Cross-calibration of eight-polar bioelectrical impedance analysis versus dual-energy X-ray absorptiometry for the assessment of total and appendicular body composition in healthy subjects aged 21-82 years.

Aim: To calibrate eight-polar bioelectrical impedance analysis (BIA) against dual-energy X-ray absorptiometry (DXA) for the assessment of total and appendicular body composition in healthy adults. Research design: A cross-sectional study was carried out. Subjects: Sixty-eight females and 42 males aged 21-82 years participated in the study. Methods: Whole-body fat-free mass (FFM) and appendicular lean tissue mass (LTM) were measured by DXA; resistance ( R ) of arms, trunk and legs was measured by eight-polar BIA at frequencies of 5, 50, 250 and 500 kHz; whole-body resistance was calculated as the sum R of arms, trunk and legs. Results: The resistance index (RI), i.e. the height 2 /resistance ratio, was the best predictor of FFM and appendicular LTM. As compared with weight (Wt), RI at 500 kHz explained 35% more variance of FFM ( vs 0.57), 45% more variance of LTM arm ( vs 0.48) and 36% more variance of LTM leg ( vs 0.50) ( p < 0.0001 for all). The contribution of age to the unexplained variance of FFM and appendicular LTM was nil or negligible and the RI 2 sex interactions were either not significant or not important on practical grounds. The percent root mean square error of the estimate was 6% for FFM and 8% for LTM arm and LTM leg. Conclusion: Eight-polar BIA offers accurate estimates of total and appendicular body composition. The attractive hypothesis that eight-polar BIA is influenced minimally by age and sex should be tested on larger samples including younger individuals.


Introduction
Sarcopenia, i.e. a decrease in skeletal muscle mass (SMM) and strength, is a common feature of ageing (Roubenoff 2000). Sarcopenia may impair the ability of the body to cope with stress and disease and contribute to morbidity and mortality in the elderly (Dutta 1997). However, the prevalence of sarcopenia is not known and this is of obstacle to the understanding of its prognostic significance (Dutta 1997).
The reference methods for the assessment of SMM are computed tomography (CT) and magnetic resonance imaging (MRI) (Lukaski 1996). Dual-energy X-ray absorptiometry (DXA) compares well with CT and MRI and has been proposed for the assessment of SMM owing to its lower cost and higher availability (Fuller et al. 1999a, Visser et al. 1999, Wang et al. 1999aElia et al. 2000, Levine 2000, Shih et al. 2000. The three-compartment DXA model separates body mass into fat mass (FM), lean tissue mass (LTM) and bone mineral content (BMC), with the sum of LTM and BMC representing fat-free mass (FFM) (Pietrobelli et al. 1996). At the appendicular level, LTM is synonymous with SMM so that DXA provides a simple means of estimating SMM (Wang et al. 1999a). A limitation of DXA is, however, that different densitometers and software versions give different estimates of body composition (Lohman 1996).
CT, MRI and DXA cannot be employed for population studies, mainly because of logistical problems. Segmental bioelectrical impedance analysis (BIA) offers a simple means of estimating appendicular LTM and is probably the best candidate for the assessment of SMM at the population level (Chumlea et al. 1995. Calibration studies of BIA versus DXA have shown that four-polar BIA gives accurate estimates of appendicular LTM or FFM in adult subjects , Fuller et al. 1999b, Nunez et al. 1999, Elia et al. 2000, Lukaski 2000, Tagliabue et al. 2000. However, while four-polar whole-body BIA has undergone many calibration studies in the elderly (Steen et al. 1987, Deurenberg et al. 1990, Brodowicz et al. 1994, Visser et al. 1995, Fuller et al. 1996, Roubenoff et al. 1997, Vache et al. 1998, Bussolotto et al. 1999, Hansen et al. 1999, Dittmar and Reber 2001, segmental BIA has not undergone a systematic evaluation. Pietrobelli et al. (1998) found, however, that age influences the estimate of LTM obtained from segmental fourpolar BIA. This is not completely unexpected because BIA, like other indirect techniques, relies on assumptions that are partly age-dependent . Another problem is whether anthropometry, and especially body weight (Wt), contributes more than BIA to the estimate of body composition. In a recent study using four-polar BIA at 50 kHz, we have found, for instance, that BIA was not superior to Wt in assessing total and leg FFM in anorexic women (Bedogni et al., in press). Use of frequencies >50 kHz may, however, improve the estimate of total and appendicular body composition from BIA because of better penetration of the electrical current into intracellular water and thus muscle cells (Wang et al. 1999b;Deurenberg et al. 2002).
Recently, an eight-polar impedance meter has been made available on the market (InBody 3.0, Biospace, Seoul, Korea). We found this method attractive for three reasons: (1) the use of very practical tactile electrodes for measuring segmental resistances at multiple frequencies, (2) the absence of need to standardize the subject's posture before BIA, and (3) the rapidity of measurement. These characteristics have the potential to reduce measurement times as compared with four-polar BIA and make this instrument ideal for epidemiological studies (Bedogni et al. 2002). The present study aimed at calibrating eight-polar BIA versus DXA in a sample of healthy subjects aged 21-82 years.

Subjects
Eligible for the study were white Caucasian subjects of both sexes fulfilling the following criteria: (1) age ! 18 years; (2) body mass index (BMI) ! 18.5 kg m -2 ; (3) absence of chronic (e.g. diabetes) and acute (e.g. influenza) disease, as determined by clinical history and physical examination; (4) menstrual cycle between the 6th and 10th day for fertile women; (5) no use of drugs known to interfere with body water homeostasis. Subjects aged 50 years were recruited mainly among the personnel working at the Departments of Biomedical Sciences and Geriatrics and those aged >50 years among the subjects visited at the Outpatient Clinic of the Department of Geriatrics. The study procedures had been approved by the local Ethical Committee and all subjects gave informed consent.

Anthropometry
All anthropometric measurements were performed by the same operator following the Anthropometric Standardization Reference Manual (Lohman et al. 1988). Weight (Wt) was measured to the nearest 100 g and height (Ht) to the nearest 0.1 cm using an electronic balance with an incorporated stadiometer (Tanita, Tokyo, Japan). BMI was calculated as Wt (kg)/Ht (m) 2 .

Eight-polar BIA
Resistance (R) of arms, trunk and legs was measured in fasting conditions (!8 h) at frequencies of 5, 50, 250 and 500 kHz with an eight-polar tactile-electrode impedance meter (InBody 3.0, Biospace, Seoul, Korea). This instrument makes use of eight tactile electrodes: two are in contact with the palm (E1, E3) and thumb (E2, E4) of each hand and two with the anterior (E5, E7) and posterior aspects (E6, E8) of the sole of each foot (figure 1).
The subject stands with his soles in contact with the foot electrodes and grabs the hand electrodes. The sequence of measurements, controlled by a microprocessor, proceeds as follows. An alternating current (a.c.) of 250 mA of intensity (I) is applied between E1 and E5. The recorded voltage difference (V) between E2 and E4 is divided for I to obtain the resistance of right arm (R RA ). The same operation is performed with V recorded between E4 and E8 to obtain trunk resistance (R T ) and with V recorded between E6 and E8 to obtain the resistance of right leg (R RL ). The a.c. is then applied between E3 and E7 and the value of V measured between E2 and E4 is used to calculate the resistance of left arm (R LA ). Lastly, the value of V measured between E6 and E8 is used to calculate the resistance of left leg (R LL ). No caution was taken to standardize the subject's posture before BIA, as suggested by the manufacturer. Segmental RI were calculated as Ht (cm) 2 /R x (), where R x is the resistance of arm or leg at frequency x. Whole-body resistance (R sumx ) was calculated as the sum of segmental R x (right arm þ left arm þ trunk þ right leg þ left leg). The whole-body resistance index (RI sumx ) was calculated as Ht (cm) 2 /R sumx (). The between-day precision of InBody 3.0, determined by three daily measurements of two subjects for five consecutive days, was 2.7% ( 5 ); within-day precision was always 2.0 ( 3 ).
2.4. DXA DXA scans were performed by the same operator using a Lunar DPX-L densitometer and adult software version 3.6 (Lunar Corporation, Madison, WI, USA). The precision of FFM and BMC assessment, as determined by three repeated weekly measurements on three subjects, was 2.5 and 1.0%, respectively. The precision of appendicular LTM assessment was 2.5%. The difference between body mass measured by DXA and Wt measured by scale was À1 AE 1 kg. In spite of its statistical significance (p < 0.0001, paired t-test), this difference is of no practical relevance.

Statistical analysis
Sample size was determined by considering that a sample of 55 subjects has a power of 0.80 to detect a slope of 0.90 at an alpha level of 0.05 when the SD of Y (FFM) is 11 kg and that of X (RI 500 ) is 5 . We enrolled 110 (55 Â 2) subjects aiming at developing BIA algorithms and cross-validating them on an equal number of subjects. Between-sex comparisons were performed by unpaired t-tests. The study hypothesis was tested using a general linear model (GLM) with FFM or appendicular LTM as the dependent variable and the corresponding RI x as the predictor variable. An interaction term between RI x and sex was added to test the influence of sex on this relationship. The adjusted determination coefficient (R 2 adj ), the root mean square error (RMSE) and the percent root mean square error (RMSE% ¼ RMSE/ mean value of Y) obtained from linear regression of FFM or LTM versus RI x were used to determine the accuracy of BIA (Guo et al. 1996). The same approach was used with Wt, Ht and R x . Statistical significance was set to a value of p < 0.05 for all tests. Statistical analysis was performed on a MacOS computer using the Statview 5.1 and SuperANOVA 1.11 software packages (SAS, Cary, NC, USA).

Characteristics of the subjects
The measurements of the 110 subjects are given in table 1. The high female:male ratio (1.6) is explained mainly by the higher probability that we had of finding women rather men not using diuretics over 60 years of age. The mean age was 54 years (range: 21-82 years) and there was no difference between sexes (p ¼ 0.786). Sixty-one subjects had <50 years of age and 49 were aged !50 years. Of these latter, 20 were aged !70 years. Wt and Ht were higher in men than women (p < 0.0001) but BMI was similar (p ¼ 0.537). BMI was between 18.5 and 33.8 kg m -2 in females and between 19.8 and 31.2 kg m -2 in males. As expected, FFM and appendicular LTM were higher in men than women (p < 0.0001). R was significantly lower in men than women at all frequencies because of their higher total body water (TBW) (table 2).

Accuracy of eight-polar BIA in the assessment of FFM
The variance of FFM explained by Wt, Ht, R sumx and RI sumx is given in table 3. RI sumx was the best predictor of FFM and there was an increase of 2% in the explained variance of FFM from 5 to 500 kHz. RI sum500 explained 35% more variance of FFM than Wt and 23% more variance than Ht. The RMSE% associated with the prediction of FFM from RI sum500 (6%) was also substantially lower than that associated with the prediction from Wt (14%) and Ht (12%). The Wt Â sex (p ¼ 0.002), Ht Â sex (p ¼ 0.0001) and R sum500 Â sex (p ¼ 0.0001) interactions were significant but RI sum500 Â sex was not (p ¼ 0.410). Thus, we modelled the predictive algorithm independently of sex, as we had previously done for TBW (Bedogni et al. 2002). Age did not contribute to the unexplained variance of FFM (R 2 adj ¼ 0.003, p ¼ 0.257) and the contribution of Wt was low (R 2 adj ¼ 0.06, p ¼ 0.004). Adding Wt as predictor with RI did not improve the accuracy of the estimate (RMSE% ¼ 6%).

Accuracy of eight-polar BIA in the assessment of LTM arm
The variance of mean LTM arm explained by Wt, Ht, mean R armx and mean RI armx is given in table 3. We used the mean values of LTM arm and RI armx because there was no significant RI armx Â hemisome interaction (p ¼ 0.07). Mean RI armx was the  best predictor of mean LTM arm and there was an increase of 2% in the explained variance of mean LTM arm from 5 to 500 kHz. Mean RI arm500 explained 45% more variance of mean LTM arm than Wt and 31% more variance than Ht. The RMSE% associated with the prediction of mean LTM arm from mean RI arm500 (8%) was also substantially lower than that associated with the prediction from Wt (22%) and Ht (19%). All the X Â sex interactions were significant (Ht Â sex, p ¼ 0.0001; R 500 Â sex, p ¼ 0.0001; Wt Â sex, p ¼ 0.034 and RI arm500 Â sex, p ¼ 0.004). Adding sex and RI 500 Â sex as predictors did not, however, improve the accuracy of the RI arm500 (RMSE% ¼ 8%). Thus, we modelled the predictive algorithm independently of sex. Age explained only a minimal portion of residual LTM arm (R 2 adj ¼ 0.04, p ¼ 0.024) and Wt no portion at all (R 2 adj ¼ 0.006, p ¼ 0.431). Adding age as predictor with RI did not improve the accuracy of the estimate (RMSE% ¼ 8%).

Accuracy of BIA in the assessment of LTM leg
The variance of LTM leg explained by Wt, Ht, R legx and RI legx is given in table 3. We used the mean values of FFM leg and RI legx because there was no significant RI legx Â hemisome interaction (p ¼ 0.502). Mean RI legx was the best predictor of mean FFM leg and there was an increase of 7% in the explained variance of FFM leg from 5 to 500 kHz. RI leg500 explained 36% more variance of FFM leg than Wt and 13% more variance than Ht. The RMSE% associated with the prediction of mean LTM leg from mean RI leg500 (8%) was also lower than that associated with the prediction from Wt (16%) and Ht (12%). The Wt Â sex (p ¼ 0.020), Ht Â sex (p ¼ 0.0001) and R 500 Â sex (p ¼ 0.002) interactions were significant but RI leg500 Â sex was not (p ¼ 0.408). Thus, we modelled the predictive algorithm independently of sex. Age explained no portion of residual LTM leg (R 2 adj = 0.002, p ¼ 0.263) and Wt explained only a minimal portion of it (R 2 adj ¼ 0.03, p < 0.388). Adding Wt as predictor with RI did not improve the accuracy of the estimate (RMSE% ¼ 8%).
Eight-polar BIA and body composition 385

BIA algorithms
To develop BIA algorithms for the prediction of FFM and appendicular LTM, we randomly split the study sample in two halves. The regression lines of the FFM-RI and LTM-RI relationships obtained in the first half (n ¼ 55) were compared with those obtained in the second half (n ¼ 55) (figure 2).
Because the slopes and intercepts were similar, the two halves were pooled together and common algorithms were developed (table 4). The RMSE% was 6% for FFM and 8% for both LTM arm and LTM leg .

Discussion
In this study, we cross-calibrated eight-polar BIA against DXA for the assessment of FFM and appendicular LTM in healthy subjects.

386
M. Malavolti et al. Eight-polar BIA was consistently superior to Wt in estimating FFM and appendicular LTM. In our studies of four-polar BIA, we have sometimes found Wt to be better than RI in estimating body composition , in press, Scalfi et al. 1997. Moreover, it is not uncommon for Wt to explain most of the variance of body compartments when employed as a predictor with RI (Bedogni et al., in press). In the present study, however, RI was substantially better than Wt in estimating body composition. Even more important, Wt did not contribute or contributed very little to residual FFM and LTM (R 2 adj 0.06). This may be a merit of eight-polar BIA but in order to establish this with certainty a direct comparison with four-polar BIA is needed. Unique characteristics of eight-polar BIA that may contribute to its very low dependency from Wt are: (1) the use of tactile electrodes, avoiding problems connected with adhesive electrodes, (2) the fact that whole-body eight-polar R is the sum of segmental resistances obtained with a five-cylinder model of the human body (see figure 1), and (3) the insensitivity of eight-polar BIA to subject's posture.
Age did not contribute or contributed very little to the variance of FFM and LTM (R 2 adj 0.04). Age is often employed in four-polar BIA algorithms because of an increase in their accuracy (Guo et al. 1996). However, this was not observed in this study, performed on subjects aged 21-82 years. Even if a sample including also children and adolescents would be better suited to test the hypothesis of age independence of eight-polar BIA, this evidence is nonetheless interesting and worth of discussion. Age is supposed to enter BIA algorithms as a surrogate marker of bioelectrical properties of the body that change with age . These properties are not well defined but changes in body fat distribution and in the relative proportions of FFM components are likely to play a role. A simple explanation for the apparent age independence of eight-polar BIA may be that with this technique the a.c. transverses both hemisomes as opposed to one hemisome in fourpolar BIA. To test this hypothesis more thoroughly, four-polar and eight-polar BIA should be directly compared on a sample of individuals widely differing in age.
The RI sum500 Â sex and RI leg500 Â interactions were not significant and the inclusion of the significant RI arm500 Â sex interaction among predictors did not improve the accuracy of the estimate of LTM arm from RI arm500 . This was observed also in our previous work on TBW (Bedogni et al. 2002). Sex is commonly employed in BIA algorithms because it increases the accuracy of the prediction (Guo et al. 1996). However, this did not happen in the present study. Sex is supposed to enter BIA predictive algorithms as a surrogate marker of bioelectrical properties of the body that differ with sex. These properties are not well defined but differences in body fat and FFM composition may play a role. Even if it is attractive to speculate Abbreviations: a 0 , intercept; a 1 , slope; R 2 adj , adjusted coefficient of determination; RMSE, root mean square error; BIA, bioelectrical impedance analysis: DXA, dual-energy X-ray absorptiometry; FFM, fat-free mass; LTM, lean tissue mass; RI x , resistance index at x kHz; sum, sum of segmental resistances (arms, trunk and legs). that eight-polar BIA can detect these differences, the limitations of this study should be kept in mind. An acceptable power (i.e.>80%) for testing the hypothesis of no RI Â sex interaction was present only for RI arm500 Â sex and a larger sample size is needed to test the hypotheses of no RI 500 Â sex and RI 500leg Â sex interaction. However, the decision of including them in a predictive model would depend substantially on the increase in the underlying accuracy.
Higher frequencies were better than lower frequencies in estimating FFM and appendicular LTM from BIA . As expected by electrical theory, the variance of FFM and appendicular LTM explained by R increased for increasing frequencies (table 3). This increase was lower when R was used as the denominator of Ht 2 in the form of RI because of the already high correlation between Ht, FFM and appendicular LTM. However, the expected pattern was still present and was especially evident for LTM leg .
In conclusion, this study shows that eight-polar BIA is an accurate method for the assessment of FFM and appendicular LTM. The rapidity with which measurements are performed make eight-polar BIA a good candidate for use in epidemiological studies of sarcopenia. The attractive hypothesis that eight-polar BIA is minimally influenced by age and sex should be tested on larger samples including younger individuals.