Development and validation of bioimpedance prediction equations for fat-free mass in unilateral male amputees

Participants

This study was approved by the Institutional Review Board (No. 1040875-201707-SB-030 and RERI-IRB-190924-1). After receiving a complete description of the study, all participants provided written informed consent. A total of 75 male, unilateral APTs were recruited and included in the study. The mean age of the participants was 43.6 ± 12 years. Seventeen participants had previously undergone upper limb amputation (trans-humeral amputation: n = 5; trans-radial amputation: n = 12), and 58 had previously undergone lower limb amputation (trans-femoral amputation: n = 32; trans-tibial amputation: n = 26). APTs with disarticulation and multilateral (bi-, tri-) amputation were excluded. Additional participant characteristics including residual limb length (cm) and onset (postoperative period, years) are presented in Table 2.

Experimental device (DXA)

DXA (Lunar Corp., Madison, WI, USA) measurements of FFM were used as the reference standard in the development of the ERE. This instrument was calibrated through the spine phantom provided by the manufacture daily. To standardize the scan, files from the original DXA system were transferred to iDXA software, version 4.0.2. The scan process was blinded and fulfilled by one radiologist who wore protective clothing. The segmentation method based on Heymsfield et al. (1990) was applied to uniform measurement (Heymsfield et al., 1990).

Participants were instructed to wear comfortable clothing for the assessment. Under the guidance of a professional examiner, each participant was asked to lay comfortably in the supine position on the assessment table, and spread both the upper and lower limbs. The whole-body DXA was performed for approximately 15 min.

Experimental device (BIA)

BIA of the tetrapolar 8-point electrode type (InBody S10 for the supine measure, InBody Co. South Korea) was used in this study. This BIA model uses eight electrodes positioned at each hand and foot and enables multifrequency impedance measurement of the arms, trunk, and legs. Impedance parameters were measured with alternating current of 80 and 100 mA at frequencies of 1, 5, 50, 250, 500, and 1,000 kHz for InBody S10. After checking for precision errors of FFM about repeatability through biplicate measurements of the same APT based on a previous study (Buckinx et al., 2015), the InBody S10 was used. It was designed for single measurements in the supine position on a non-conductive surface through a SFBIA of only 50 kHz. In the sound limb, defined anatomical sites were cleaned with alcohol, on which adhesive gel electrodes were placed on the dorsal surfaces of the hand, wrist, ankle, and foot as follows: the proximal edge of the wrist electrode was attached from an imaginary line bisecting the styloid process of the ulna and the proximal edge of the finger electrode on an imaginary line from the imaginary line bisecting the metatarsophalangeal joint of the middle finger. The proximal edge of the ankle electrode was attached from an imaginary line bisecting the medial malleolus and the distal edge of the toe electrode was placed from an imaginary line through the metatarsophalangeal joints of the second toe as shown in Fig. 1B. In the residual limb, the distal and proximal electrodes were attached from the end of the stump (=distal part) to the region (=proximal part) by keeping the distance according to the instructions of InBody S10. Additionally, fixed-distance of electrodes was used with a 5-cm standard distance as shown in Fig. 1C (Kaysen et al., 2005; Kriemler et al., 2009). The device was calibrated every morning using the standard control circuit supplied by the manufacturer. We confirmed that the precision error was less than 2% (Fig. 1).

Definition of segmental ZI (ZI) and regions of interest (ROI) in DXA

Impedance indices (ZI) for each body part were determined considering the residual limb length in each participant. ZI values were calculated in accordance with methods described by Tanaka et al. (2007). As shown in Eq. (1), ZI was calculated by dividing height² by Z (based on values for non-APTs). We then calculated the body part length ZI (ZI_BPL) by dividing the body part length (BPL)² by Z. (Tanaka et al., 2007). BPL was measured for the following five areas: left arm (LA), right arm (RA), left leg (LL), right leg (RL), and trunk (TR). The reference positions for the lengths of the upper/lower limbs and trunk were defined as follows: the area between the humeral head (=acromion) and styloid processes of the wrist (ulna) for the upper limbs; the area from the anterior superior iliac spine to the medial malleoli for the lower limbs; and the posterior length between the seventh cervical vertebra prominens to the center of the posterior superior iliac spine for the trunk (Beattie et al., 1990; Hackenberg et al., 2003; Jamaluddin et al., 2011; Lee, Cha & Lee, 2016; Neelly, Wallmann & Backus, 2013; Ross, 1972). The results of the experiment were analyzed by matching the physical measurements to the BIA electrode locations and DXA ROI based on these measurement standards.

According to the Heymsfield’s protocol (1990), the boundaries of the ROI are defined as follows: (1) for the upper limbs of the ROI (right and left), the arms are isolated by running a line through the humeral head and (2) for the lower limbs, the pelvis cut is placed just above the pelvic brim and the computer automatically draws the lower pelvic lines to bisect the hip joints (Heymsfield et al., 1990; Jeon et al., 2020). (1)${\displaystyle ZI=\frac{H^2}{Z}}$ H: Height of the whole body, Z: Impedance (2)${\displaystyle Z{I}_{BPL}=\frac{BP{L}^2}{Z}}$ BPL: Body part length, Z: Impedance

Independent variables for segmental BIA

Independent variables included in the EREs were determined for the five body areas as follows: R was applied by differentiating among the LA (R_LA), RA (R_RA), LL (R_LL), RL (R_RL), and trunk (R_TR). Using notations identical to those for R, impedance (Z), reactance (Xc), and PA were also calculated for each body part and expressed in terms of R_BPL, Z_BPL, Xc_BPL, and PA_BPL.

Experimental procedures

Prior to the measurements of FFM, participants were instructed to abstain from excessive dehydration-accompanied exercises and excessive alcohol use. In addition, they were instructed to fast for at least 6 h and abstain from alcohol consumption for at least 4 h.

We ensured that the urinary bladder was voided in all participants within 30 min before measurement, all participants wore non-conductive and comfortable sportswear. All conductive materials, prosthetic limbs, and amputation covers (silicone, amputation protection, etc.) were removed. First, DXA and then BIA test was conducted. For the measurements, after lying on a DXA sheet, stability (stabilization of BC) measurement was taken for 5 min in the supine position considering the potential for variable fluid shifts, and the DXA scan was then performed, and BIA was measured on the place immediately after the DXA scan without changing the position. The shoulder and hip joints of all participants were maintained at an abduction angle of approximately 15° for the shoulder and hip joints. The elbow and knee joints were extended in a straight anatomical position. The physical contact of each electrode was ensured in accordance with the criteria recommended by the manufacturer. Each measurement took approximately 5–15 min. Participants were instructed to maintain a comfortable position without any movement during the examination (Brantlov et al., 2017; Kyle et al., 2004d).

Sum of the segmentally-predicted FFM values by sERE (Sum_sEREs)

After the development of the sEREs of the five body parts in the APTs, the sum of the segmentally-predicted FFM values by sERE was calculated. We confirmed the validity and accuracy between DXA and the sum of sERE_BIA about FFM through bivariate linear regression and the Bland_Altman plot.

Statistical analysis

The physical characteristics of the APTs group are presented as means with SDs. A normality test is required if there are fewer than 30 participants. However, since the number of participants in our study was 75, we assumed normality and analyzed all data (Kwak & Kim, 2017; Sang Gyu et al., 2019). The forward stepwise multiple linear regression analysis was used to develop sEREs in the APTs group. The significance level was set to p < 0.05. Variables included in the initial analyses contained ZI_BPL, R_BPL, Xc_BPL, PA_BPL, age (yr), height (cm), and weight (kg). The developmental equations were selected by measures of goodness-of-fit statistics, including coefficient of determination (R²), the standard error of estimate (SEE), acceptable subjective rating of SEE (i.e., good to excellent) according to the minimally acceptable standard for prediction errors (Buckinx et al., 2018; Heyward & Wagner, 2004a; Heyward & Wagner, 2004b; Lohman, 1992), and the variance inflation factor (VIF). The SEE measures the variation in the actual values from the predicted values. The SEE represents the degree of deviation of individual scores form the regression line. It is computed using the following formula:

SEE = $\sqrt{\sum {\left(Measured\mathrm{FFM}-Estimated\mathrm{FFM}\right)}^2∕\left(N-p-1\right)}$

where p = number of predicter variables. The VIF assesses how much the variance of an estimated regression coefficient increases when predictors are correlated for estimating collinearity/multicollinearity. In case of values more than 10, it can be assumed that the regression coefficients are poorly estimated due to multi-collinearity to remove predictors from the model. In our study, with values less than 10 (Lee et al., 2018; Wickramasinghe et al., 2008), we could proceed with our regression analysis. In the cross-validation, the group predictive accuracy of the Sum_sEREs was tested by calculating R², total error (TE: The TE represents the degree of deviation from the line of identity using the formula:

Total Error = $\sqrt{\sum {\left(Measured\mathrm{FFM}-Estimated\mathrm{FFM}\right)}^2∕N}$),

and acceptable subjective rating of TE (Heyward & Wagner, 2004a; Heyward & Wagner, 2004b; Lohman, 1992). The individual predictive accuracy of these equations was also tested by Bland-Altman plots, whitch included the bias of the mean difference between measured values of DXA and predicted values of Sum_sEREs. We used the 95% limits of agreement (LOA) between equations, and concordance correlation efficient (r_{y−y′,mean}). Data were analyzed using Microsoft Office Excel Ver. 2013 (Microsoft, Redmond, WA, USA) and SPSS version 18.0 (IBM, USA).

Segmental Estimated Regression Equation

DXA measurements of FFM were used as the dependent variable in the development of the EREs for use in APTs. Various independent variables were entered to ensure optimal model development. Our model considered factors such as R², multicollinearity (tolerance and variance inflation factor [VIF]), and standard error estimates (SEE). Using these factors, we developed sEREs for the left and right upper/lower limbs as well as the trunk. ZI_LA, Xc_LA, height, and age were entered as independent variables in the final sERE model for FFM in the left arm (LA_FFM). Values for the final sERE for LA_FFM were as follows: R = 0.95, R² = 0.90, and adjusted R² = 0.89. ZI_RA and Xc_RA were entered as independent variables in the final sERE model for FFM in the right arm.

Values for the final sERE for RA _FFM were as follows: R = 0.86, R² = 0.74, and adjusted R² = 0.73. R_BPL, Xc_BPL, ZI_BPL, and weight were entered as independent variables in the final models for both the left and right lower limbs. The LL_FFM model included R_LL, Xc_LL, ZI_LL, and weight. The final sERE values for LL_FFM were as follows: R = 0.95, R² = 0.91, and adjusted R² = 0.90. The highest correlation coefficients were observed for RL_FFM: R = 0.97, R² = 0.94, and adjusted R² = 0.93. In contrast, the lowest correlation coefficients were observed for TR_FFM: R = 0.88, R² = 0.78, and adjusted R² = 0.76. A total of five independent variables were entered for the TR_FFM ERE: ZI_TR, weight, height, age, and R_TR (Table 3).

Cross-validation between BIA and DXA about FFM

Residuals and bias in Bland–Altman plots

To evaluate the validity of the final ERE, the residuals (i.e., the difference between BIA-estimated and DXA-measured FFM) and means of the two methods were assessed using the Bland–Altman plot (Fig. 2). The bias of the difference between the two FFM measurements was –4.60 kg. BIA estimates obtained using the ERE tended to be higher than those obtained using DXA values. Furthermore, there was a tendency for the residuals of the bias to be evenly distributed, and increased bias tended to be associated with decreased residuals.

Limitation

In this study, limitations were placed on the use of only 50kHz frequency to characterize the cell membrane; however, ICF and ECF have limitations in reflecting the characteristics at a normal level. We intend to proceed with the MFBIA study including 1, 50, 250, 500, 1000kHz, and over frequencies, as a future research project. In the validation procedure, we recognized the importance of validating a method through a split group design, K-fold or LOOV-type. However, the sEREs were developed without external cross-validation through group of predictive power test. We confirmed only cross-validation between DXA and Sum_sEREs of FFM values. Additionally, we also did not perform a thoughtful analysis of the PA, but we intend to do it for the APTs in the future.