Introduction

Body composition is an important element to consider when assessing individuals of a wide range of ages, given that age-specific patterns of changes occur naturally throughout the life span [1,2,3]. Skeletal muscle mass (SMM), for example, is a major predictor of physical function and survival rate later in life [4], and decreases throughout adulthood [5, 6], which leads to natural changes in whole-body lean soft tissue mass (LSTM) and appendicular LSTM (ALSTM). Thus, following body composition during adulthood will allow not only to target vulnerable groups exposed to SMM decline, but also to tailor effective interventions delaying this age-related deterioration.

Dual-energy X-ray absorptiometry (DXA) is a widely accepted method with high validity and reliability for the estimation of LSTM [7, 8] and has been shown to provide an adequate estimation of SMM [7]. Given the constraints regarding the use of this methodology (i.e., high cost and low portability), bioelectrical impedance analysis (BIA) has emerged as a more feasible alternative for body composition assessment [9,10,11]. The indirect estimation of body composition (i.e., fat-free mass (FFM) and LSTM) through BIA derives from measures of body tissue electrical conductivity and depends on the addition of biological and physiological variables to the prediction model [12]. Still, the accuracy of such prediction equations are device-specific and rely on the characteristics of the population in which they were developed [12].

The availability of tetrapolar BIA analyzers enables the estimation of regional body composition, which differs from the traditional whole-body technique [11]. To date, most measures of regional BIA are performed in a standing position on a platform embedded with stainless tetrapolar electrodes. Even though this is a convenient and applicable approach, several predicting errors resulting from inappropriate contact between the skin and the stainless electrodes have been reported [13]. As an alternative, the use of a tetrapolar BIA approach with eight pre-gelled Ag/AgCl surface electrodes, while performed in a supine position, has been proposed to overcome some of these issues and compensates for differences in distinct body types. Currently, there is a research gap regarding the validation of regional BIA measurements for predicting DXA regional LSTM, with most of the equations being developed in specific populations, such as athletes [14, 15] and older adults [16,17,18]. Even though some recent equations predicting the overall LSTM of arms, legs and trunk have been developed in adults [19, 20], to the best of our knowledge, no previous investigations using tetrapolar BIA measurements with eight point electrodes have validated equations predicting DXA-derived LSTM of each body segment independently using healthy adults as the reference population. The present investigation aimed to develop and validate specific BIA-derived equations predicting regional DXA LSTM using a sample of adults varying in age, sex and body composition.

Materials/methods and subjects

Sample

This cross-sectional investigation included 149 adults (i.e., 77 women) with distinct body composition profiles (i.e., normal weight, overweight, and obese profiles). Participants with active pregnancy, implantable medical devices, amputated limbs, orthopedic prosthesis, skin wounds at the electrode placement sites, and under the administration of medication or clinical conditions with impact on water compartments or body cell mass were excluded. All data were collected between July and December 2020 at the Exercise and Health Laboratory, Faculty of Human Kinetics, University of Lisbon.

Anthropometry

Body weight and height were measured with participants wearing minimal clothing on an electronic scale with an integrated stadiometer (SECA 796 Hamburg, Germany) according to standardized procedures [21]. Body mass index was calculated as body weight divided by height squared.

Bioelectrical impedance analysis (BIA)

The impedance measurements were performed in the early morning following 12 h of fasting and after bladder voiding using a phase-sensitive single-frequency BIA (BIA 101 BIVA PRO, Akern S.R.L., Pisa, Italy), which applies an alternating current of 245 microamperes at 50 kHz. Prior to participation, participants were asked to maintain regular dietary habits, to refrain from exercising in the 24 h before to the assessment day, to not smoke in the 8 h before morning assessments, and to remove all metal accessories (e.g., ring, earrings, neckless, watches) before the assessment moment. After five minutes of rest to stabilize body fluids, the impedance measurements were performed with the subjects in the supine position with a leg opening of 45° compared to the median line of the body and the upper limbs positioned 30° away from the trunk.

Four injecting current electrodes were placed on the dorsal surface of both hands and feet, in the plane of the head of the third metacarpal and third metatarsal, respectively, while the remaining four electrodes (i.e., sensing electrodes) were placed on the dorsal surface of both wrists and tibia-tarsal joints, in the middle of an imaginary plane between the two styloid apophyses of each hand and in the middle of an imaginary plane between the two malleoli of each, respectively [14].

The resistance index (RI) of each body region was calculated as the full height (cm) squared divided by each regional R value (height2/R) and used to represent the relative contribution of each body region to whole-body conductivity. In addition, the relative contribution of trunk and extremities to whole-body conductivity was calculated from the ratio between trunk and extremities RI [22]. The coefficient of variation in our laboratory for repeated within-day R and Xc measures was, respectively, 1.6% and 1.9% for the right arm, 1.9% and 1.7% for the left arm, 1.9% and 1.9% for the right leg, 1.1% and 2.0% for the left leg, 1.3% and 0.1% for the right trunk, and 3.4% and 0.7% for the left trunk.

Dual-energy X-ray absorptiometry (DXA)

Total and regional FFM were estimated on a DXA fan-beam densitometer (Hologic Explorer-W, fan-beam densitometer, software QDR for Windows version 12.4, Waltham, MA, USA) using a standardized protocol, with the same laboratory technician performing daily calibrations, consisting of scanning a step phantom with six fields of acrylic and aluminum of varying thickness and known absorptive properties to serve as an external standard for the analysis of different tissue components. In addition, the technician positioned the participants, performed the whole-body scans, and analyzed the data. Regional FFM was measured through partial analyses with Hologic APEX Version 3.3.0.1 analysis software, based on regions of interest defined by default. Regional LSTM was calculated by subtracting bone mineral content from FFM of each segment. The coefficient of variation of measurement in our laboratory for LSTM and ALSTM were 1.1% and 1.8%, respectively [23].

Statistical analysis

All analyses were performed using IBM SPSS Statistics, version 25 (SPSS Inc., Chicago, IL). Descriptive characteristics were presented as means, standard deviations and ranges (minimum-maximum). Normality was assessed using the Kolmogorov–Smirnov test (n ≥ 50) for the development sample and the Shapiro–Wilk test (n < 50) for the validation sample. Differences between both samples were assessed using the Mann–Whitney U test (for continuous variables) and Fisher’s Exact test (non-continuous variables). Statistical significance was set at ɑ = 0.05.

The prediction equations were established using a cross-validation method in which 100 participants (i.e., 2/3 of the sample) were selected in a random fashion using the basic random number generator in Excel (participants with the lowest 100 numbers were allocated to the development group), and used for equations development, while the remaining 49 (i.e., 1/3 of the sample) were used for equations validation. A sample size of 100 participants provided sufficient power to perform the development models to achieve a moderate effect size for the R2, while considering the inclusion of four independent predictors (i.e., ≥2 participants per predictor rule) [24], a type-1 error of 5% and a power of 80%.

Equation development

A multiple linear regression approach with stepwise and backward selection procedures was used to develop equations for predicting DXA-derived LSTM of specific body regions using the development group. Predictor variables tested included sex, age, ethnicity, side dominance, standing height, sitting height, and raw BIA-derived measures (Xc, R, RI). Considering that no sex interactions were observed with the BIA parameters, this parameter was used as a predictor in the final models. The resulting equations were tested for multicollinearity using the variance inflation factor and, if multicollinearity was detected, the predictors with the lowest correlation with DXA-derived LSTM were eliminated. The model selection was based on the model coefficient of determination adjusted for the number of predictor variables in the model (i.e., adjusted R2) and the standard error of estimation (SEE).

Equation cross-validation

The predicting model with the highest adjusted R2 and the lowest SEE was selected for cross-validation analysis, and further compared with DXA-derived LSTM to assess the equality of the corresponding regression slopes and intercepts. To estimate the closeness of fit between BIA-predicted and DXA-measured LSTM of each segment, the 95% confidence intervals of each approach were assessed by testing the slope and intercept against the null values of 1 and 0, respectively.

The developed equations were applied to the validation group to predict the DXA-determined LSTM values. The paired sample t-test was used to compare mean values of BIA-predicted and DXA-measured LSTM in the validation group. The pure error (PE), which is the square root of the mean of squares of differences between the measured and predicted LSTM [25], was calculated to test the performance of the predictive equations. The Bland and Altman analysis [26] determined the limits of agreement (LOA) (mean difference ±2 SD) between BIA-derived and DXA-determined LSTM of all body regions. The statistical significance was set on p value <0.05.

Results

There were no differences in the characteristics of the development and validation groups. All descriptive data is presented by group in Table 1.

Table 1 Characteristics of development and validation groups.
Full size table

Table 2 shows the developed prediction equations for estimating regional LSTM. In the regression analysis, sex, left body RI (RILB) and RI resulting from the ratio between trunk to extremities (RITE) explained 90% (Adj.R2 = 0.900, SEE = 1.823) of the variance in DXA-measured left body LSTM, while right body RI (RIRB) and right trunk RI (RIRT) explained 89% (Adj.R2 = 0.894, SEE = 1.902) of the variance in DXA-measured right body LSTM. For lower body LSTM prediction, sex, lower body RI (RILwB) and RITE explained 89% (Adj.R2 = 0.894, SEE = 1.413) of the variance in DXA measured lower body LSTM, while sex, upper body RI (RIUB) and RITE explained 92% (Adj.R2 = 0.918, SEE = 0.550) of the variance in DXA measured upper body LSTM. For both right and left legs and arms, variables including sex, RI specific for each segment and RITE explained 89% (Adj.R2 = 0.893, SEE = 0.724), 88% (Adj.R2 = 0.876, SEE = 0.751), 90% (Adj.R2 = 0.902, SEE = 0.302) and 91% (Adj.R2 = 0.911, SEE = 0.285) of the variance in LSTM determined by DXA, respectively. The RITE and mean trunk RI (RIT), calculated as the mean of the right and left trunk RI, explained 78% (Adj.R2 = 0.783, SEE = 2.623) of the variance in trunk LSTM measured through DXA.

Table 2 Developed prediction equations using development group (N = 100) for estimating DXA-derived LSTM from BIA and performance analysis of developed prediction equations using the cross-validation group (N = 49).
Full size table

Figure 1 displays the relationship between predicted LSTM and DXA measured LSTM for each body segment. No differences were found between predicted and DXA-derived LSTM for all body regions, with exception to trunk LSTM (p < 0.05). In addition, strong relationships (p < 0.001) were observed between predicted and measured LSTM for upper and lower body (R = 0.96; R = 0.89), right and left sides of the body (R = 0.94; R = 0.94), right and left arms (R = 0.94; R = 0.96), right and left legs (R = 0.88; R = 0.88), and trunk (R = 0.90). In addition, slopes and intercepts of predicted LSTM and DXA-derived LSTM did not differ from 1 and 0, respectively, for each segment (p ≥ 0.05), with the exception for the trunk model (p < 0.05).

Fig. 1: Relationship between the measured and predicted LSTM.
figure 1

For all body regions are indicated the with line of identity (solid line) and the regression line of predicted and observed LSTM (dashed line). Adj.R2 adjusted coefficient of determination, BIA bioelectrical impedance, DXA dual X-ray absorptiometry, LSTM lean soft tissue mass, SEE standard error of estimation.

Full size image

Bland-Altman analyses (Fig. 2) revealed no associations between the differences and the means of the predicted and DXA-derived LSTM of the upper and lower body (r = 0.05, p = 0.72; r = −0.24, p = 0.10), right and left sides of the body (r = −0.01, p = 0.93; r = −0.04, p = 0.81), right and left arms (r = −0.19, p = 0.20; r = −0.06, p = 0.68), right and left legs (r = −0.26, p = 0.08; r = −0.26, p = 0.07) and trunk (r = 0.17, p = 0.26).

Fig. 2: Bland-Altman plots of the difference between observed and predicted LSTM.
figure 2

For all body regions are indicated the mean of the observed and predicted LSTM (long dashed line) and 95% LOA (short dashed line). LOA limits of agreement, LSTM lean soft tissue mass.

Full size image

Discussion

The present investigation led to the development and validation of BIA equations for predicting DXA LSTM of distinct body regions in a population of healthy adults. To date, no such equations are available for predicting regional LSTM derived from raw BIA parameters assessed using a tetrapolar BIA device with eight electrodes. This investigation offers multiple nonproprietary equations that accurately estimate LSTM of each body region independently and can be used in the future as a cost-effective and time-efficient alternative to DXA for the assessment of regional LSTM in adults with similar characteristics to the development sample.

The LSTM has been highlighted as an independent predictor of mobility and mortality across all ages [27]. Based on the construct that the largest portion of whole-body LSTM is primarily SMM and that approximately three-fourths of whole-body SMM is located in the limbs [28], a greater emphasis has been given to the assessment of ALSTM. Due to the influence of ALSTM on physical performance [29], a growing number of BIA prediction equations for upper and lower limbs LSTM have been developed in special populations (i.e., athletes and older adults), using whole-body [14] and raw BIA parameters [17] (see Table 3). Previous studies showed that BIA-derived equations to predict ALSTM in athletes accounted for 84 to 89% of the variability in upper limbs LSTM and for 81 to 93% of the variability in lower limbs LSTM as measured by DXA [14]. In addition, a similar variance was found in older adults with BIA prediction equations accounting for 83–90% of the variability in upper limbs LSTM and for 83 to 85% of the variability in lower limbs LSTM assessed through DXA [20]. Even though the BIA prediction equations emerging from the present investigation accounted for a similar variance in upper limbs LSTM as those in athletes and older adults (88 to 91%), the percent of variability in lower limbs LSTM was slightly lower (78%). Since the validation sample included adults from a wide spectrum of ages and body composition profiles, the estimation of body composition through BIA-derived parameters may have contributed to inconsistent relations between lower limbs LSTM and bioelectrical resistance in some individuals. Although there were no statistical differences between observed and predicted LSTM in the lower limbs, the BIA models for lower limbs LSTM exhibited wide LOA and high PE, similar to findings reported in athletes [14]. Although the use of eight electrodes is expected to perform better compared to using a traditional tetrapolar BIA approach [30], the wide LOA reported in the present investigation may be mostly explained by the high levels of individual variability in regional LSTM within our sample [30]. For instance, our findings revealed that in almost all body regions, with the exception of the lower body, the agreement between DXA and BIA predicted LSTM fell outside of the 95% limits for some cases. Therefore, even though caution should be taken when using these equations to estimate regional LSTM at the individual level (due to high individual variability), these equations have shown to provide accurate estimations when used at the group level.

Table 3 Summary of regional tetrapolar BIA (8-electrodes) findings.
Full size table

Since the main component of whole-body and appendicular LSTM is SMM, an important health and nutritional indicator particularly during older adulthood [29], most of BIA-derived equations predicting ALSTM have been developed for the older adult population using DXA as the reference method [17, 18]. By contrast, only few investigations using multifrequency or tetrapolar BIA measurements with eight-point electrodes have developed and validated BIA-derived equations predicting ALSTM assessed by DXA using a sample of healthy adults [19, 20]. Even so, no studies have developed equations predicting DXA-derived ALSTM of each body segment independently in adults. Thus, by developing and validating BIA-derived equations predicting LSTM for each limb, as well as LSTM of other underexplored body regions (e.g., upper and lower body), our investigation makes a significant contribution to research in the field of body composition.

It is important to highlight that among all predictor variables considered, regional RI was the only bioelectrical parameter to fit in all equations, suggesting a predominant role of this parameter in estimating LSTM [17]. This finding reinforces the previous evidence highlighting RI as the main contributor of LSTM (with a variance accounting for approximately 80-90%) [17, 31]. Even though additional raw BIA variables, such as R and Xc, have been considered when estimating regional LSTM [17, 32], as presented in Table 3, these variables did not significantly contribute to our models and were not included. This is somewhat surprising, considering that Xc, for example, is mainly related to the capacitance function of the cell membranes, including skeletal muscle cells, which are the main component of LSTM [12]. Nevertheless, our findings concur with previous available evidence suggesting that Xc does not contribute [14] or contributes very little [17] to the prediction of regional LSTM in adults, regardless of their BMI, possibly due to a greater stabilization of Xc during this period of life [33]. Therefore, this parameter should rather be considered and included as a predictor of regional LSTM in children and older adult populations [12].

Despite the validity of our newly developed equations, there are some limitations that are worth noting. First, the use of DXA as the reference method is expected to overestimate specific parameters (i.e., SMM), when compared to other reference methods (i.e., computed tomography) [34]. This difference relies on the fact that LSTM is a heterogeneous body component at the molecular level of analysis, composed by water, protein, soft-minerals, and glycogen [35], and was used as a surrogate for assessing SMM, a tissue-level component. As a consequence, other tissue-level components such as skin, blood vessels, connective tissue and fat-free adipose tissue may have not been discriminated from SMM when using LSTM as a predictor [36]. Nevertheless, DXA measured LSTM in the limbs has been found to be significantly interrelated to SMM and is considered an accurate method to assess both appendicular and whole-body composition [37]. In addition, the development of equations that include the trunk segment may be a limitation, considering that significant differences were found between the observed and predicted values of LSTM in this segment. Even though the trunk represents approximately 50% of the body mass, it only has a slight contribution (5–12%) to whole-body resistance. This phenomenon can be attributed to the trunk geometry (i.e., short length and large diameter) and fluid compartmentalization [38], which have been shown to adversely compromise the estimation of trunk impedance components using tetrapolar BIA devices [39]. Although a strength of our study was the development and validation of our equations in a diverse, heterogenous group of adults, which allows our equations to be applied to participants from a wide spectrum of ages and body composition profiles, our equations may not be applicable to populations with different characteristics (e.g., children, older adults, people with chronic diseases, and other ethnicities) and when other BIA devices and approaches (frequencies different from 50 Hz) are considered. Finally, the estimation of regional body composition using whole-body raw BIA parameters may contribute to inconsistent relations between DXA regional LSTM and the BIA-derived parameters. Thus, to maximize BIA estimations, a major strength of our investigation was the use of a tetrapolar BIA device, which allowed for the assessment of regional raw BIA parameters. However, given that some tetrapolar BIA devices do not discriminate the boundaries of the body regions of interest, it remains unclear whether region of interest used by BIA devices can be comparable to the DXA regions defined by default according to the manufacturer.

In summary, BIA-derived equations predicting LSTM of distinct body regions were developed and validated using an independent sample of healthy adults. These equations have the potential to be applied to monitor changes in regional LSTM that naturally occur throughout adulthood, and to follow the changing patterns resulting from health-related programs. Further research is needed to examine the validity of these equations in different multi-ethnic populations.