical impedance values predict TBW well (r = -0.86,P < 0.01; ref. 13) when compared with isotope dilution techniques and have a coefficient of variation for repeated measurements of 2 to 3.4% (5,13). Today, single-frequency bioelectrical impedance analyzers are being replaced by multiple-frequency analyzers because the latter instruments offer the potential of differentiating between TBW, ECV, and ICV.
Although it has been postulated(7,9,21,24,25) that low- and high-frequency impedance measurements can be used to distinguish ECV from intracellular volume (ICV), no scientific consensus has been reached(8,14,17,18,28,29). Also, because instruments today cannot measure impedance reliably at very low(<1 kHz) and very high (>500 kHz) frequencies(8,29), statistical modelling techniques have been used. Bioimpedance spectroscopy (BIS) is a statistical modeling technique(7,9,29) that extrapolates values for resistances at very high and very low frequencies from resistance values in the frequency range that is reliable (i.e., 1-500 kHz), then calculates ECV, ICV, and TBW (29). However, three limitations of the BIS technique are relevant to this investigation. First, BIS in humans is likely to be influenced by differences in body composition(17,29). Second, the following factors are known to affect impedance measurements: electrode placement, side of the body, posture, limb position, composition of the examination surface, ambient temperature, hydration status, food and liquid intake, plasma osmolality, and recency of exercise (4,17,23). Third, BIS is linked with TBW, ECV, and ICV by theory(7,8,12,14), not proven biophysical principles (17). These limitations emphasize the need for validation studies.
Because BIS methods and commercial instruments have been developed only during the present decade, the validities of BIS modeling techniques have not been firmly established (23). Therefore, the purpose of this investigation was to test the validity of a technique that provides TBW, ECV, and ICV estimates from BIS analyses. This was accomplished by using deuterium oxide and sodium bromide dilution techniques as reference measures. We hypothesized that volumes derived via BIS would be statistically different from those derived with tracer dilution.
Thirteen healthy male college students (Table 1) gave their written, informed consent to participate in this study after all risks and procedures had been explained. The test protocol was approved by the Institutional Review Board at the University of Connecticut and complied with the policy statement regarding the use of human subjects as published byMedicine and Science in Sports and Exercise. The percentage of body fat of each subject was determined by hydrostatic weighing(20) and a determination of residual lung volume with a bell spirometer. Body density was used to calculate fat free mass with the Siri formula (22). Height was measured to the nearest 0.5 cm and body mass was recorded on a calibrated electronic floor scale(± 50 g; SR Instruments, model 700 M, Tonawanda, NY).
Fluid and tracer analyses. Subjects arrived at the laboratory after fasting for at least 8 h. Their fluid intake had been increased during the previous 24 h by consuming > 1300 mL of fluid beyond ad libitum intake. Their hydration status was verified with measurements of urine specific gravity (refractometer, Spartan Co. model A300CL, Japan) and triplicate analyses of plasma osmolality (osmometer, Advanced Instruments model 3MO, Needham Heights, MA). Subjects were considered to be normally hydrated when plasma osmolality was between 283-290 mOsm·kg-1 and urine specific gravity was between 1.017-1.029 (2). In addition, pretest plasma sodium and potassium (ion-sensitive electrodes, AVL Scientific Corp., model 984-S, Roswell, GA) and hematocrit(microhematocrit technique) were measured in triplicate to verify that these values fell within normal ranges (1).
Baseline urine and blood samples were obtained after 45 min of lying supine, commencing between 0700 and 0715 h. Blood (15-20 mL) was drawn from a forearm vein before the initial BIS measurement. Subjects next arose, voided their bladder, and ingested two tracers simultaneously in a solution containing 1.70 g of bromide for ECV determination (NaBr, certified ACS; Fisher Scientific, Pittsburgh, PA) and 7.27 g deuterium oxide for TBW analysis(D2O, 99.9% enriched; Cambridge Isotope Laboratories, Andover, MA), between 0800 and 0835 h. This was followed by ingestion of 100 mL of distilled deionized water from the same container. Subjects then rested for 4 h in a supine position while tracers equilibrated in body fluids. Deuterium and bromide dilutions were determined in blood and urine samples that were collected 4 h after dosing; blood samples were centrifuged and frozen at -80°C for subsequent analysis. Urine volume (± 0.1 g) and tracer concentrations were used to correct for tracer excretion during this period.
Isotope abundances in blood and urine were determined with an isotope-ratio mass spectrometer (Nuclide Co., model 3-60, Bellefonte, PA) following the method of Wong et al. (30). The coefficient of variation of this technique is 1%. TBW was calculated from the dilution of isotopic water (20): Equation where d is the ingested dose of deuterium minus the amount of deuterium excreted in urine(g); MW is the molecular weight of deuterium (20.02); APE is the atom percentage enrichment of deuterium in the consumed dose (99.9%); 18.02 is the molecular weight of water; Rstd is the ratio of 2H/1H in the VSMOW water standard (0.00015576); Δδ D is the arithmetic difference of deuterium in the plasma before and after isotope administration(parts per thousand). This TBW value was divided by 1.04, to correct for the nonexchangeable deuterium which binds to protein(20).
Bromide analyses were performed in triplicate by a high-performance liquid chromatography system (Shimadzu Corp., Columbia, MD) equipped with an ultraviolet detector and a Partisil column (Alltech Associates, Inc., Deerfield, IL), following the method of Miller et al.(16). The mobile phase contained 6 g of potassium dihydrogen phosphate dissolved in 2 L of water. The column retention time for bromide was 17.5 to 19 min. The bromide concentration was determined by comparing the area under the curve with those of known serum and urine standard curves (acceptable at r = 0.999). The coefficients of variation for this technique are 1-2% (within-in run) and 3% (between-day)(15). The corrected bromide space (CBS) was calculated from the amount of bromide consumed (Br dose), the amount of bromide excreted in urine (U), and the change in plasma bromide concentration ([ΔBr plasma]) (16): Equation where 0.90 is the correction factor for the distribution of bromide in nonextracellular sites, principally red blood cells; 0.95 is the correction factor for the Donnan equilibrium; and 0.94 is the correction factor for the water in serum. Intracellular volume (ICV) was calculated as the difference between TBW and ECV (ICV = TBW - ECV).
Bioimpedance. A multiple frequency bioimpedance analyzer (Xitron Technologies, model 4000B, San Diego, CA) was employed immediately before blood sample collections. Subjects rested in a supine position for 4 h on a nonconductive mattress with arms and legs abducted 10 ° while wearing shorts, a T shirt, and no jewelry. Each of the four instrument cables was individually shielded to minimize the effect of cable placement. Before each test, the analyzer was self calibrated using a 422-ohm test resistor supplied by the manufacturer. A tetrapolar arrangement of gel electrodes (2.0 × 7.5 cm) was applied to the skin after alcohol preparation per the manufacturer's suggestions, and skin hair was removed with an electric shaver. Specifically, electrodes were placed (a) on the dorsal surface of the left foot 2 cm proximal from the metatarsophalangeal joint of the third toe, (b) on the anteriodorsal surface of the left ankle over the axis of the medial and lateral malleoli, (c) on the dorsal surface of the left hand 1 cm proximal from the metacarpophalangeal joint of the third finger, and (d) on the dorsal surface of the left wrist 7.5 cm proximal from hand electrode c.
A maximum measurement current of 250 μA was applied to the skin, immediately before the collection of each blood sample (see above). Measurements of resistance, reactance, impedance, and phase angle were recorded at 48 frequencies ranging from 50 to 500 kHz, and data files were analyzed using an iterative, nonlinear, curve-fitting algorithm (Xitron Technologies). This software fit each impedance and phase spectrum (50-500 kHz) according to the Cole-Cole (7,14) model of reactance as a function of resistance and removed any frequency that significantly decreased the total weighted least-square error for fit to that model. This modeling procedure then computed the extrapolated values for resistances at zero frequency (R0) and the infinite frequency(R∞). The software for the Cole-Cole modelling and calculations of R0 and R∞ were supplied by Xitron Technologies. The estimated R0 and R∞ were then substituted into equations derived by Hanai (8,11). Estimated ECV was calculated asEquation and estimated ICV was determined by an iterative process using the equation where Recf is the resistance because of the extracellular fluid, Ricf is the resistance because of the intracellular fluid, W is the subject mass (kg), L = subject height (cm), Kecf = 0.337, and Kp = 2.905; the constants Kecf and Kp were supplied by Xitron Technologies. R0 and R∞ are the extrapolated resistances from the Cole-Cole model corresponding to Recf and Ricf, respectively. The computer algorithm for the Hanai equations was not supplied by Xitron but was separately developed in the form of a spreadsheet. TBW was calculated as the sum of ECV plus ICV.
All tests were conducted in an environmental chamber maintained at a dry-bulb air temperature of 22 ± 1 °C. Several factors that alter impedance were controlled in this investigation, including electrode placement, side of the body, posture, limb position, nonconducting examination surface, ambient temperature, hydration status, food and liquid intake, plasma osmolality, and recency of exercise(4,17,23). Skin temperatures at the chest, forearm, and lateral calf were recorded with an infrared temperature sensor(Exergen Corp., Ototemp model 3000, Newton, MA) immediately after BIS were recorded, and mean weighted skin temperature was calculated by using the formula of Burton (6).
Statistical analyses. Linear regression analyses provided correlation coefficients (r) between TBW, ECV, and ICV, as derived from BIS measurements and the corresponding dilution techniques. Pearson product moment correlations (r) were computed for the relationships between body composition variables and the differences between BIS and tracer dilution volumes. These differences were related to body composition. To determine if mean BIS volumes for TBW, ECV, and ICV differed from mean dilution volumes, t-tests for dependent samples were used without alpha level correction. The reliability of BIS measurements was assessed by making repeated determinations on the same day. Regarding the size of the sample (N = 13), a retrospective power analysis (Welch approximation test) found a small effect size (statistical power <0.20 for TBW, ECV, and ICV). All statistics were computed with a commercial software data management package (Statistica, StatSoft, Inc., Tulsa, OK) and were reported as mean ± SD.
The thirteen test subjects were young nonobese adults who had no signs, symptoms, or reports of disease, and who had been purposefully selected to allow measurements over a wide range of body compositions. The six subject characteristics shown in Table 1 were normally distributed; this was verified by calculations of skewness (range: -0.12 to 0.73), kurtosis (range: -1.55 to 0.03), and the Shapiro-Wilks test of frequency distribution (all P > 0.05). Plasma osmolalities (286± 4 mOsm·kg-1) and urine specific gravities (1.018± 0.006) indicated that subjects were euhydrated at the onset of testing (2). The following baseline measurements also fell within clinically accepted ranges for males (1), indicating that subjects were normally hydrated before testing: hematocrit, 45± 3%; plasma sodium, 145 ± 1 mEq·L-1; and plasma potassium, 4.4 ± 0.2 mEq·L-1.
The mean weighted skin temperature of these subjects did not change significantly from baseline (32.7 ± 0.6 °C) to the 4-h point of testing (32.6 ± 0.7 °C) because of the controlled 22 °C environment. Body mass at the beginning of testing (80.3 ± 15.3 kg) was significantly different (P < 0.00005) from the body mass at the end of daily testing (79.7 ± 15.2 kg). This loss of 0.6 kg was attributed to insensible water evaporation during the 4-h test period (0.15 kg·h-1) because no food or fluid was ingested. The mean urine production was 322 ± 235 mL·4 h-1.
The mean ± SD values for the dilution and BIS methods, plus the between-method differences (Δ), appear in Table 2. Despite the fact that large differences existed for some subjects, the linear regression comparisons of dilution techniques versus BIS(Fig. 1) were strongly correlated (r = 0.93 to 0.96, allP < 0.001). The SEEs for these regression analyses were 2.23 L for TBW, 1.26 L for ECV, and 1.71 L for ICV; these were 4.4%, 6.3%, and 5.5%, respectively, of the dilution volumes for TBW, ECV, and ICV.
Linear regression comparisons of dilution techniques and BIS estimates appear in Figure 1, in the determination of TBW (panel A), ECV (panel B), and ICV (panel C). The test-retest reliabilities of BIS, assessed by intraclass r values, were very strong (TBW, r = 0.98; ECV, r = 0.98; ICV, r = 0.94).
Table 3 presents the percentage of body mass, fat free mass, and TBW that were a result of fluid volumes for both dilution and BIS methods. Despite differences between techniques (see * and ** symbols), all values fell within the range of previously published multiple-frequency bioimpedance and BIS studies(9,18,21,26,29).
An analysis of the differences between dilution measures and BIS estimates is presented in Table 4. These differences (see Δ columns in Table 2) were correlated to body composition. As fat free mass and body mass index increased, the Δ of TBW and theΔ of ICV (but not ECV) became increasingly negative (r = -0.56 to -0.71,P < 0.05) and BIS increasingly underestimated dilution measurements.
Three primary limitations of the BIS technique were described in the introductory section. The first, involving body composition, was evaluated by correlating fat-free mass, body mass index, and percent body fat values with BIS fluid compartment volumes (see below). The second limitation, involving the factors known to influence bioimpedance, was addressed by controlling these factors during testing. The third, involving BIS theory, emphasizes the need for the present validation study.
Measurements of TBW by dilution and BIS were not statistically different. The SEEs determined in regression analyses were all within the range of those reported by other investigators (1.5-3.3 L)(18,25,27,29) and reflected differences that ranged from -5.40 to +2.87 L (Table 2). Individual TBW differences (and those of ICV, see Table 2) were sometimes large. In contrast, the mean ECV derived from bromide dilution was significantly different from the BIS value for ECV (P< 0.01), reflecting the fact that BIS overestimated bromide volumes in 12 of 13 subjects; a similar finding was reported by Patel et al.(18). The SEE (1.26 L) for our ECV comparison was similar to errors reported in previous studies(18,25,27,29), amounting to 6.3% of the mean ECV and reflecting a range of differences from -2.23 to +2.53 L(Table 2). Because dilutional ICV was calculated as the difference between TBW and ECV, the characteristics of ICV were influenced by the deuterium oxide and bromide volumes. The ICV means for BIS and dilution techniques (Table 2) were different by 2.08 ± 3.08 L (P < 0.05).
Table 4 demonstrates that the differences between dilution measures and BIS estimates (see Δ columns inTable 2) were related to body composition. To our knowledge, this has not been reported previously. As fat-free mass and body mass index (but not body fat%) increased, the Δ of TBW and the Δ of ICV (but not ECV) became increasingly negative (r = -0.56 to -0.71,P < 0.05) and BIS increasingly underestimated dilution measurements. This Δ ICV effect probably was influenced by the calculation of ICV using TBW (ICV = TBW - ECV) in the dilution method. It is likely that percent body fat did not exhibit this relationship because fat does not conduct electricity well, regardless of the total fat mass. Also, it is possible that one of the modelling components played a role in this phenomenon. Specifically, the Hanai body tissue mixing theory(8,11,14) calculates TBW, ECV, and ICV with fixed coefficients for ECV and ICV resistivity (i.e., Kecf = 0.337 and Kp = 2.905, see Methods) that may not apply to all human body types (25). Future adjustments of these coefficients, to account for intrasubject differences in fat-free mass or body mass index, may improve the accuracy of BIS estimates.
Although acute fluid shifts and illness generally reduce the accuracy of impedance methods (3,17), the accuracy of BIS(Table 2) will likely be acceptable in clinical settings, especially if the change in ECV/ICV ratio (Table 3) is the primary focus (14,21) and not the absolute volumes. Clinicians should recognize that several factors may affect inpedance measurements (see introductory section) and attempt to use a consistent protocol. Nevertheless, in clinical settings it is likely that error/variance terms will be larger than in the present controlled experiment. Because the BIS technique correlates strongly with dilutional TBW, ECV, and ICV(P < 0.0001, Fig. 1), and because BIS exhibited a strong test-retest reliability (r = 0.94 to 0.99), it is valid for between-subject research comparisons of body fluid compartments. However, the variance observed in our data (Table 2) will likely make BIS inappropriate for measurements of relatively small changes in fluid volumes (i.e., < 1 L) in response to experimental treatments. Unfortunately, our findings do not allow us to formulate an unequivocal recommendation regarding the validity of BIS to all research protocols. Investigators should evaluate the appropriateness of using BIS by considering the level of accuracy desired for each research application, the experimental design, and the error inherent in the BIS technique (i.e., SEEs were TBW, 2.23 L, 4.4%; ECV, 1.26 L, 6.3%; ICV, 1.71 L, 5.5%). Finally, the determination of the precision of BIS, which is beyond the scope of this investigation, will require delineation of the errors associated with both BIS and dilutional methods.
The authors are grateful for the technical and administrative contributions of the following individuals: Douglas Casa, Ph.D., Marcos Echegaray, Ph.D., Kent LaGasse, M.D., Ralph Matott, Paul Schloerb, M.D., Neema Shakibai, and Michael Whittlesey.
This investigation was supported by technical support and a gift from the Procter & Gamble Company, Miami Valley Laboratories, Cincinnati, OH. Sodium bromide analyses were conducted at the University of Kansas Medical Center, Kansas City, KS. Deuterium oxide analyses were conducted at the laboratories of Metabolic Solutions, Inc., Merrimack, NH.
The results of the present study do not constitute endorsement of any product by the authors or the American College of Sports Medicine.
Current addresses: Robert W. Kenefick, University of New Hampshire, Durham, NH; John W. Castellani, U.S. Army Research Institute of Environmental Medicine, Natick, MA; Deborah Riebe, University of Rhode Island, Kingston, RI; James T. Kuznicki, Procter & Gamble Co., Cincinnati, OH.
Address for correspondence: Lawrence E. Armstrong, Ph.D., University of Connecticut, Box U-110, 2095 Hillside Road, Storrs CT 06269-1110. E-mail:ARMSTRON@UCONNVM.UCONN.EDU.
1. Alba's Medical Technology,
10th Ed. Anaheim, CA: Berkeley Scientific Publications, 1984, pp. 572-574.
2. Armstrong, L. E., C. M. Maresh, M. F. Bergeron, R. W. Kenefick, K. E. Lagasse, and D. Riebe. Urinary indices of hydration status.Int. J. Sport Nutr.
3. Azcue, M. D., Wesson, M. Neuman, and P. Pencharz. What does bioelectrical impedance spectroscopy (BIS) measure? In: Human Body Composition,
K. J. Ellis and J. D. Eastman (Eds.) New York: Plenum Press, 1993, pp. 121-123.
4. Baumgartner, R. N., W. C. Chumlea, and A. F. Roche. Bioelectric impedance for body composition. Exerc. Sport Sci. Rev.
5. Boulier A., A. L. Thomasset, and M. Apfelbaum. Bioelectrical-impedance measurement of body water. Am. J. Clin. Nutr.
6. Burton, A. C. Human calorimetry: the average temperature of the tissues of the body. J. Nutr.
7. Cole, K. S. and R. H. Cole. Dispersion and adsorption in dielectrics: I. alternating current characteristics. J. Chem. Physiol.
8. Deurenberg, P., A. Andreoli, and A. de Lorenzo. Multi-frequency bioelectrical impedance: a comparison between the Cole-Cole modelling and Hanai equations with the classical impedance index approach.Ann. Hum. Biol.
9. Deurenberg, P., A. Tagliabue, and F. J. M. Schouten. Multi-frequency impedance for the prediction of extracellular water and total body water. Br. J. Nutr.
10. Gamble, J. L., J. S. Robertson, C. A. Hannigan, C. G. Foster, and L. E. Farr. Chloride, bromide, sodium, and sucrose spaces in man.J. Clin. Invest.
11. Hanai, T. Electrical properties of emulsions. In:Emulsion Science,
P. H. Sherman (Ed.) London: Academic Press, 1968, pp. 354-477.
12. Lukaski, H. C. Methods for the assessment of human body composition: traditional and new. Am. J. Clin. Nutr.
13. Lukaski, H. C., P. E. Johnson, W. W. Bolonchuk, and G. I. Lykken. Assessment of fat-free mass using bioelectrical impedance measurements of the human body. Am. J. Clin. Nutr.
14. Mathie, J. R. and P. Withers. Impedance measurements of body-water compartments. Am. J. Clin. Nutr.
15. Miller, M. E. and C. J. Cappon. Anion-exchange chromatographic determination of bromide in serum. Clin. Chem.
16. Miller, M. E., J. M. Cosgriff, and G. B. Forbes. Bromide space determination using anion-exchange chromatography for measurement of bromide. Am. J. Clin. Nutr.
17. National Institutes of Health. Bioelectrical Impedance Analysis in Body Composition Measurement
. Bethesda, MD: NIH Technology Assessment Conference Statement, 1994, pp. 1-35.
18. Patel, R. V., J. R. Mathie, P. O. Witheres, E. L. Peterson, and B. J. Zarowitz. Estimation of total body and extracellular water using single- and multiple-frequency bioimpedance. Ann. Pharmacother.
19. Schoeller, D. A., R. F. Kushner, P. Taylor, W. H. Dietz, and Bandini L. Measurement of total body water: isotope dilution techniques. In: Body Composition Assessment in Youth and Adults,
A. F. Roche (Ed.) Columbus, OH: Ross Laboratories, 1985, pp. 24-29.
20. Schoeller, D. A., E. Van Santen, D. W. Peterson, W. Dietz, J. Jaspen, and P. Klein. Total body water measurements in humans with18
O and 2
H labeled water. Am. J. Clin. Nutr.
21. Segal, K. R., S. Burastero, A. Chun, P. Coronel, R. N. Pierson, and J. Wang. Estimation of extracellular and total body water by multiple-frequency bioelectrical-impedance measurement. Am. J. Clin. Nutr.
22. Siri, W. E. Body composition from fluid spaces and density, analysis of methods. In: Techniques for Measuring Body Composition,
J. Brozek and A. Henschel (Eds.). Washington, D.C.: National Academy of Sciences, 1961, pp. 223-244.
23. Thomas, B. J., B. H. Cronish, and L. C. Ward. Bioelectric impedance analysis for measurement of body fluid volumes: a review. J. Clin. Eng.
24. Thomasset, A. Electrochemical volume measurements of extracellular liquids. Biophysics significance of impedance at one kilocycle.Lyon Med.
25. Van Loan, M. D., L. E. Kopp, J. C. King, W. W. Wong, and P. L. Mayclin. Fluid changes during pregnancy: use of bioimpedance spectroscopy. J. Appl. Physiol.
26. Van Loan, M. D. and P. L. Mayclin. Use of multi-frequency bioelectrical impedance analysis for the estimation of extracellular fluid. Eur. J. Clin. Nutr.
27. Van Loan, M. D., P. Withers, J. Matthie, and P. L. Mayclin. Use of bioimpedance spectroscopy to determine extracellular fluid, intracellular fluid, total body water, and fat-free mass. In: Human Body Composition,
K. J. Ellis and J. D. Eastman (Eds.). New York: Plenum Press, 1993, pp. 67-70.
28. van Marken Lichtenbelt, W. D., K. R. Westerterp, and L. Wouters. How solid is the theoretical basis for bioelectrical impedance analysis? A reply. Am. J. Clin. Nutr.
29. van Marken Lichtenbelt, W. D., K. R. Westerterp, L. Wouters, and S. C. M. Luijenduk. Validation of bioelectrical-impedance measurements as a method to estimate body-water compartments. Am. J. Clin. Nutr.
30. Wong W. W., L. S. Lee, and P. D. Klein. Deuterium and oxygen-18 measurements on microliter samples of urine, plasma, saliva, and human milk. Am. J. Clin. Nutr.