Assessment of bioimpedance spectroscopy devices: a comparative study and error analysis of gold-plated copper electrodes

Objective. Bioimpedance spectroscopy (BIS) is a non-invasive diagnostic tool to derive fluid volume compartments from frequency dependent voltage drops in alternating currents by extrapolating to the extracellular resistance (R 0) and intracellular resistance (R i). Here we tested whether a novel BIS device with reusable and adhesive single-use electrodes produces results which are (in various body positions) equivalent to an established system employing only single-use adhesive electrodes. Approach. Two BIS devices (‘Cella’ and the ‘Body Composition Monitor’ [BCM]) were compared using four dedicated resistance testboxes and by measuring 40 healthy volunteers. In vivo comparisons included supine wrist-to-ankle (WA) reference measurements and wrist-to-wrist (WW) measurements with pre-gelled silver/silver-chloride (Ag/AgCl) electrodes and WW measurements with reusable gold-plated copper electrodes. Main results. Coefficient of variation were <1% for all testbox measurements with both BIS devices. Accuracy was within ±1% of true resistance variability, a threshold which was only exceeded by the Cella device for all resistances in a testbox designed with a low R 0/R i ratio. In vivo, WA-BIS differed significantly between BIS devices (p < 0.001). Reusable WW electrodes exhibited larger resistances than WW-BIS with Ag/AgCl electrodes (R 0: 738.36 and 628.69 Ω; R i: 1508.18 and 1390 Ω) and the relative error varied from 7.6% to 31.1% (R 0) and −15.6% to 37.3% (R i). Significance. Both BIS devices produced equivalent resistances measurements but different estimates of body composition both in silico and in WA setups in vivo, suggesting that the devices should not be used interchangeably. Employing WW reusable electrodes as opposed to WA and WW measurement setups with pre-gelled Ag/AgCl electrodes seems to be associated with measurement variations that are too large for safe clinical use. We recommend further investigations of measurement errors originating from electrode material and current path.


Introduction
Bioimpedance spectroscopy (BIS) is used to assess fluid volume compartments (Moissl et al 2006) and body composition (Chamney et al 2007) by applying weak alternating currents (AC) to biological tissue and measuring voltage drops within this bioelectrical circuit (Cornish et al 1996).Impedance data gathered from frequency sweeps within the b-dispersion range allow extrapolation to theoretical resistances at zero (R 0 ) and infinite frequency (R Inf ).This process is often 'Cole' modelling, referring to the original author (Cole 1928, Cole andCole 1941).Extracellular water (ECW) and total body water (TBW) volumes are indirectly proportional to these resistances and can be approximated with corresponding models (Moissl et al 2006).Depending on the electrode setup (octapolar segmental versus tetrapolar wrist-to-ankle [WA]), measurements can either be taken at face-value or have to be corrected to represent the entire body (De Lorenzo et al 1997, Ward et al 2022).
BIS devices are equipped with modules for (a) signal generation, (b) amplification and (c) conversion to a current signal, (d) signal detection as well as (e) analogue-to-digital conversion (Naranjo-Hernández et al 2020).Judging by patents from the 1990s and early 2000s (Withers 1994, Chamney 2002, Kennedy 2005) and comparing them to today's 'smart' BIS devices (Naranjo-Hernández et al 2020), the technique has remained similar, although the overall size and measurement time have been reduced.Bioimpedance technology is offered in consumer electronic devices such as scales and wearables at an affordable price (Rossi et al 2017, Naranjo-Hernández et al 2019).However, BIS devices designed for medical use still lack versatility in that they are large, more expensive, and frequently dependent upon non-reusable adhesive electrodes.
The 'Body Composition Monitor' (BCM), a whole-body BIS device manufactured by Fresenius Medical Care (FMC, Bad Homburg, Germany), owes its popularity to a dedicated three-compartment body composition model (Chamney et al 2007) distinguishing between euvolemic lean tissue mass (LTM), euvolemic fat tissue mass (FTM) and volume overload (VO).This model has been validated in numerous isotope studies (Wabel et al 2006, Moissl et al 2007, 2008, Wabel et al 2009).VO is of particular interest in patients with chronic kidney disease on haemodialysis (HD) treatment, where accurate information on volume status is important for optimal therapy (Dekker et al 2017, Zoccali et al 2017, Hecking et al 2018).One disadvantage of the BCM, and the original motivation for this study, is its reliance on impractical pre-gelled silver/silver-chloride (Ag/AgCl) electrodes and their cumulative costs for longitudinal measurements.
The present study therefore aimed at fulfilling the following objectives: (I) To compare two BIS devices of different manufacturers (established and new) with equivalent electrical circuit models and in healthy volunteers, (II) to investigate whether wrist-to-wrist (WW) BIS measurements with reusable electrode systems produce similar results as whole-body WA reference measurements with pre-gelled electrodes and, if not, to (III) adapt the former to the latter.

Population and ethics
The present investigation was conducted in healthy volunteers who were recruited by personal invitation, flyers, word-of-mouth, and subsequently online registration in a university hospital (Vienna, Austria) and surrounding areas.Candidates were considered healthy if their medical history, blood, and urine analyses showed no signs of chronic kidney disease.All participants provided written informed consent and were financially compensated for their participation.The study was approved by the Ethics Committee of the Medical University of Vienna (EK #2057/2020).

BIS devices
Bioimpedance measurements were conducted with BCM (FMC) and Cella (Cella Medical, California, USA) BIS devices.A version 3.2.5 BCM (manufactured in 2011) and three version 1.92 Cella devices (all manufactured in 2019) were studied.The BCM was chosen due to its established status in HD practice.The Cella device was used because its three-compartment tissue model is closely related to the BCM's model, therefore allowing a more representative comparison as opposed to BIS devices using other tissue models.Cella Medical was willing to provide custom made electrode equipment (detailed below).
Both manufacturers' frequency sweeps include 50 discrete AC frequencies ranging from 4.88 to 1000.98 kHz (supplemental table S1).The injected currents are 50-800 mA (stepless) and  100 mA for the BCM and Cella, respectively.Measurement precision as provided by the manufacturers is 0.01 Ω impedance, 0.01 degrees phase, 1% (ECW, intracellular water [ICW], TBW) and 4% (Fat, LTM, body cell mass) coefficients of variation for repeated measurements with the BCM and <4% intracellular resistance (R i ) and <2% R e for Cella.Note that impedance and phase, R 0 , R Inf , and R i , as well as fluid volume compartments (ECW, ICW, TBW) and body composition (LTM, FTM, VO) presented herein are taken directly from BCM and Cella devices and software, with equations potentially varying between both manufacturers' devices.No additional data modelling was performed unless specified.
In addition to measurements with standard electrode leads and pre-gelled Ag/AgCl electrodes on wrist and ankle, Cella was also connected to prototype hand-rest ('Board') and hand-held ('Tablet') enclosures (figure 1) for WW measurements via direct skin contact to gold-plated copper plates of printed circuit boards.One Cella device was permanently mounted to standard electrode leads, one was connected to the Tablet device, and one was connected to the Board device to reduce potential wear on the cables by avoiding frequent de-and reattachment.All three Cella devices produced practically identical results in a pre-analysis.The BCM, the smart cards, the Fluid Management Tool (FMT) for data storage and the disposable pre-gelled Ag/AgCl electrodes were commercially supplied by FMC.Cella devices, the Tablet and Board devices and Samsung 4J+ smartphones (Samsung, Seoul, South Korea) which were required to operate the dedicated Cella smartphone application (Alpha version 1.0.1.14)were supplied by Cella Medical.

Resistance testboxes
Testboxes were designed to simulate electrical properties of the human body in a simple equivalent electrical circuit (Kyle et al 2004).The circuit design is laid out in supplemental figure S1.A resistor representing R 0 was soldered in parallel to a serial connection of a resistor for R i and a capacitor emulating the properties of cell membranes.Each electrode connection had a downstream resistive (51 Ω) and capacitive (100 nF) component corresponding to the electrode-to-skin interface.Box A was constructed in analogy to the testbox shipped with the BCM, while the other boxes simulate a healthy individual (Box B ), the median results for BIS measurements of an HD population (Mussnig et al 2023) (Box C ), and an outlier from the same HD population (Box D ).Boxes A−D and their specifications are shown in supplemental figure S2 and are referred to in table 1.

Study procedures
Participants received ten unique BIS measurements as depicted in figure 1. Immediately after arrival at the study facility, volunteers were measured in a sitting position using the Board (Setup 9) and Tablet (Setup 10) device.Next, they were asked to lie down in a supine position for 10 min to achieve fluid equilibrium after which BCM (Setup 2) and Cella (Setup 1) measurements were conducted with adhesive electrodes applied to the wrist and ankle of the same side, as instructed by the manufacturer.Then, Board (Setup 7) and Tablet (Setup 8) measurements were conducted with both legs extended horizontally and the upper body reclined at 30 to 45 degrees to simulate a hospital bed.Board and Tablet measurements were subsequently repeated in a sitting position (Setups 5 and 6).Finally, participants laid down their hands on a chipboard tabletop while remaining seated, adhesive electrodes were applied to the back of both hands, and BCM (Setup 4) and Cella (Setup 3) measurements were repeated.Setups 1-2 are summarized as WA-BIS-Leads, Setups 3-4 as WW-BIS-Leads and Setups 5-10 as WW-BIS-Plates.On principle, each setup was conducted once per participant.BCM measurements were considered invalid if FMC's proprietary quality parameter was <85%.According to the Cella software, a measurement is considered as invalid if one of the following conditions is met on the results provided by the device: R 0 > 1200 or fit error of the equivalent circuit model >7.0 or capacitance <0.7 nF or fc > 300 kHz or fc < 20 kHz.These thresholds were set Figure 1.Illustrations of all in vivo BIS measurements conducted within this study grouped by current path and electrode material (Cella: Setups 1, 3 and 5-10; BCM: Setups 2 and 4).Setups 1 and 2 were carried out after participants lay flat on their backs for 10 min to achieve fluid equilibrium and were followed by (in that order): Setups 7 and 8, 5 and 6 and finally 3 and 4. Setups 9 and 10 took place prior to the 10 minute long supine body positioning right after the participants' arrival at the study facility.This figure was adapted with permission from Mussnig 2023.BCM, Body Composition Monitor; BIS, bioimpedance spectroscopy, WA-BIS-Leads, wrist-toankle BIS measurements with electrode leads and pre-gelled silver/silver-chloride electrodes; WW-BIS-Leads, wrist-to-wrist BIS measurements with electrode leads and pre-gelled silver/silver-chloride electrodes; WW-BIS-Plates, wrist-to-wrist BIS measurements with gold-platted copper electrodes on printed circuit boards.by Cella Medical and could not be altered.Invalid measurements prompted up to three retries after which the specific setup was deemed immeasurable in this participant.Setups were visualized in figure 1 using Blender version 3.0.1 3D animation software (Blender Foundation, Amsterdam, Netherlands).Input weight was derived from measurements with a Kern MTS standing scale (Kern and Sohn GmbH, Balingen-Frommern, Germany) for all participants, by subtracting pre-defined garment-dependent values from their fully clothed but shoeless weight.Height was determined through measuring tape.Participants were scheduled to undergo blood and urine analyses on the same day as the BIS measurements.
For measurements of Boxes A−D , electrode cables were first connected to the electrode connectors of the testboxes.The cables were then rolled out to their maximum length and fixed to a table with adhesive tape.For each testbox, measurements were repeated 20 times per BIS device in rapid succession.

Data retrieval and handling
Data from the BCM was stored on dedicated FMC smart cards, imported into the FMT program, and exported as a .csvfile.Data from Cella measurements were stored using the Cella smartphone application.Impedance data and modelled equivalent circuit data were wirelessly uploaded to a secure cloud server.Laboratory results were extracted from the hospital database.Anthropometric information and relevant comorbidities of all participants were collected at the start of the study visit and added to an Excel file (Microsoft, Redmond, USA).All individual datapoints were ultimately merged into a single data frame using R programming language version 4.3.0(21-04-2023) (The R Foundation for Statistical Computing, Vienna, Austria) and RStudio for macOS version 2023.06.2+561 (Posit Software, Boston, Massachusetts, USA).

Sample size calculation
We previously showed that the mean difference and standard deviation for R Inf and R i between two BCM devices of the same version number were 2.69 (±3.33)Ω and 40.41 (±63.24)Ω, respectively (Mussnig et al 2023).A sample size calculation to detect similar differences in R Inf and R i between BCM and Cella with a paired t-test (power = 80%, a = 0.95) yielded n = 15 and n = 22, respectively.To account for invalid measurements, the final sample size was set to n = 40.

Statistical analyses
Shapiro-Wilk tests were used to test for normal distribution.Population characteristics, laboratory and BIS data were described using absolute values (%), means (±standard deviations) or medians [quartile 1; quartile 3], depending on variable type and distribution.Coefficients of variation were calculated for repeated testbox measurements per BIS device.Impedance data were visualized in resistance-reactance diagrams and as locally weighted running line smoother (loess) curves by plotting impedance magnitude and phase against the respective AC frequency.Equivariant Passing-Bablok regression models were calculated according to Dufey et al (Dufey 2020) using the R package 'mcr' (Potapov et al 2023).Concordance correlation coefficients were calculated using the R package 'DescTools' (Stevenson 2024).Repeated measures analyses of variance (ANOVA) were performed to test differences between multiple groups, and pairwise t-tests were used post-hoc to test specific differences.Inter-device intra-subject differences were depicted using Bland-Altman plots and error bars.Generally, results were interpreted as statistically significant if the two-sided p-value was <0.05 except for pairwise t-tests, where multiple testing was corrected for following the Bonferroni-Holm method.All statistical analyses were carried out using R programming language version 4.

Measurements of resistance testboxes
Impedance measurements of 20 frequency sweeps per BIS device and testbox are shown in figure 2. Differences between measured values and true resistances were largest for Cella in Box D (R i difference 101.1 Ω (5.2%) and VO difference 0.3 L, table 1).Here, the device only computed resistances but no volume compartments or body composition parameters due to Cella prompting an error message whenever the measured capacitance is <0.7 nF (true capacitance of Box D is 0.68 nF).The coefficients of variation were less than 1% for all measurements in both devices (supplemental table S2).

Comparison of supine reference measurements
Results from Setups 1 and 2, the supine reference measurements with Cella and the BCM, are shown entirely in table 3. Equivariant Passing-Bablok regression models revealed no differences for R 0 , R Inf , ECW and ICW, a systematic error for VO and combined proportional and systematic errors for R i , LTM and FTM according to the respective means and 95% confidence intervals of the models' intercepts and slopes (figure 3, Panels A1-H1).Mean differences are visualized as Bland-Altman plots in (figure 3, Panels A2-H2).R i exhibited the largest mismatch with an average of 76.12 Ω (6.0%) (figures 3, Panel C2, figure 4(C)).Applying measured R 0 and R Inf to equations (4)-( 8) (appendix) showed smaller differences between Setups 1 and 2 (supplemental figure S3).
Agreement between different current paths and electrode materials WW measurements (Setups 3-10) produced larger resistances than their WA counterparts (Setups 1-2) (table 3).Pairwise t-tests for R Inf and R i were calculated for Setups 1-6 and yielded statistically significant differences for all comparisons except for R i between Setup 3 and 6 (p = 0.238) and R Inf between Setup 3 and 2 (p = 0.223).Loess curves of impedance magnitude and phase plotted against AC frequency revealed larger impedance magnitude of Setups 5-6, compressed phase peaks for Setups 3-6 and a hook-up effect of phase in the high-frequency AC range for Setup 6 (figure 5).Resistances of all setups were normalized to Setup 1 (figure 4 Panels A-C) and to Setup 3 (figure 4 Panels D-F).On average, Board measurements had the largest relative error while supine BCM data (Setup 2) showed the best agreement with supine Cella measurements (Setup 1).

Discussion
Here we show that two BIS devices of different manufacturers produce very similar impedance measurements but varying estimates of body composition both when measuring whole-body BIS in healthy volunteers as well as under controlled testbox conditions.
Objective I: Measurements of testboxes revealed high accuracy and precision of both machines.The relative error of BIS measurements compared to true resistor specifications exceeded the natural inaccuracy of resistors (1%) in some instances (Cella: R i of Box C [1.2%] and R 0 [2.1%],R Inf [3.6%] and R i [5.2%] of Box D ; BCM: R i of Box D [1.5%]) but overestimated VO only by a maximum of 0.3 L (table 1).
Resistances of fluid-equilibrated supine WA-BIS-Leads differed significantly between the Cella device (Setup 1) and the BCM (Setup 2), but the relative error was only 1% for R 0 and R Inf (figure 4 Panels A-B) and both parameters exhibited neither proportional nor systematic errors according to Passing-Bablok regression analyses (figure 3, Panels A1 and B1).Bland-Altman analyses did however suggest a linear relationship between magnitude and inter-device differences, with a negative slope for R 0 and a positive slope for R Inf (figure 3 Panels A2 and B2).Since R i is calculated from R 0 and R Inf according to equation (3), its larger absolute and relative difference, and proportional and systematic errors (figure 3 Panels C1-2) can be explained by error propagation.The high-frequency performance of both devices was good and warranted only little correction for time delay (T d , table 3).Differences in volume compartments and body composition were unsurprisingly large and the result of discrepancies in impedance measured, and models applied.These large offsets inhibit Cella and BCM from being used interchangeably.Especially the difference in FTM followed a peculiar linear pattern (figure 3 Panels G1-2).The BCM calculates tissue compartments according to the three-compartment model by Chamney et al (2007) all while utilizing BIS derived extra-and intracellular volumes relating to formulae by Moissl et al (2006).A patent assigned to Cella Medical Inc. (Baricevic 2022).suggests that while the Moissl model was initially used to set up calculations for Cella, an additional linear correction of ECW and ICW incorporating dual-energy x-ray absorptiometry measurements was implemented to account for body fat, thereby explaining the linear increase in difference of FTM between Cella and BCM (figure 3, Panel G2).FTM must not be confused with fat mass, since the former is considered normohydrated as part of the three-compartment tissue model while the latter refers to fat mass without water contents (Chamney et al 2007).It is worth noting that all volume and body composition compartments are calculated differently for each device, therefore either limiting the comparability or highlighting the fact that BIS devices are not just machines to measure resistances but are highly dependent on their proprietary models.In a supplementary analysis, we applied equations (4)-( 8) to impedance measurements with Setups 1-2 which revealed substantially less inter-device differences, no proportional and only systematic errors (supplemental figure S3).
The Cella device was previously compared to the BCM in 42 maintenance HD patients undergoing predialysis WA measurements (Matthie et al 2021).In this preprint the authors concluded that 'K the BCM and the Cella device can be used interchangeably', although VO exhibited a larger upper and lower limit in differences compared to our results (−1.192 to 1.081 L and −0.37 to 0.70 L, respectively).Further comparisons of both studies' results are limited because their analysis only included VO and ECW and ICW normalized to body weight, omitted information on resistances, absolute ECW, ICW and body composition and was conducted in HD patients.Our study offers a more thorough and conclusive comparison of these two BIS devices and arrives at a different conclusion: the devices must not be used interchangeably.We observed that WW-BIS with pre-gelled Ag/AgCl electrodes were on average equivalent to WA-BIS-Leads but the relative errors of R 0 , R Inf and R i to Setup 1 exhibited large variabilities between subjects (maximum −minimum = 27%, 41% and 88% for Setup 3; 27% 41% and 89% for Setup 4 (figure 4)).Impedance magnitude was equivalent to WA-BIS-Leads over the entire measured frequency spectrum (figure 5, Panel A) but phase showed a depressed peak (figure 5, Panel B).Electrical resistance is directly proportional to resistivity and conductor length, and indirectly proportional to the conductor's cross-sectional area.While the interelectrode distance is smaller for WW-BIS than for WA-BIS, the cross-sectional area of arms is proportionally even smaller than that of legs and the trunk.We may therefore expect larger resistances for WW than for WA measurements, although potential differences in resistivity per body segment is not factored in here.Cox-Reijven et al (2002) compared WA-to WW-and ankle-to-ankle-BIS using adhesive electrodes.The latter two produced similar results to regular WA-BIS with the relationship following a linear regression model.In their study, WW-BIS showed a bias towards less R 0 and R i than WA-BIS.While the same is true for our population in terms of R 0 , R i values were larger across all WW measurements (table 3 and figure 1).Keane et al (Keane and Lindley 2015) tested the BCM in supine WA and WW setups in HD patients.They also found a bias towards more ECW and ICW with WW-BIS, but the 95% confidence interval was balanced rather equally around a zero-litre-difference.
Objective II: On average, R 0 , R Inf and R i of WW-BIS-Plates were larger than WA-BIS-Leads (Setup 5: 15.4%, 19.3% and 28.2%; Setup 6: 15.9%, 15.3% and 15.3%).This offset was expected, again referring to the differences in current paths.However, compared to WW-BIS using Ag/AgCl electrodes, Board and Tablet measurements also produced a mean positive measurement error (figure 4, Panels D-F), suggesting that the error is not solely due to the current path but also to the electrode material.Bipolar electrode setups with polarized electrodes exhibit polarization impedance at the electrode-skin interface (Webster 2020).Since current injection and signal measurement are conducted with different electrode pairs in tetrapolar BIS setups, electrode polarization was initially thought to be negligible.This assumption turned out to be partially wrong (Grimnes and Martinsen 2007), with tetrapolar BIS also introducing a host of other potential errors (Mazzeo andFlewitt 2007, Marcôndes et al 2022).Impedance measurements also depend on electrode specifications.The contact impedance of pre-gelled Ag/AgCl electrodes varies between manufacturers (Nescolarde et al 2016) and is highly dependent on the electrode's surface area (Buendía et al 2012).'Dry' electrodes (for example gold-plated copper or stainless steel on printed circuit boards) may be more practical in day-to-day practice but were shown to measure larger impedance magnitude compared to pre-gelled Ag/AgCl electrodes (Kusche et al 2018).The positive relative offset of impedance measurements with Board and Tablet prototypes, which also showed AC frequency dependence (figure 5 Panel A), may therefore be the sum of parasitic effects and material differences.Koelmeyer et al (2020) compared WA-BIS-Leads to the SOZO WA-BIS device which is equipped with stainless steel electrode plates (ImpediMed Limited).The authors observed on average 15% larger R 0 values from the SOZO device, a finding which they attribute to the slightly larger interelectrode distance but not to differences in the electrode material.
Objective III: Implementing WW-BIS-Plates into day-to-day practice would consequently require adjustment for the current path and the contact impedance of gold-plated copper electrodes.Both corrections in sequence however seem unlikely to work with sufficient accuracy.Considering a male reference subject (R 0 = 575 Ω, R i = 1115 Ω, weight = 75 kg, height = 175 cm), even a 1% increase in R 0 and R i would amount to 0.12 L more ECW and 0.168 L more ICW.The large variability of the relative error of WW-BIS-Leads to WA-BIS-Leads between subjects and the unsystematic impedance offset of WW-BIS-Plates from WW-BIS-Leads cannot be corrected for by invoking simple empirically derived correction factors.The resulting estimates would exhibit excessive margins of error and would render these devices unusable in clinical practice.
This study is limited by the young age and low BMI of the healthy population.Additional anthropometric data such as length of limbs and circumference of the trunk (much like Cox-Reijven et al (2002)) could have provided useful information on the differences between WW and WA measurements and should be included in future investigations.More stringent pre-study measures (fasting, emptying of the bladder) could have produced more appropriate inter-subject comparisons.The intra-subject comparability however was unaffected by the lack of standardized measures.All measurements were conducted within roughly 30 min for each participant.No food or water intake, defecation or urination took place, and no serious shifts of volume compartments were expected during this period (except, of course, for the redistribution of volume during the ten minute long supine equilibrium manoeuvre).Electrode cables of the BCM were longer than those of Cella.Cable length is known to introduce substantial capacitive effects to BIS measurements.Measurements with the  B) derived from frequency sweeps in 40 healthy volunteers using BIS Setups 1-6 are depicted as loess curves.Each measurement covered 50 AC frequencies ranging between 4.88 kHz and 1.098 MHz (more measurements with higher frequencies, logarithmically scaled x-axis for better visibility).Note that the frequencies were similar but not identical between Cella and BCM (see supplemental table S1).AC, alternating current; BIS bioimpedance spectroscopy; BCM, Body Composition Monitor.
Tablet prototype (Setups 6, 8 and 10) were potentially limited because their larger T d values suggested stronger correction of stray capacitive effects in the high frequency range than Board measurements (table 3).This suspicion was further confirmed by the presence of a hook-up effect in the high-frequency range of phase measurements with Setup 6 (figure 5 Panel B) and because Setup 8 did not follow the same pattern of relative errors as other measurements with electrode prototypes.Although the BCM was validated against reference methods (Wabel et al 2006, Moissl et al 2007, 2008, Wabel et al 2009), we note that this is not a validation study against a reference method assessing the accuracy of either device.
In conclusion, we showed that two BIS devices of different manufacturers produced equivalent impedance measurements, but different fluid volume and body composition estimates in silico and in vivo.Due to clinically relevant differences in VO, LTM and FTM estimates, the Cella device should currently not be used interchangeably with the BCM in HD practice.The in vivo comparison was complemented by the introduction of gold-plated copper electrodes which produced larger resistance values both compared to WA-BIS and WW-BIS with Ag/AgCl electrodes.These results add to the growing literature detailing substantial differences between BIS machines originating from electrode materials, current paths and compartment models.R i and R 0 served to calculate ECW (in L) and ICW (in L) in accordance with formulae reproduced from Moissl et al (2006) in equations (4) and (5).Body mass index (BMI, in kg/m 2 ) was calculated by dividing body weight (W, in kg) by the square of body height (H, in m).VO (in L), LTM (in kg) and FTM (in kg) were calculated from ECW and ICW following equations by Chamney et al (2007), reproduced in equations ( 6)-( 8).

( ) =
´- ´ Note that the terminology we used is consistent with that of Chamney et al (2007), although VO is also called 'overhydration' in the BCM and 'hydration' in Cella.We refer to them more universally as VO because bioimpedance is capable for assessment of volume compartments rather than water alone (Spital 2007, Bhave andNeilson 2011).

Figure 2 .
Figure 2. Nyquist plots for measurements of Boxes A−D .Panels A−D depict data from Boxes A-D , respectively.Each panel's upper graph shows impedance data of 20 measurements per BIS device and resistance testbox with each individual frequency sweep covering 50 discrete alternating current frequencies between 4.88 kHz and 1.098 MHz as well as mean modelled results of Cm (in nF), T d (in ns) and a (in radian).Resistance and reactance were calculated from impedance magnitude and phase according to equations (1) and (2) (see appendix).Vertical lines in the panels below detail true R Inf (left end) and R 0 (right end) values of the respective testbox with coloured points presenting the mean results of R Inf and R 0 for the BCM (red, square) and Cella (blue, circular).BCM, Body Composition Monitor; BIS, bioimpedance spectroscopy; Cm, capacitance; R 0 , resistance at zero frequency; R Inf , resistance at infinite frequency; T d , time delay.

Figure 3 .
Figure 3. Equivariant Passing-Bablok regression and Bland-Altman plots of Setups 1 and 2. Panels A1-H1: Scatter plots of measurements with Setup 1 plotted against Setup 2 and corresponding equivariant Passing-Bablok regression models (Dufey 2020, Potapov et al 2023).The red solid line represents the Passing-Bablok fit, with the 95% confidence interval shaded grey.Panels A2-H2: Solid horizontal lines show the mean difference while dashed horizontal lines delimit mean difference ±1.96 standard deviations of the mean difference.Note that volume and body composition models differed between BCM and Cella.BCM, Body Composition Monitor; ECW, extracellular water; FTM, fat tissue mass; ICW, intracellular water; LTM, lean tissue mass; R 0 , extracellular resistance; R i , intracellular resistance; R Inf , resistance at infinite frequency; VO, volume overload.

Figure 4 .
Figure 4. Normalized errors of R 0 , R Inf and R i values in % compared to Setup 1 (WA-BIS-Leads, Panels A-C) and to Setup 3 (WW-BIS-Leads, Panels D-F).Values withing circles depict mean relative error in % and error bars delimit the mean error ±1.96 standard deviations of the mean error.R 0 , extracellular resistance; R i , intracellular resistance; R Inf , resistance at infinite frequency, WA-BIS-Leads, wrist-to-ankle measurements with adhesive electrodes; WW-BIS-Leads, wrist-to-wrist measurements with adhesive electrodes.

Figure 5 .
Figure5.Bode plots for impedance magnitude (Panel A) and phase (Panel B) derived from frequency sweeps in 40 healthy volunteers using BIS Setups 1-6 are depicted as loess curves.Each measurement covered 50 AC frequencies ranging between 4.88 kHz and 1.098 MHz (more measurements with higher frequencies, logarithmically scaled x-axis for better visibility).Note that the frequencies were similar but not identical between Cella and BCM (see supplemental table S1).AC, alternating current; BIS bioimpedance spectroscopy; BCM, Body Composition Monitor.

Table 2 .
Population characteristics.Values of n = 40 participants are presented as absolute numbers (%), mean (standard deviation) or median [quartile 1; quartile 3] depending on the distribution.BMI, body mass index.

Table 3 .
BIS results of all measurement setups.Values are presented as absolute numbers (% completeness i.e. number after exclusion divided by all measurements with the respective setup multiplied by 100), mean (standard deviation) or median [quartile 1; quartile 3] depending on the distribution.Note that negative values for T d should theoretically not exist since they do not meet the model critera.BIS, bioimpedance spectroscopy; Cm, capacitance; ECW, extracellular water; FM, fat mass; FTM, fat tissue mass; ICW, intracellular water; LTM, lean tissue mass; N, number; R 0 , resistance at zero frequency, R i , intracellular resistance, R Inf , resistance at infinite frequency; T d , time delay; VO, volume overload.