Variations in bioelectrical impedance devices impact raw measures comparisons and subsequent prediction of body composition using recommended estimation equations

SUMMARY

Background & aims:

Bioelectrical impedance analysis (BIA) for body composition estimation is increasingly used in clinical and field settings to guide nutrition and training programs. Due to variations among BIA devices and the proprietary prediction equations used, studies have recommended the use of raw measures of resistance (R) and reactance (X_c) within population-specific equations to predict body composition.

Objective:

We compared raw measures from three BIA devices to assess inter-device variation and the impact of differences on body composition estimations.

Methods:

Raw R, X_c, impedance (Z) parameters were measured on a calibrated phantom and athletes using tetrapolar supine (BIA_SUP4), octapolar supine (BIA_SUP8), and octapolar standing (BIA_STA8) devices. Measures of R and Xc were compared across devices and graphed using BIA vector analysis (BIVA) and raw parameters were entered into recommended athlete-specific equations for predicting fat-free mass (FFM) and appendicular lean soft tissue (ALST). Whole-body FFM and regional ALST were compared across devices and to a criterion five-compartment (5C) model and dual energy X-ray absorptiometry for ALST.

Results:

Data from 73 (23.2 ± 4.8 y) athletes were included in the analyses. Technical differences were observed between Z (range 12.2–50.1Ω) measures on the calibrated phantom. Differences in whole-body impedance were apparent due to posture (technological) and electrode placement (biological) factors. This resulted in raw measures for all three devices showing greater dehydration on BIVA compared to published norms for athletes using a separate BIA device. Compared to the 5C FFM, significant differences (p < 0.05) were observed on all three equations for BIA_SUP8 and BIA_STA8, with constant error (CE) from −2.7 to −4.6 kg; no difference was observed for BIA_SUP4 or when device-specific algorithms were used. Published equations resulted in differences as large as 8.8 kg FFM among BIA devices. For ALST, even after a correction in the error of the published empirical equation, all three devices showed significant (p < 0.01) CE from −1.6 to −2.9 kg.

Conclusions:

Raw bioimpedance measurements differ among devices due to technical, technological, and biological factors, limiting interchangeability of data across BIA systems. Professionals should be aware of these factors when purchasing systems, comparing data to published reference ranges, or when applying published empirical body composition prediction equations.

1. Introduction

Body composition is increasingly recognized for its role as a diagnostic and prognostic measure in clinical practice, with a similar role in field measurement to guide nutritional and exercise interventions [1]. Body composition assessment by bioelectrical impedance analysis (BIA) garners attention across clinical and field settings owing to its non-invasive measurement, low-cost, ease of use, and rapidity, allowing for the collection of longitudinal change or use in large-scale studies [2,3]. BIA devices are commonly supplied with proprietary empirical equations or generalized (non-population specific) empirical equations to estimate body composition parameters such as fat-free mass (FFM), fat mass (FM), or appendicular lean soft tissue (ALST) [4]. However, the accuracy of these equations has been challenged as they are assumed to be developed in healthy adult populations, limiting their broad applicability [5].

The ability to directly measure raw impedance measurements of resistance (R; a surrogate of total body fluids), reactance (Xc; a reflection of cell membrane integrity and cellularity) and phase angle (PhA; a marker of overall cellular health) also provides distinct advantages over other body composition assessment methods as bioelectrical impedance vector analysis (BIVA, plotted with parameters of R and Xc adjusted by height) is effective in monitoring changes in hydration both in athletes and clinical populations [6-8]. These raw parameters preclude the use of body composition prediction equations that may be inaccurate at the individual level and thus can be used to track changes in fluid levels and cellular health, besides being beneficial for monitoring injury, recovery, and training adaptations [7,9]. Raw bioimpedance parameters are better predictors of factors such as handgrip strength compared to estimations of body composition [10]. Similarly, PhA is a predictor of bone density and muscular performance including handgrip strength, jump performance, and sprint time in addition to its value in predicting clinical outcomes [7,11-15].

Differences in measures between BIA systems result from variation in the type (technical) and placement of electrodes (biological), as well as number of electrodes and subject positioning (technological) among different BIA machines [16,17] that could limit comparison of raw parameters of PhA and BIVA and particularly when used in recommended body composition estimation equations [18-20]. The resulting differences in impedance could complicate body composition estimation and subsequent classification of individuals, particularly when comparing results across devices.

Though these differences have impacts across clinical settings, athletes serve as an ideal sample to explore the impact of BIA devices on the assessment of FFM as hydration, body size, and muscularity vary based on training and competition in sport. This population enables us to explore how differences in raw bioimpedance measurements impact the agreement of these values in recommended population-specific empirical body composition estimations in comparison to criterion methods, as any inaccuracies could adversely impact training or nutrition strategies and result in failing to qualify for a specific sporting event [21]. While the hydration of athletes differs from clinical populations, the variance observed in this population reflects variance in hydration that occurs from a diverse range of clinical conditions.

The purpose of this analysis was to [1]: to assess the inter-device variability in raw BIA parameters due to technical, technological, and biological variation and [2] identify how the differences in raw measures between devices impact body composition estimates when applied to recommended BIA prediction equations. We expected that differences between BIA methods would result in raw impedance differences that limit their comparison across devices. Further, differences would limit the ability to broadly use raw BIA measures across devices within published prediction equations for body composition assessment.

2. Methods

This analysis was a cross-sectional comparative measures study on a sample of convenience recruited from student-athletes at the University of Hawai'i at Manoa, known as the Da Kine study [22]. Recruitment included student athletes participating across a variety of sports, with a minimum age of 18 y for enrollment. This study was approved by the University of Hawai'i Research Compliance and Institutional Review Board (IRB), Protocol # 2018-01102.

Participants: The purpose of this study was to recruit a sample of athletes to understand how varying hydration status and total body water (TBW) may lead to inaccurate body composition measurements when assuming normal hydration. Participants were excluded if they were pregnant or breastfeeding, had metal implants (knee, hip, shoulder), had received body-altering surgical procedures (ex; liposuction or breast reduction/enhancement), or taking performance enhancing drugs or other medications. Written consent was obtained from all participants. To standardize testing conditions, participants were required to abstain from alcohol and caffeine at least 8 h before testing and avoid extraneous exercise for 24 h. On the day of the test, participants arrived at the University of Hawai'i Cancer Center body composition laboratory in the morning (8 AM). Participants confirmed they were overnight fasted and properly hydrated. They were then instructed to void prior to testing. Height and body mass (BM) were measured to the nearest 0.1 cm and 0.1 kg, respectively with a stadiometer/digital scale combination (Seca 264, Chino, CA, USA). A complete overview of the study protocols has previously been reported, including the diversity of sport that can impact training and, subsequently, body composition [22,23]. That said, details of each measurement method and their purpose for use in the criterion multicompartment model of body composition are provided below.

2.1. Bioelectrical impedance analysis (BIA)

Three BIA devices were included in the analysis, two supine (BIA_SUP) and one standing device (BIA_STA). Raw parameters of resistance (R), reactance (Xc), and phase angle (PhA) were recorded in duplicate and compared as the primary objective of the analysis. Raw parameters graphically presented in BIVA were compared to athlete-specific reference values published previously by Campa et al. (2019) using data from a supine tetrapolar RJL/Akern BIA-101 BIA (RJL/Akern Systems, Clinton Township, MI, USA) device [24]. Next, the raw parameters along with demographics data were used within the proprietary algorithms respective of each device to estimate fat-free mass (FFM), fat mass (FM), and body fat percent (BF %) to compare to a criterion multicompartment model of body composition. The raw parameters were used in two published empirical estimation equations suggested for use in athletic populations [25,26]. Specifically, an equation for TBW was included due to the potential for BIA-based TBW estimations for more accessible multicompartment body composition estimation [27]. The two equations and the device used to generate these equations are provided, below:

$$\begin{gathered} (Matiasetal.,2016;RJL∕AkernBIA\text{-}101): \\ TBW=0.286+0.195\ast RI+0.385\ast Wt+5.086\ast sex;male=1 \end{gathered}$$ (1)

$$\begin{gathered} (Matiasetal.,2021;RJL∕AkernBIA\text{-}101): \\ FFM=-2.261+0.327\ast RI+0.525\ast Wt+5.462\ast sex;male=1 \end{gathered}$$ (2)

For each equation, resistance index (RI) was expressed as height (Ht) in centimeters squared and resistance (R) measured at 50 kHz on the right side of the body using a tetrapolar approach, with weight (Wt) measured in kilograms (kg).

The raw impedance parameters were then included into an empirical equation (Sardinha et al., 2020 using an RJL/Akern BIA-101) for the prediction of appendicular lean soft tissue (ALST) in athletes. The bone- and fat-free soft tissue of the arms and legs primarily reflects muscle mass, a key component for athletic performance and monitoring responses to training or disuse [28]. The ALST predicted from the BIA devices were compared to the ALST derived from dual energy X-ray absorptiometry (DXA). The BIA ALST prediction equations for arms and legs are presented in Eqs. (3) and (4):

$$\begin{gathered} ALS{T}_{arms}=0.940\ast sex+0.042\ast Wt+0.080\ast RI+0.024\ast Xc-3.927; \\ male=0 \end{gathered}$$ (3)

$$\begin{gathered} ALS{T}_{legs}=1.983\ast sex+0.154\ast Wt+0.127\ast RI-1.147; \\ male=0 \end{gathered}$$ (4)

For these two equations, weight (Wt) was measured in kilograms (kg), and resistance (R) reactance (Xc) were measured at 50 kHz on the right side of the body using a tetrapolar approach. The RI was calculated as height (in cm) squared over resistance (ht²/R). The estimated ALST from Eq. (3) and Eq. (4) were summed for comparison to DXA ALST ([DXA arm lean + leg lean] − [DXA arm BMC + leg BMC]).

Finally, the estimated body composition parameters from the proprietary device-specific prediction equations were also compared to the criterion multicompartment model to determine the accuracy of the integrated algorithms. To note, because BIA technology reflects the conductive materials of the FFM, these devices provide estimations of FFM (ALST + bone mass) only. As a result, we were unable compare the device-specific estimations to DXA ALST.

The BIA_SUP4 device used was the ImpediMed SFB7 (ImpediMed Ltd, Brisbane, Queensland, Australia), a bioimpedance spectroscopy tetrapolar device. The SFB7 device includes a resistor/capacitor circuit along with manufacturer-provided calibration procedures. Each day prior to testing, leads were attached to the test circuit and a test was performed to ensure proper modeling within ±5 Ω of the calibration circuit. Participants were required to lay on the scan table for 10 min prior to testing to ensure fluid normalization [18]. Arms were separated to roughly 30° from the torso and the legs were separated to ensure no contact between limbs. The tetrapolar system consists of two source and two sensor leads connected to four electrodes (3 M Red Dot Ag/AgCl; 3 M Health Care, St. Paul, MN, USA). Two electrodes were placed on the right hand with the sensor electrode on the wrist in line with the ulna and the source electrode on the hand at least five centimeters apart. Two electrodes were placed on the right foot with the sensor electrode on the dorsal surface of the ankle between the lateral and medial malleoli with the source electrode being placed on the foot at least five centimeters apart. The device tests over a range of 256 frequencies, with a proprietary manufacturer algorithm based on Hanai mixture theory for bioimpedance spectroscopy used for body composition estimation [29]. Measures at 50 kHz were used for raw bioimpedance comparisons. The measurement technique of this device most closely resembles the BIA devices used to generate BIA prediction equations including the same tetrapolar electrode configuration and placement as well as testing position.

The BIA_STA8 device used in this study was the ImpediMed SOZO (ImpediMed Ltd, Brisbane, Queensland, Australia), a bioimpedance spectroscopy octapolar device with two stainless steel electrodes in contact with the bottom of each foot as well as the palms of the hands. The SOZO device contains an internal calibration circuit that performs an internal calibration upon system startup; no error codes were reported throughout the study. Participants were required to stand for 5 min prior to standing testing to ensure fluid normalization [18]. Data (height, weight, age, and sex) was entered via a Bluetooth-connected tablet computer application. Hands and feet were cleaned with an isopropyl wipe prior to stepping onto the device. The participant stood on the foot electrodes and placed hands on the hand electrodes. The device provides an octapolar electrode configuration, however the tetrapolar right side 50 kHz measurement was used for the raw impedance analysis to ensure equal comparison between measurement methods. The device uses the same approach as the BIA_SUP4 to estimate body composition.

The BIA_SUP8 device used was the InBody S10 (InBody USA, Cerritos, CA, USA), a multifrequency octapolar bioimpedance device. The S10 device contains an internal calibration circuit that performs an internal calibration upon system startup; no error codes were reported throughout the study. The S10 touch-type system uses silver chloride (AgCl) touch-type finger electrodes on the thumb and middle fingers of the hands, while the foot electrodes are placed between the ankle bone and the heel, based on manufacturer recommendations. To note, the S10 system also offers the ability to utilize adhesives in place of the touch-type electrodes, however these were not used in this study. Recordings of segmental impedance and reactance of the right side of the body were used to calculate resistance. Whole-body phase angle from the device reflects the 50 kHz right-sided measurement of tetrapolar BIA methods, thus this value was used for the raw comparison.

Testing of the supine devices was performed in a randomized fashion following the required 10 min to allow for fluid normalization [18]. Participants were then required to stand for 5 min to allow fluid compartmentalization to the legs prior to completing the standing test to minimize potential effects of insensible fluid loss.

In order to further explore the technical variation in raw bioimpedance parameters, we created a circuit board calibration phantom with separate resistors for the arms, legs, and trunk to mimic the resistance and capacitance effects observed when testing BIA in the human body [16]. We attached 3M adhesive electrodes to the circuit board clips and the electrodes on each BIA device. Using this known circuit of resistors, we measured resistance, reactance, and impedance at 50 kHz on the right side of the body (to account for the availability of only whole-body measures available on the BIA_SUP4) for all three devices to compare measured impedance to the known impedance on the resistor board. This allowed us to define the agreement to a known resistance to account for technical variation between BIA devices that is separate from the biological (tissue properties) or technological variation that could occur from the differences in postural fluid shifts or electrode placement, respectively.

2.2. Muticompartment body composition

The multicompartment model of body composition served as the criterion FFM measurement. Multicompartment models provide accurate measures of FFM regardless of hydration status, which we previously showed to vary in this sample of athletes [22]. Because this model was used to generate the FFM prediction in Eq. 2, we used this approach for validation purposes. Based on the Wang et al. (2002) five-compartment (5C) model [30] the model utilizes TBW from deuterium dilution (D₂O), bone mineral content (BMC) from DXA, body volume (BV) from air displacement plethysmography (ADP), and body mass (BM) from scale weight to obtain fat mass composition, as shown in Eq. (5):

$$F{M}_{5C}=2.748\ast BV-0.715\ast TBW+1.129\ast M_o+1.222\ast M_s-2.051\ast BM(30)$$ (5)

Where osseous mineral (${\mathrm{M}}_{\mathrm{o}}=\text{BMC}\ast 1.0436$) and soft tissue mineral (${\mathrm{M}}_{\mathrm{s}}=\text{TBW}\ast 0.0129$) were calculated.

FFM was derived by subtracting the FM_5C from body mass.

Each participant completed whole-body DXA scans on a Hologic Discovery/A system (Hologic Inc., Marlborough, MA, USA) for BMC, and whole-body BV measurements using a BodPod (v5.4.1, COSMED USA, Inc., Concord, CA, USA). For TBW and Ms measures, participants consumed a 30 g (99.9% pure) dose of D₂O (Cambridge Isotope Laboratories, Inc., Tewksbury, MA, USA) in 100 mL drinking water. Two post-dose saliva samples were obtained at three- and 4-h time points which were compared to ensure equilibration <5% difference, and the 4-h time point was used for the determination of TBW.

2.3. Appendicular lean soft tissue (ALST)

Measures of arm and leg lean soft tissue were used to construct the criterion measure of ALST. Segment-specific bone and fat mass were subtracted from regional DXA mass measures to obtain ALST.

2.4. Statistical methods

Kolmogorov-Smirnov with Lilliefors-correction and Shapiro-Wilk tests were performed to confirm normal distributions of data. Data with normal distribution are expressed as mean ± standard deviation (SD). A within-subjects ANOVA was performed to assess the agreement between raw bioimpedance parameters (R, Xc, PhA) among the three devices. The raw resistance, reactance, and phase angle values were compared via linear regression, with the BIA_STA8 selected for the x-axis for comparison to the two supine devices. Test-retest precision of duplicate measurements using the same BIA device were compared (via root mean square error, RMSE) to the differences in raw measurements among devices to model the errors associated with BIA. The BIVA Excel software by Piccoli & Pastori (2002) was used to generate BIVA graphs for each BIA method [31]. Additionally, raw bioimpedance parameters were used for the calculation of FFM and ALST based on equations recommended for use in athletic populations [5,25,26,28]. Of note, because the Matias et al. (2021) equation is used to predict TBW, we used the theoretical hydration constant of 73.2% (FFM = TBW/0.732) to estimate FFM [32]. Differences between these estimations and the criterion are reported via a within-subjects ANOVA for FFM. Similarly, the DXA ALST was compared to the device-specific predictions using a within-subjects ANOVA.

For comparisons to criterion measurements using the device-calculated body composition, a within-subjects ANOVA was performed for FFM, FM, and BF%. Constant error (CE) between methods was calculated as the mean difference from each alternate method to criterion (CE = alternate – criterion). Lin's concordance correlation coefficients (CCC) were calculated to assess accuracy and precision in body composition estimates by each BIA compared to criterion using the scale: poor (<0.90), moderate (0.90–0.95), substantial (0.95–0.99); almost perfect (>0.99) [33]. Bland-Altman analyses were performed to evaluate the agreement and potential bias between methods for all estimations of body composition, while confidence intervals of the 95% limits of agreement (LOA) were calculated to determine the likelihood of the LOAs containing 95% of the population [34]. Statistical analysis was performed using SAS version 9.4 for Windows (SAS Institute, Cary, NC, USA).

3. Results

Athletes ≥18 years and were competing in-season or participating in an off-season strength and conditioning routine specific to their sport were recruited. Of the 80 subjects recruited, a total of five participants had their data removed due to having PhA values that were deemed to be abnormal (<4°, n=4; >15°, n = 1) on at least one BIA device based on normal ranges reported by Norman et al. (2012) [7]. Additionally, two subjects were removed due to failure to save data on BIA devices, leaving a total of 73 participants for the raw bioimpedance data analysis. The demographic data for these participants is presented in Table 1. Because bioimpedance data is reliant upon proper height and weight values for body composition estimation, two subjects were removed from the body composition comparisons due to height or weight input errors on one or more BIA device, resulting in a sample of 71 (34 female) participants for body composition comparisons [7]. A full STROBE breakdown of exclusions is presented in Supplemental Fig. 1.

Table 1.

Participant demographics (n = 73, 35 female).

Variable	Mean ± SD	Min	Max	Males (n = 38)	Females (n = 35)
Age (years)	23.2 ± 4.8	18.0	37.0	24.5 ± 5.2	21.8 ± 3.8b
Height (cm)	175.0 ± 11.5	154.7	203.0	181.1 ± 10.2	168.5 ± 9.3b
Weight (kg)	73.7 ± 14.6	43.9	102.5	82.7 ± 10.3	64.1 ± 12.8b
BMI (kg/m2)	24.0 ± 3.6	17.8	32.8	25.3 ± 3.3	22.5 ± 3.5b
5C FFMa (kg)	60.7 ± 13.9	33.9	84.3	71.5 ± 8.5	49.1 ± 8.2b
5C FMa (kg)	13.1 ± 7.2	3.2	46.3	11.2 ± 5.8	15.1 ± 8.0b
5C FMa (%)	17.9 ± 8.7	4.5	45.5	13.3 ± 6.2	22.7 ± 8.4b

Abbreviations: 5C: 5-compartment Wang (2002) model, cm: centimeters, FFM: fat-free mass, FM: fat mass, kg: kilograms, m: meters, SD: standard deviation.

^an = 71.

^bValues differed significantly (all p < 0.05) compared to males.

3.1. Technical variation with a known resistor

The assessment of technical variation among the three devices is presented in Supplemental Table 1. Using a known resistor with an Z of 598.8Ω on the right side of the body, the BIA_SUP4 was 50Ω greater (648.9Ω), while the BIA_SUP8 and the BIA_STA8 were both within 15Ω (611.0Ω for BIA_SUP8 and 614.7Ω for BIA_STA8), indicating a technical variation between BIA devices. While a small technical variation of <15Ω is to be expected in electrical circuitry models, these differences occur regardless of the biological variation of posture or electrode positioning, indicating a limited ability to interchange the raw impedance parameters between devices.

3.2. Test-retest precision vs. inter-device differences

Technical error resulting from duplicate testing are presented in Supplemental Fig. 2. The test-retest measures (RMSEs) were greatest for BIA_STA8 (43.6 Ω), which accounted for 7.8% error between duplicate measurements. The range of inter-device differences between the three BIA devices (−62.7 to +123.9 Ω) accounted for up to 24.6% error in agreement between BIA devices. For reactance values, RMSE values were greatest for BIA_SUP4(24.3 Ω), which accounted for up to 39.2% average error between duplicate measurements. The range of differences between the three BIA devices (−10.8 to +27.1) accounted for up to 43.7% error between BIA devices.

3.3. Raw measures comparisons

To compare the effects of biological differences (electrode placement, testing position), the raw bioimpedance parameters from each BIA device were compared and presented in Fig. 1. The measured R was similar between BIA_SUP8 (565.9 ± 93.6Ω) and BIA_STA8 (556.2 ± 85.7Ω), with both values significantly different (p < 0.05) from the BIA_SUP4 device (503.9 ± 85.8Ω). This is despite the higher measured Z for BIA_SUP4 using the calibration circuit phantom, indicating the electrode placement (on the wrist next to the ulnar head and on the ankle at the malleoli) results in a significantly lower measured impedance compared to the electrode placement of the remaining two BIA devices (technological and biological variance). These effects resulted in a greater RI for BIA_SUP4of 6.9 cm²/Ω and 6.0 cm²/Ω compared to BIA_SUP8 and BIA_STA8, respectively.

Fig. 1.

Comparison of BIA devices at 50 kHz frequency for R (A), RI (B), Xc (C), and PhA (D) (n = 73, 35 female). Data are expressed as mean ± standard deviation and reported as R (Ω), RI (cm²/Ω), Xc (Ω), and PhA (°). Abbreviations: BIA_STA8: ImpediMed SOZO, BIA_SUP4: ImpediMed SFB7, BIA_SUP8: InBody S10, PhA: Phase angle, R: resistance, RI: resistance index, Xc: reactance. * significant (p < 0.05) mean difference between devices.

For Xc, BIA_SUP8 device reported a significantly higher measured reactance value (68.0 ± 9.0U; p < 0.05) compared to the BIA_SUP4 (62.0 ± 7.6Ω) and BIA_STA8 (59.8 ± 6.6Ω). While these differences have smaller effects in prediction equations using Xc, these differences resulted in the BIA_STA8 having a significantly lower PhA value (p < 0.05) compared to the two supine measurements. Individual differences and group-level comparisons are presented in Supplemental Fig. 3, where differences among some individuals were large (For example, a high Xc [100.6 Ω] for BIASUP8 compared to a similar Xc for BIASUP4 and BIASTA8 [73.5–76.3 Ω] resulted in a 2.3° greater PhA for the BIASUP8 device). That said, significant heterogeneity in agreement was observed across devices for all measures. To reduce the discrepancies between devices, we provide Deming regression models for R, Xc, and TBW measures between BIA devices in Supplemental Table 2.

3.4. Raw measures in BIVA

When applying the raw bioimpedance parameters to the BIVA vector graph for clinical interpretation (Fig. 2), observable differences between the tolerance ellipses were present, where no overlap occurred between ellipses of any of the three BIA methods. The BIA_STA8 and BIA_SUP8 ellipses had no overlap with the Campa et al. (2019) reference range for athletes, meaning these devices did not conform to the reference range previously published [24]. Despite the similarity in technique of the BIA device used by Campa et al. (2019) and the BIA_SUP4 devices, the BIA_SUP4 only had roughly 50% overlap and showed greater average resistance (suggesting lesser hydration). Both the BIA_SUP8 and BIA_STA8 showed even lesser hydration values, limiting comparison of data between these devices and the reference values.

Fig. 2.

Bioelectrical impedance vector analysis (BIVA) for each BIA device (n = 73). Center of circle represents group mean, with 95% tolerance ellipse. Devices: BIA_STA8: ImpediMed SOZO, BIA_SUP4: ImpediMed SFB7, BIA_SUP8: InBody S10, Campa₂₀₁₉: Akern/RJL BIA-101.

3.5. Raw measures for predicting body composition

The impact of the different device-specific measures on body composition estimates compared to the criterion methods are presented in Table 2. A statistically significant (p < 0.05) difference in FFM was observed between the criterion and both BIA_STA8 and BIA_SUP8 for both equations, with no significant differences observed between sexes (p = 0.016). The >2 kg mean difference between BIA_SUP4 and the other BIA devices was the result of the large impact of the RI, noted previously, on predicting FFM in this equation (RI * 0.327). Figure 3 shows the criterion methods for TBW, FFM, and ALST as well as the equation-predicted values for each of the three BIA devices. Overall, both equations resulted in a lower TBW or FFM prediction for all three BIA devices. While the interquartile range for TBW was larger for BIASUP8 and smaller for BIASTA8 compared to the criterion, all three devices showed larger interquartile ranges for FFM estimation.

Table 2.

Body composition estimates based on raw bioimpedance parameters used in prediction equations developed for athletes (n = 73, 35 female).

	Criterion	Matias (2016)a			Matias (2021)			Sardinha (2020)
	Mean (SD)	BIASUP8	BIASUP4	BIASTA8	BIASUP8	BIASUP4	BIASTA8	BIASUP8	BIASUP4	BIASTA8
FFM (kg)	60.7 ± 13.9	57.8 ± 13.7b	59.6 ± 14.1	58.0 ± 13.5b	57.7 ± 13.9b	59.9 ± 14.3	57.9 ± 13.7b
TBW (L)	42.8 ± 9.9	42.1 ± 9.9b	43.5 ± 10.2	42.3 ± 9.8b
ALST (kg)	26.9 ± 6.7							24.0 ± 4.3c	25.3 ± 4.5c	24.0 ± 4.2c

Criterion methods for FFM: Wang_5C multicompartment model, TBW: deuterium dilution, ALST: dual energy X-ray absorptiometry.

Abbreviations: ALST: appendicular lean soft tissue, BIA_STA8: ImpediMed SOZO, BIA_SUP4: ImpediMed SFB7, BIA_SUP8: InBody S10, FFM: fat free mass, kg: kilograms, L: liters, TBW: total body water.

^aMatias (2016) calculates TBW. FFM was calculated using the equation FFM = TBW/0.732.

^bMean value significantly (p < 0.05) differs from criterion.

^cMean value significantly (p < 0.01) differs from criterion.

Fig. 3.

Comparison of BIA devices for fat-free mass (a, b, c) and appendicular lean soft tissue (d) compared to criterion methods (n = 73, 35 female)

Box-whisker plots represent the interquartile range expressed as the shaded box, with the mean (diamond) and median (blue line) of the measures for each device. Equations used were Matias (2016, left) and Matias (2021, right). Data are expressed as mean ± standard deviation and reported as total body water (L), fat-free mass (kg). Abbreviations: 5C: Wang multicompartment model, BIA_STA8: ImpediMed SOZO, BIA_SUP4: ImpediMed SFB7, BIA_SUP8: InBody S10, FFM: fat-free mass, TBW: total body water. * significant (p < 0.05) mean difference between measurements.

When applying the device-specific algorithms (Supplemental Table 3), a nonsignificant (all p > 0.05) difference was observed for all three BIA devices when compared to the criterion multicompartment model for FFM, FM, and BF%. The improved agreement in addition to the high concordance correlation prompted the comparison of the agreement between empirical-prediction equations and the device-specific proprietary equations, which are shown in Fig. 4. While the mean differences for each BIA device to the criterion were smaller compared to the published prediction equations, considerable improvement in the 95% limits of agreement (LoAs) show a clear improvement of the device-specific equations. The largest improvement in LoAs was for BIA_SUP8, with improvements from −5.9–12.1 kg to −5.2–6.2 kg. Notably, the significant outliers (>10 kg difference) observed for the three BIA devices when using published prediction equations were improved and within LoAs with the device-specific formulas, though outliers >10 kg remained (n = 3) for BIA_SUP4 and BIA_STA8.

Fig. 4.

Bland-Altman comparisons of proprietary device-specific fat-free mass estimations in comparison to the criterion Wang 5C model (n = 71). Devices: BIA_STA8: ImpediMed SOZO, BIA_SUP4: ImpediMed SFB7, BIA_SUP8: InBody S10. Abbreviations: FFM: fat-free mass, kg: kilogram, 5C: Wang multicompartment model.

Performing comparisons for ALST showed that all three equations were significantly (p < 0.05) different compared to DXA ALST (mean differences all >1.6 kg), with a significant bias across the study sample. As observed in Fig. 5, a compressed interquartile range for all three devices prompted further investigation into the overall agreement. Sex-specific comparisons resulted in significant (all p < 0.01) sex-specific differences in agreement for all devices, with overestimation of ALST in females and underestimation in males. After substituting the coding for sex variables (male = 1 instead of male = 0), the absolute agreement (Fig. 6) and LoAs (Fig. 6) were greatly improved and sex-specific differences eliminated (all p > 0.05), indicating an error in reporting of the equations. Ultimately, the use of the correct sex variable resulted in an improvement of LoAs, with the largest LoAs (BIASTA8) improving from ±6.9 kg to ±1.8 kg. After correction, a statistically significant (p < 0.01) difference remained for each BIA device compared to DXA, despite a better overall average difference of 1.5 ± 2.0 kg for BIA_SUP4.

Fig. 5.

Bland-Altman comparisons of proprietary device-specific fat-free mass estimations in comparison to the criterion Wang 5C model (n = 71). Devices: BIA_STA8: ImpediMed SOZO, BIA_SUP4: ImpediMed SFB7, BIA_SUP8: InBody S10. Abbreviations: ALST: appendicular lean soft tissue, kg: kilogram.

Fig. 6.

Comparison of BIA devices for appendicular lean soft tissue before and after equation correction (n = 73, 35 female). Box-whisker plots represent the interquartile range expressed as the shaded box, with the mean (diamond) and median (blue line) of the measures for each device using the Sardinha (2020) equation as reported (left) and after correction of sex variable (right). Data are expressed as mean ± standard deviation and reported as appendicular lean soft tissue (kg). Abbreviations: ALST: appendicular lean soft tissue, BIA_STA8: ImpediMed SOZO, BIA_SUP4: ImpediMed SFB7, BIA_SUP8: InBody S10, DXA; dual energy X-ray absorptiometry. * significant (p < 0.05) mean difference between measurements.

4. Discussion

This study shows the significant group and, more importantly, individual differences in body composition estimations that occur when raw bioimpedance parameters are used from BIA devices with different technologies. While other studies have reported the differences in raw impedance factors, we show the impact when applying these impedance parameters to equations reported to provide accurate body composition estimates broadly across BIA systems [35]. We observed that technical (device-specific variance), technological (differences in electrodes, posture, body regions measured) and biological (location of electrodes) differences between measured impedance parameters impacted body composition estimation and classification of hydration status. Our results failed to find improvements when applying published raw prediction approaches over the use of device-specific proprietary estimation equations. These findings are significant as interchangeable use of BIA technologies could impact tracking of hydration and body composition, critical factors when monitoring health status, risk, and outcomes related to various interventions.

Because the biological, technical, and technological differences between the three BIA devices used in this study resulted in significant differences in the measures or R and Xc, the use of these raw parameters in BIVA graphs resulted in different tolerance ellipses with no discernible overlap across all three devices. While the BIA_SUP4 device had roughly 50% overlap to the BIVA reference presented for use in athletes, the difference in raw measures (or populations) resulted in the interpretation of “lesser hydration” in our sample compared to the proposed athletic reference group. Though the device used for these references (RJL/Akern BIA-101) are most similar to the BIA_SUP4 used in this study, Guedes et al. (2023) [36] showed that the raw parameters between two tetrapolar, supine-based BIA devices had differences in raw parameters that impact their cross-comparison. The results of this study show that the differences between devices of similar approaches still show effects that impact their comparison to these previously published reference ranges.

When performing duplicate measures of the same BIA devices, the amount of error observed within devices resulted in large differences in raw intra-device measures across all three devices, resulting in inter-device differences in R up to 24.6% and Xc up to 43.7%. These differences could be the result of factors such as the orthostatic and compartmental (intracellular to extracellular) fluid shifts observed when standing/laying supine, or due to changes in whole-body and regional impedance due to the differences in contribution of trunk and limbs to overall body impedance [18,37]. That said, the use of a known calibration tool causing differences in the raw impedance measures also shows the limited ability to compare across devices. Results from these in vitro and in vivo analyses show that without any changes in the placement of electrodes or testing conditions, factors such as biological or technical differences can impact the raw impedance measures across BIA devices [38]. Further, biological differences appear to contribute to a higher PhA during supine testing as compared to the standing position [39]. This highlights the importance of using consistent measurement techniques for tracking as well as the need to utilize the proper measurement device (and equation, when necessary) for comparison to a reference to avoid erroneous comparisons [40]. Also importantly, this highlights how large-scale meta-analyses of PhA and BIVA may be skewed by the interchangeable use of the different systems or reliance on a single system to derive measurements [41]. Further, these results identify a need for additional data from these various devices to serve as references for researchers and practitioners.

To derive body composition estimations from raw BIA predictions, manufacturers have developed proprietary prediction equations [16]. Due to the proprietary nature of these equations, including the populations from which these equations were derived, researchers have suggested the development and use of population-specific prediction equations for body composition estimation [5,26]. This has ultimately led to the development of prediction equations in athletic populations as well as across clinical conditions [42-44]. Validating these equations using different BIA devices, as performed in this study, shows that the body composition estimations do not agree when predictions are applied across systems. Further, the improvement in the agreement of body composition measures when applying the device-specific prediction equations show that the broad development and recommendation of prediction equations may not capture the complexity of equations developed for specific BIA-devices. Ultimately, the recommendation for the broad use of published equations across devices does not appear to provide robust and accurate body composition estimations.

Finally, we showed that the proper presentation of BIA prediction equations in the literature is essential for accurate body composition estimations. Generally, prediction equations provide an integer that estimates greater FFM in males due to a greater musculature and lesser body fatness compared to females. The adjustment of the prediction for ALST showed a significant improvement in the overall agreement to DXA, further highlighting the potential errors when developing and reporting of estimation equations or the use of these equations for calculating body composition in the field [45]. It also shows how the use of these equations may require validation to the criterion in a small sample of study participants prior to their broadscale use in research or clinical practice [27]. While applying these raw bioimpedance parameters into equations for fat-free mass estimation in athletes appears to be problematic, this effect may be more significant in other populations, particularly clinical populations. Because athletes are generally more focused on athletic performance versus body composition, inaccurate body composition estimations resulting from raw BIA parameters from different devices may be more significant in older adults who are at risk for frailty and sarcopenia [40].

Older adults have lower muscle mass (and subsequently higher bioelectrical resistance), meaning the disagreement between these equations could be larger in older adult populations, underscoring the potential effect of interchanging these equations and the subsequent risk of improper sarcopenia diagnosis in clinical settings [46]. Because our sample was based on a recruitment of athletes with greater musculature (and lower measured impedance) and their deviation from the reported FFM hydration (we observed an average FFM hydration of 70.0 ± 2.0% [22,30]), the comparisons of the estimations provided by the three BIA devices may differ from results observed in other adult populations. We recommend the collection of similar measures in older adults to better understand the impact of raw BIA comparisons and their effects on body composition estimations in at-risk older adult populations. Because we assumed euhydrated status of these individuals prior to testing, further work should also explore the impact of hydration on the agreement between BIA and criterion measures [47].

Population-level data reference data has proven to be pivotal in identifying abnormal body composition associated with increased risk for a variety of clinical conditions [48-50]. In athletics and in the general population, the rapidity of BIA assessments makes this technology an appealing method to estimate body composition for the assessment of disease risk, monitor changes from diet and exercise, and evaluate hydration status. However, clinicians should be aware of how the biological, technological, and technical differences among BIA devices influence the interpretation and comparison of data across systems. We recommend that professionals: 1) ensure proper comparisons of raw BIA measures to systems with similar configurations, 2) if system differences exist, samples be collected to compare across systems and against criterion methods (e.g.; 5C or DXA), and 3) if comparing body composition estimates, that the proprietary device-specific estimations be validated and potentially prioritized over empirical estimations (while acknowledging the accuracy limitations of these equations).

As noted by Ward (2019), a 1994 National Institutes of Health conference on BIA technology recommended researchers and manufacturers collaborate and develop standardizations for BIA manufacturers [51,52]. Since then, more than 30 manufacturers of BIA devices have developed a heterogeneous mix of technologies, measurement techniques, and algorithms used to predict body composition, with minimal standardization performed [52]. Further advances to BIA technology allow for technical differences as well as variations in testing postures (technological) and positions (biological). Our findings show that these differences result in small but significant differences in raw bioimpedance parameters that can impact the ability to compare results across devices or use these measurements in published prediction equations. These results highlight the need for caution when comparing results across BIA devices to avoid creating errors that impact the quality of research or application in clinical practice.

5. Conclusion

Raw bioimpedance parameters differ between BIA devices as a result of technical, technological, and biological differences. This limits the interchangeability of BIA devices, especially comparing between tetrapolar and octapolar devices that differ in all three aspects. Use of raw measures from BIA devices does not bring agreement for measures of PhA, BIVA, and body composition and does not fix the current challenges to the accuracy of BIA. It is recommended that BIA measurements only be compared to the same BIA technology to avoid improper tracking and classification in athletics and clinical practice. Further, we highlight the need for recognition that broad recommendation of a published empirical prediction equation may result in its misuse without clear and precise recommendations.

Abbreviations:

ADP

air displacement plethysmography

AgCl

silver chloride

BIA

bioelectrical impedance analysis

BIA_STA8

standing octapolar BIA

BIA_SUP4

supine tetrapolar BIA

BIA_SUP8

supine octapolar BIA

BIS

bioelectrical impedance spectroscopy

BIVA

bioelectrical impedance vector analysis

BF

body fat

BM

body mass

BMC

bone mineral content

BV

body volume

CCC

Lin's Concordance Correlation Coefficient

D₂O

deuterium dilution

DXA

dual energy X-ray absorptiometry

FFM

fat-free mass

FM

fat mass

Ht

height

LOA

limits of agreement

M_o

osseous mineral

M_s

soft tissue mineral

PhA

phase angle

R

resistance

TBW

total body water

W_t

weight

X_c

reactance

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.clnesp.2024.07.009.