Bioimpedance spectroscopy for the estimation of fat-free mass in end-stage renal disease

Abstract

Background & Aims

Bioimpedance spectroscopy may provide reliable estimates of fat-free mass in end-stage renal disease patients. We aimed to evaluate the ability of bioimpedance spectroscopy to estimate fat-free mass in end-stage renal disease patients using dual-energy X-ray absorptiometry as a reference.

Methods

Fat-free mass measured by bioimpedance spectroscopy was compared to fat-free mass measured by dual-energy X-ray absorptiometry in 16 end-stage renal disease patients on hemodialysis, 12 undialysed end-stage renal disease patients and 23 control subjects.

Results

Methods were highly correlated for fat-free mass in all subject groups (r = 0.87, P < 0.001). Mean bioimpedance spectroscopy measures of fat-free mass were not different from the dual-energy X-ray absorptiometry measures in any subject group. Individual comparisons revealed wide limits of agreement between methods in hemodialysis (11.6 to −9.72 kg) and undialysed patients (10.95 to −14.73 kg).

Conclusions

Although bioimpedance spectroscopy estimates of fat-free mass in the end-stage renal disease patient groups were not different from dual-energy X-ray absorptiometry and the methods were highly correlated, there was great individual variability. From these data it is clear that future studies are warranted before bioimpedance spectroscopy can be recommended as a valid clinical tool for assessing fat-free mass in end-stage renal disease patients.

INTRODUCTION

End-stage renal disease (ESRD) is characterized by a decreased ability of the kidneys to excrete and regulate body water, minerals and organic compounds¹. Hemodialysis (HD) attempts to manage these symptoms by restoring fluid and electrolyte balance¹. As a result of HD, shifts from the intracellular fluid (ICF) to the extracellular fluid (ECF) compartment may occur¹. Loss of lean tissue related to malnutrition and other consequences of renal impairment may also contribute to altered body composition¹. The prevalence of malnutrition among patients at the initiation of HD has been reported to be as high as 40%². Nutritional status worsens as the disease state progresses, and malnutrition is a significant predictor of morbidity and mortality in ESRD³. Accurate and reliable methods to assess body composition are needed to effectively assess the symptoms and severity of the disease, monitor changes in fluid and tissue levels and evaluate the efficacy of therapeutic treatment interventions.

Anthropometric measurements of height, weight and BMI are commonly used to assess nutritional status and estimate body composition⁴. This method is not sensitive enough to identify fluid shifts and malnutrition, because changes that occur at the cellular and tissue levels cannot be evaluated⁵. DXA is capable of evaluating body composition with greater sensitivity and is considered a reference method for estimating FFM, FM and bone mineral content⁶. It is recommended by the National Kidney Foundation's Kidney Disease Outcomes Quality Initiative (K/DOQI) clinical practice guidelines as a valid tool for measuring body composition in ESRD². Factors limiting its utility include the cost of instrumentation, exposure to ionizing radiation, inability to quantify fluid status, non-portability and the need for trained technicians², ⁷.

Bioimpedance spectroscopy (BIS) is a relatively inexpensive, safe and portable method of body composition that has the potential to overcome many of these limitations. BIS devices measure impedance over a spectrum of frequencies and use software to fit the impedance data to the Cole model⁸. The modeled data can be applied to equations for the calculation of ECF and ICF. Estimates of fat-free mass (FFM) can be calculated based on the manufacturer's software. Application of the current over a spectrum of frequencies allows for the differentiation of the ECF and ICF compartments and provides estimates of FFM⁹ making this method especially appealing to the renal community. A consensus regarding the use of BIS to evaluate body composition in ESRD and other clinical populations has not been reached. The aim of the present study was to evaluate the ability of the BIS to estimate FFM compared to DXA in ESRD patients.

SUBJECTS AND METHODS

Subjects in this analysis were part of a larger study designed to evaluate responses to exercise in patients with ESRD after changing treatments to daily dialysis or transplantation. The analyses presented here are from the baseline data collected on 51 of the study subjects (9 females, and 42 males) aged 21 – 64 years (45 ± 11) with mean body mass index (BMI) 27.29 kg/m². Subjects were recruited into three groups. Group I consisted of 16 ESRD patients treated with HD (25.8 ± 23.4 mo; range: 3 – 72 mo) who had a mean age of 42.30 ± 14.80 years and a mean BMI of 27.19 ± 5.21 kg/m². Group II consisted of 12 undialysed (UD) ESRD patients who had a mean age of 47.68 ± 10.00 years and a mean BMI of 28.83 ± 4.15 kg/m². Lastly, group III consisted of 23 healthy control subjects who had a mean age of 45.85 ± 8.40 years and mean BMI of 26.55 ± 4.35 kg/m². ESRD patients were recruited from several dialysis centers in the San Francisco Bay area and Minneapolis/St Paul as described by Painter et al (2010) in the American Journal of Kidney Disease (in press). ESRD patients scheduled for transplant were referred from the University of California at San Francisco and the University of Minnesota. Patients were referred to the study by nurse coordinators and/or physicians. Sedentary healthy control subjects were recruited from kidney donors (> 1 year post kidney donation with normal renal function as indicated by estimated GFR) at the University of Minnesota. An attempt was made to match the group of control subjects to the HD and UD patient groups by percentage of women, and age decades, although there were inadequate donors < 30 years of age.

The present study was initiated at the University of California at San Francisco, and completed at the University of Minnesota. All testing was performed in the General Clinical Research Center of the respective institutions. Testing was done on a mid-week non-dialysis day, at least 15 hours after completion of the dialysis treatment. Subjects were instructed to avoid vigorous physical activity 24 hours prior to testing. Due to the duration of testing and the multiple procedures involved in the overall study, a small breakfast was provided to subjects 30–45 minutes prior to the body composition analysis. Body composition was assessed in all subjects by DXA and BIS. DXA measurements of body composition were taken first, followed by instruction to walk 150 feet and then rest in the supine position for 10 minutes prior to the BIS measurements. All subjects signed consent documents approved by the Committee on Human Subjects at the University of California-San Francisco and the Institutional Review Board at the University of Minnesota.

DXA

A DXA whole-body scan was performed using a General Electric Lunar Prodigy scanner (Lunar Radiation Corp., Madison, WI) and provided estimates of FFM calculated according to the manufacturer's software (version 8.8). DXA passes two X-ray energies through the body. The initial energy emitted is known and the final energy is detected after passing through the body. These values are then applied to known body mass attenuation equations and the proportion of FM to FFM is used to determine whole body composition⁷, ¹⁰. To help ensure accuracy, subjects were asked to remove all metal items prior to the DXA scan. None of the female subjects were pregnant at the time of examination.

BIS

BIS measurements of ICF, ECF, and FFM were taken 10 minutes after the subjects were asked to assume a supine position, using the ImpediMed SFB7 device (ImpediMed Limited, Eight Mile Plains, Australia). BIS measures the body's impedance to an electrical current and based on impedance values, fluid volumes and FFM can be determined¹¹. Electrodes were arranged in the traditional tetrapolar configuration, with injection electrodes placed wrist-to-ankle and sensing electrodes placed hand to foot¹². For patients on HD therapy, electrodes were placed on the side of the body contralateral to the arteriovenous fistula. The ImpediMed SFB7 device made ten measurements at 1-second intervals by applying an electrical current at 256 frequencies ranging from 4 kHz to 1000 kHz. The Cole method of biophysical modeling and the Hanai mixture theory equations using standard resistivity constants were then used to calculate ECF, and ICF¹³. Estimates of FFM were calculated per the manufacture's software.

Statistical Analysis

Data are reported as means ± standard deviation (SD). Comparisons between methods were made in all subject groups (HD, UD and controls) for measures of FFM. For group-level comparisons: A one-way analysis of variance (ANOVA) was used to compare the mean values of subject characteristics and body composition measures between all subject groups, and where ANOVA showed statistically significant differences, post hoc comparisons for unequal variances were made using the Tamhane's test. The Student's paired t-test was used to compare mean-level accuracy of BIS within each group and the Pearson's Product Moment Correlation analysis was used to determine the relative agreement between DXA and BIS. To assess the individual variability between the DXA and BIS methods for FFM, the Bland-Altman method¹⁴ was used. The BIS percent differences from the DXA reference value for estimates of FFM were calculated in each individual ESRD patient and control subject. BIS bias was calculated by subtracting the BIS value from the DXA value (DXA-BIS). This bias was evaluated for each subject and also across subject groups, by ANOVA. Correlations between subject characteristics of age, gender and BMI were evaluated for their contribution to the BIS bias in estimates of FFM. To determine if these variables were significant predictors of BIS bias, a general linear model with backwards step-wise model selection was fit to the data. Statistical analyses were performed using SPSS 17.0 for Windows software (SPSS, Inc., Chicago, IL) and statistical significance was set at an alpha value ≤ 0.05. Bland-Altman plots were created using Sigma Plot 10.0 (Systat Software, Inc. Chicago, IL).

RESULTS

Subject Characteristics

Mean physical and clinical characteristics of all subjects and each of the groups are summarized in Table 1. No significant differences were observed between any of the groups for age, weight, height or BMI. One ESRD subject had a BMI in the underweight category (BMI: 16.5 – 18.4 kg/m²) of 17.0 kg/m². All other ESRD patients and healthy control subjects had BMIs in the normal (BMI: 18.5 – 24.9 kg/m²) to obese class II (35 – 40 kg/m²) categories that ranged from 20.3 – 35.5 kg/m². Mean clinical lab values of all subjects and each of the groups are also presented in Table 1. For all subjects, the mean values for calcium, c-reactive protein, glucose, potassium, and sodium fell within the expected normal ranges (data not shown). Compared to controls, the HD and UD ESRD patient groups had significantly higher mean values of blood urea nitrogen (BUN) and creatinine (Cr) and significantly lower mean values of hematocrit (Hct) and hemoglobin (Hgb). None of the ESRD patients or healthy control subjects presented with hypoalbuminemia.

Table 1.

Physical and Clinical Characteristics by Subject Groups^a

	Controls (n = 23)	Hemodialysis (n = 16)	Undialysed (n = 12)
		Mean ± SD
Age (y)	45.85 ± 8.40	42.30 ± 14.80	47.68 ± 10.00
Weight (kg)	82.68 ± 18.65	85.06 ± 21.18	90.66 ± 14.25
Height (m)	1.76 ± 0.10	1.76 ± 0.09	1.77 ± 0.06
BMI (kg/m2)	26.55 ± 4.35	27.19 ± 5.21	28.83 ± 4.15
BUN	17.78 ± 5.38	34.94 ± 10.85b,c	59.18 ± 18.40c
Cr	1.15 ± 0.17	8.97 ± 2.27b,c	5.05 ± 2.54c
Hgb	14.49 ± 0.94	13.01 ± 0.93c	12.19 ± 1.41c
Hct	43.30 ± 2.49	38.69 ± 2.96c	36.23 ± 4.11c

Abbreviations: BUN, blood urea nitrogen; Cr, creatinine; Hgb, hemoglobin; Hct, hematocrit

^aSubject group comparisons by one-way ANOVA

^bSignificantly different from undialysed (Tamhane multiple comparisons of means with post hoc tests): P < 0.05

^cSignificantly different from controls (Tamhane multiple comparisons of means with post hoc tests): P < 0.05

Mean body composition measurements of all subjects and groups are summarized in Table 2. No significant differences in body composition by either method were observed between groups, by ANOVA. The UD patients had a significantly higher ECF volume than controls and the HD patients had a significantly higher ECF/TBF and ECF/ICF ratios than controls.

Table 2.

Body Composition Characteristics by Subject Groups^a

	Controls (n = 23)	Hemodialysis (n = 16)	Undialysed (n = 12)
		Mean ± SD
FFMDXA (kg)	56.57 ± 13.34	61.18 ± 10.26	62.31 ± 8.85
FFMBIS (kg)	57.67 ± 12.84	60.24 ± 13.07	64.20 ± 12.44
TBFBIS (L)	42.18 ± 9.39	44.09 ± 9.57	47.00 ± 9.10
ECFBIS (L)	18.37 ± 3.86	21.74 ± 5.02	22.04 ± 2.93b
ICFBIS (L)	25.40 ± 8.52	22.35 ± 6.90	24.95 ± 7.57
ECFBIS/ TBFBIS	0.44 ± 0.02	0.50± 0.08b	0.48 ± 0.08
ECFBIS/ ICFBIS	0.75 ± 0.11	1.06± 0.36b	0.96 ± 0.30

Abbreviations: FFM, fat-free mass; DXA, dual-energy X-ray absorptiometry; BIS, bioimpedance spectroscopy; TBF, total body fluid; ECF, extracellular fluid; ICF, intracellular fluid

^aSubject group comparisons by one-way ANOVA

^bSignificantly different from Controls (Tamhane multiple comparisons of means with post hoc tests): P < 0.05

Group-level Comparisons

DXA and BIS values of FFM for the different study groups are shown in Table 2. Estimates of FFM by both methods were not significantly different between any of the treatment groups by ANOVA. Within all subject groups, BIS produced estimates of FFM that were not different from DXA by paired t-tests (Table 3). DXA and BIS measures of FFM were highly correlated in HD patients (r = 0.924, P < 0.001), UD patients (r = 0.871 P < 0.001) and control subjects (r = 0.950, P < 0.001) (Figure 1).

Table 3.

Within Subject Group Comparisons^a

	Controls (n = 23)		Hemodialysis (n = 16)		Undialysed (n = 12)
	Mean ± SD	P	Mean ± SD	P	Mean ± SD	P
DXAFFM – BISFFM	−1.102 ± 4.167	0.218	0.944 ± 5.326	0.489	− 1.891 ± 6.424	0.330

^aWithin subject group comparisons by Student's paired t-test

Figure 1. Correlation Between Dual-energy X-ray Absorptiometry and BIS for FFM by Subject Group.

Correlation between fat-free mass (FFM) measured by bioimpedance spectroscopy (BIS) and dual-energy X-ray absorptiometry (DXA). The dependent variable was FFM measured by using BIS. In hemodialysis patients (◆), the correlation coefficient (r) was r = 0.924, P < 0.001; in undialysed patients (▲), the correlation coefficient was r = 0.871, P < 0.001and in control subjects (●), the correlation coefficient was r = 0.950, P < 0.001.

Individual Variability

Agreement between DXA and BIS for estimates of FFM was further evaluated using the Bland-Altman method. The estimates obtained by DXA and BIS were averaged and plotted against the difference between the two devices for each group of subjects. These results are shown in Figure 2(a – c). The mean difference between DXA and BIS was smaller for HD patients (0.94 ± 5.33 kg) and larger for UD patients (−1.89 ± 6.42 kg) compared to control subjects (−1.10 ± 4.17 kg). The upper- and lower-limits of agreement were wider for all ESRD subjects compared to controls. Six (38%) HD patients, 4 (33%) UD patients and 11 (48%) control subjects had BIS estimates of FFM that were within ± 5% of the reference value (DXA-measured FFM) (Table 4).

Figure 2. (a–c) Bland-Altman Plots of Fat-Free Mass Measures by the ImpediMed Device and Dual-energy X-ray Absorptiometry (Reference) Method.

Agreement between bioimpedance spectroscopy (BIS) and dual-energy X-ray absorptiometry (DXA) for measurement of fat-free mass (FFM) in hemodialysis (HD) patients (a), undialysed (UD) patients (b) and control subjects (c) by the Bland & Altman method. Figures show the mean difference (solid line) and 95% limits of agreement between methods (dashed line).

Table 4.

BIS Percent Difference from DXA - All Subjects

BIS – FFM
Difference from reference, DXA	No. of Subjects	Percent of total	Σ Percent
< ± 5.0%	21	41.2	41.2
± 5.0 – 9.9%	16	31.4	72.6
± 10.0 – 14.9%	9	17.6	90.2
≥ ± 15.0%	5	9.8	100.0

Factors Predicting Bias in BIS Measurements of FFM

Age and BMI were significantly correlated with the bias in BIS (Table 5). BMI and the interactions between gender and BMI and gender and age showed a significant unique effect in the model predicting BIS bias for estimates of FFM. While holding all other variables constant, a one-unit increase in BMI resulted in a BIS overestimation of FFM by 2.588 kg in female subjects and 1.383 kg in male subjects. The effect of BMI was significant (P = 0.001) and the interaction between gender and age was also significant (P = 0.017). The interaction between BMI and age approached significance (P = 0.06) (Table 6). The R² value for this model was 0.429; indicating 42.9% of the variation in BIS bias for estimates of FFM can be explained by these relationships.

Table 5.

Correlation of BIS and Subject Characteristics – All Subjects

	FFMDXA - FFMBIS
Age (y)	0.294a
Gender	0.127
BMI (kg/m2)	−0.372b

^aCorrelation is significant at the 0.05 level (2-tailed)

^bCorrelation is significant at the 0.01 level (2-tailed)

Table 6.

General Linear Regression for BIS Prediction Errors of FFM – All Subjects

	Type III Sum of Squares	F - Statistic	P	β ± SE
Age	17.323	1.008	0.321	−0.573 ± 16.26
Gender	2.183	0.127	0.723	4.415 ± 12.39a
BMI	200.639	11.671	0.001	−1.383 ± 0.552
Gender * BMI	130.545	7.594	0.008	−1.175 ± 0.427a
Gender * Age	106.011	6.166	0.017	0.494 ± 0.199a
BMI * Age	61.824	3.596	0.064	0.023 ± 0.012

β = unstandardized coefficient

^aMale was used as the reference category for gender

R² = 0.429

DISCUSSION

We found that in all treatment groups BIS estimates of FFM were not significantly different from and were highly correlated with DXA measured FFM values. Despite the apparent good agreement between methods, individual variability was notable, as evidenced by wider limits of agreement and fewer BIS estimates that fell within ± 5% among ESRD patients compared to the control subjects. This large individual variability is further illustrated by the fact that less than half of the control subjects had FFM values that fell with ± 5% of DXA estimates. We also demonstrated that BMI, age and gender were significantly correlated with the BIS bias.

Given the observed differences in clinical lab values and fluid volumes between subject groups, it was somewhat surprising that there were no differences in FFM (by either method) between treatment groups. Although it has been previously reported that overhydration in HD patients yielded significantly different estimates of FFM compared to controls, as measured by DXA¹⁵ we did not find this to be the case. This lack of difference in FFM between treatment groups may be partially explained by the characteristics of our ESRD patient population. As part of the inclusion criteria of the larger overall study, patients were required to engage in exercise testing without exacerbating their symptoms of ESRD. Thus our patient population may have been healthier than average ESRD patients, explaining why no differences in body composition were observed between treatment groups. These data are supported by Stall et al (1996)¹⁶ who found no differences in %BF between dialysis patients and control subjects when measured by DXA and single frequency bioelectrical impedance analysis (SF-BIA).

Group level Comparisons of DXA and BIS

To our knowledge no other studies have evaluated the use of BIS against DXA for estimates of FFM in ESRD patients. However, a few studies evaluated the use of SF-BIA^15–17, and one evaluated multi frequency bioelectrical impedance analysis (MF-BIA)¹⁷ against DXA for measures of FFM in ESRD. In these previous studies the methods were highly correlated; however, in contrast to our findings two reported significant mean differences between methods^15–17. The disparate results between the present study and those conducted previously may be attributed to differences in bioimpedance methods used. SF- and MF-BIA require the use of population-specific prediction equations that may not have been valid for the study population to which they were applied. The use of Cole biophysical modeling and the Hanai mixture theory equations in BIS represents a different approach from other BIA methods in that regression-derived prediction equations are not utilized, although resistivity constants are required that are likely to vary in different populations. Theoretically, the BIS Cole modeling and mixture theory approach with appropriate resistivity constants could produce better estimates of body composition than BIA in ESRD patients. This notion is supported by Ho et al. (1994)¹⁸ who showed that BIS was more effective than BIA methods in evaluating fluid volumes in ESRD patients on HD therapy.

Individual Variability of DXA and SF-BIA and MF-BIA

One previous study evaluated SF-BIA and MF-BIA estimates of FFM against DXA in ESRD patients and reported individual variability by the Bland-Altman method¹⁷. The authors reported upper and lower limits of agreement for both SF-BIA and MF-BIA estimates of FFM that were narrower than the limits of agreement observed across all treatment groups in the present study. However, in the present study one subject in the control group and one subject in the HD group had FFM difference values that were outliers (−11.39 kg and −10.10 kg) of the mean differences for FFM in their respective group populations. Removal of these outliers would result in a tighter clustering of data points around the mean difference in both the healthy subjects and HD patients. Regardless of the outliers, the Bland-Altman plots constructed in the present study revealed relatively wider limits of agreement among all ESRD patients compared to healthy control subjects, and thus, the individual variability may be too large to produce meaningful measures of FFM in the clinical setting. That said, nearly 61% and 36% of the ESRD patients had BIS-measured FFM values that fell within ± 10% and 5% of DXA-measured values, respectively, suggesting that for at least some individuals, BIS may be an accurate method of evaluation.

BIS Bias

Our multiple linear regression models show BMI and the interactions between gender and BMI and gender and age had a significant unique effect on predicting BIS bias. These findings are supported by a previous study conducted in healthy individuals by Sun et al (2005)¹⁹ who reported that age, gender and BMI were significant explanatory contributors to the variation in MF-BIA estimates of %BF. In addition we were able to show a large portion (at least 42%) of the variance in BIS bias could be explained by these factors. The remaining bias may be due to measurement error, or to additional covariates that warrant further exploration.

Limitations

The use of cross-sectional data for the evaluation of body composition assessment methods may limit the findings of the present study because paired t-tests can yield significantly different results even though strong correlations between methods may exist²⁰. Furthermore, BIS is typically scaled to multiple dilution because it was originally designed to assess fluid volumes. FFM estimation is calculated as a secondary measure and thus scaling differences may introduce error. Comparison of BIS estimates of fluid volume to multiple tracer dilution would have even greater clinical utility. Unfortunately multiple tracer dilution was not conducted in this study, and thus we could not compare BIS estimates to reference estimates of fluid volume in our subjects. Although some investigators have evaluated BIS against multiple dilution for measuring absolute fluid volumes in dialysis patients^{18, 21, 22}, additional studies are warranted before BIS can be adopted as a viable tool to estimate fluid volumes, and secondarily the FFM compartment in ESRD patients. In addition the high individual variability observed in the HD and UD patient groups may have been attributed to the small number of patients in each group. Future studies that include a greater number of HD and UD patients are needed in order to confirm or refute the results of the present study.

The BIS device and software used in the present study utilized the Hanai mixture theory to calculate measures of body composition. This theory accounts for the non-conducting elements in the ECF and ICF, but makes assumptions about their volume and concentrations in the body. In addition, assumptions are made about the body proportions of the leg, arm, trunk and height²³. These assumptions may have been violated due to the altered fluid status in the ESRD groups.

Two other factors may be a source of error. First, FFM was calculated from measures of ECF and ICF volumes determined from standard specific resistivity constants that may not apply to individuals with altered body composition. Second, BIS relies on the interpretation of the body as a series of 5 cylinders where changes in the composition of the limbs and/or trunk, or fluid shifts occurring during HD may be disproportionately detected²⁴. Although BIS is thought to provide more individualized estimates of body composition than SF- and MF-BIA, the validity of the assumptions and resistivity constants used has been questioned in populations with abnormal fluid status²⁵.

In summary, the present study has demonstrated that at the group mean-level, a BIS device using Cole modeling and Hanai mixture theory equations with standard resistivity constants produces estimates of FFM that are not different from DXA. To the contrary, variability between methods at the individual-level was large and may be attributed to altered fluid status in the patient population and may also be attributed to the small number of subjects in the HD and UD patient groups. Age, gender, and BMI partially explain the variation in BIS bias for estimates of body composition. This study provides novel data to the field of body composition assessment in ESRD patients. We have shown that in some patients BIS provides estimates of FFM that are in good agreement with DXA. Therefore BIS, when used in conjunction with other body composition tools, should not be ruled out as a viable method to assess body composition in ESRD patients. However, future studies that evaluate the ability of BIS to measure changes in FFM over time, compare the ability of BIS to measure fluid volumes against multiple isotope dilution, establish appropriate specific resistivity constants that can refine the measurements, explore other potential confounding variables in BIS bias, and involve a greater number of and a more heterogeneous mix of ESRD patients are warranted before BIS can be recommended as a valid clinical tool for assessing FFM in ESRD patients.

Non-standard abbreviations

BIS

Bioimpedance spectroscopy

FFM

fat-free mass

ESRD

end-stage renal disease

HD

hemodialysis

UD

undialysed