The need for the impedance measures with a skeletal tissue-specific high-frequency that can be directly comparable to ALM measured by DXA in epidemiological and clinical settings has led us to develop and cross-validate accurate MF-BIA prediction equations for the ALM in elderly men and women. The MF-BIA prediction equation developed in this study included the impedance index (ZI) for the skeletal muscle-specific frequency at 2 MHz, reactance (Xc) at 5 kHz, and sex as variables. The accuracy and precision of the newly developed prediction equation for MF-BIA were high, not only at the group level (R2 = 0.931, SEE = 0.97 kg ALM) but also at the individual level (Bias = -0.01 ± 0.97, LoA = -1.92 ~ 1.90 kg ALM, PIA = 82.1%). The equation’s sensitivity, specificity, and overall agreement in the diagnosis of sarcopenia make its use suitable for clinical settings and epidemiological studies. Thus, it is demonstrated that the newly established appendicular lean-specific 2 MHz high-frequency BIA prediction equation can be applied to clinical settings and epidemiological studies with high accuracy.
The main achievement of the present study was the development of a new and accurate prediction equation that validates its within-group and individual error based on the 2 MHz high-frequency impedance index. Said index expresses the characteristics of appendicular lean mass, without anthropometric variables in the equation. Previous BIA prediction equations for appendicular lean mass included age, weight, and waist circumference as explanatory variables [14,15,16,17,18,19,20, 22, 25, 28], in addition to ZI, RI, and Xc as original variables of bioimpedance parameters to reduce the error from the assumption that the human body has a cylindrical conductive volume [29]. However, the use of multiple empirical variables (or anthropometric variables) in the regression, causes problems of multicollinearity linked to the unreliability of the estimated regression coefficient because of the correlation between explanatory variables and increased variance of the regression coefficient. This results in the derivation of an inaccurate regression equation [26, 30, 31].
This study excluded the explanatory variable of sex included in the prediction equation and added reactance based on the 2 kHz high-frequency impedance index, thereby ensuring that the VIF (ZI = 3.54, Xc = 1.07, Sex = 3.57) of each explanatory variable does not have multicollinearity. Therefore, each explanatory variable had a valid prediction equation with independent explanations for the ALM. In particular, the explanatory variability of the 2 MHz high-frequency impedance index had among the three independent variables an of R2 = 0.915 (R = 0.957) with SEE = 1.0822 kg AMS.
The explanatory variability and SEE of ZI or RI in the 50 kHz BIA prediction equations of previous studies were each: below 0.860 and above 1.34 kg ALM [28], 0.836 and above 1.450 kg ASM [15], below 0.852 and above 1.283 kg ALM [17], 0.883 and 1.401 kg ALM [16], 0.856 and 1.26 kg ALM [18]; and 0.917 and 1.53 kg ALM [19]. The explanatory variability and error of recently reported 250 kHz BIA prediction equations were each: 0.906 and 1.490 kg ALM [26], 0.415 ~ 0.635 and 2.27 ~ 1.88 kg ALM [14], and 0.825 and 1.35 kg ALM [20]. The addition of Xc of 5 kHz and sex along with ZI as predictor variables improved the explanatory variability and SEE of the prediction equation. Thus, the explanatory variance of R2 = 0.930 (R = 0.965) and group predictive accuracy of SEE = 0.97 in the new high-frequency prediction equation was superior to that of most previously published predictive equations (R2 = 75.7 ~ 92.5%, SEE = 1.02 kg ~ 1.46 kg) [15, 18, 19, 22, 26]. Its values were further comparable to results from the equations [16, 20] developed through large-scale studies by Peniche and Kyle which had predictive powers of R2 = 0.910 R2 = 0.952 and SEE = 1.01 kg and SEE = 1.12 kg ALM, respectively. Therefore, the appendicular lean-specific 2 MHz high-frequency impedance index used in this study showed the highest explanatory variability and precision for the single impedance index among all reported studies.
The reason for this is that the 2 MHz ZI has high explanatory variability specific to appendicular lean mass and induces conductivity ideal for the intracellular fluid, which composes up to 80% of skeletal muscles [11, 32]. To date, there have been reports of prediction equations for fat-free mass and appendicular lean mass using impedance index at 50 kHz, which is shown to have the highest reactance on the Cole–Cole model [13]. However, Deurenberg & Schouten [1992] reported a frequency range up to 1350 kHz that did not have high reactance, but high conductivity in both the extracellular and intracellular fluid through penetration of the cellular membrane [12]. Their study demonstrated that conductivity in both the extracellular and intracellular fluids under high frequency results in measurements of extracellular and intracellular fluids with higher precision and accuracy. However, the lack of advancements in technology and screening required for multifrequency measurements has led the 50 kHz index to be used most often hitherto. Said index also allows for simple measurements and conductivity of not only the extracellular but also the intracellular fluid [13, 33].
On the other hand, recent innovations in technology allowed high-multifrequency range measurements at 5 kHz, 50 kHz, 250 kHz, 500 kHz, 1 MHz, 2 MHz, and 3 MHz to be made easily and safely at lower frequencies, resulting in a higher accuracy of the measurements of the intracellular fluid. This would have enabled this study to precisely measure the conductivity of the intracellular fluid of skeletal muscle, which constitutes the highest proportion of total body water. Therefore, it may be predicted that high frequency ranges could be used in the conversion of ZI using the conductive volume model for appendicular skeletal muscles to ALM to obtain precise and accurate prediction equations in the future.
Simultaneously, this study had no bias in its individual predictive accuracy, and the percentage of individual agreement (PIA) within the margin of error was above 81%. In particular, this study had improved LoA of -1.90 ~ + 2.08 kg ALM in comparison to LoAs of previous studies such as: -2.68 ~ 2.41 kg ALM [28], -2.82 ~ 2.81 kg ALM [15], -2.467 ~ 2.562 kg ALM [17], -2.22 ~ 2.22 kg ALM [16], -2.23 ~ 2.46 kg ALM [18], -2.2 ~ 2.1 kg ALM [19], -1.95 ~ 1.98 kg ALM [26], and -2.2 ~ 1.9 kg ALM [20]. This improved LoA make the novel equation suitable for predicting individual ASMs.
The present study conducted an external cross-validation between previously existing prediction equations for the ASM of the older Korean population and those developed in this study. Acceptance standards for the predictive accuracy of regression equations were R2 above 0.800 without bias, SEE below 1.45 kg ALM for males, SEE below 1.16 kg ALM for females, and PIA above 70% [13, 26, 27, 30].
The two-MHz high-frequency BIA regression equation, the regression equation reported by Vermerien et al., and that reported by Perniche met the acceptance standards for group and individual precision. On the other hand, regression equations by Scafogliery, Sergi, Kyle, and Kim met standards for explanatory variance, but had large SEE, and had PIAs largely below the acceptance standard. This prevented these equations from meeting the acceptance standards for predictive accuracy and was unsuitable for application to older Korean people. The predictive equations reported by Scafogliery, Sergi, and Kyle were mostly developed based on Caucasian and African American populations, which made their application to Koreans presumably difficult. White and African Americans have a different relationship between muscle mass and body electrical conductivity and resistance due to their relatively shorter torso and longer limbs in comparison to East Asians and a higher body density in comparison to Asians [8, 13].
These discrepancies remark the need for the development of prediction equations specific to Asian populations. In contrast, while the BIAKim [20] developed based on Korean or Asian Seniors had the advantage of using a relatively high frequency of 250 kHz, the study shared the problems of low accuracy observed during development of the prediction equation and large margins of error to have the lowest accuracy overall. Indeed, the TE of BIAKim was 8.21 kg ALM in contrast to the TE of 0.97 kg ALM of the BIA prediction equation developed in this study [20]. This made it necessary for Kim’s prediction equation for Koreans and Asian populations to be replaced by the new prediction equation proposed in this study. The overall results of External Cross-Validation demonstrated that the prediction equation of Vermeiren, that by Peniche, and the prediction equation newly developed in this study were able to provide accurate predictions for Korean seniors in both group and individual levels.
Finally, BIA@2 MHz, BIAVermeiren, and BIAPeniche were verified to check their applicability for sarcopenia diagnosis. The two BIA prediction equations by Vermeiren and Peniche were 55.4% and 73.3% overall agreement and Cohen’s Kappa of 0.216 and 0.300, causing detailed analysis on sensitivity, specificity, positive predictive value, and negative predictive value to be impossible. In contrast, the newly developed appendicular lean-specific high-frequency BIA prediction equation had an overall agreement of 94.9% and Cohen’s Kappa of 0.779 (almost perfect), making evaluation possible due to its sensitivity, specificity, positive predictive value, and negative predictive value to be determined suitable for individual sarcopenia diagnoses and clinical application. In particular, the specificity, negative predictive value, and positive predictive value showed potential for accurate diagnosis above 90%, while the sensitivity of 71.4% was much higher than the previously reported 37%–55% [34] and showed improved results with a sensitivity of 63.3% from the recently reported multifrequency prediction equation using 250 kHz ZI and 50 kHz Xc. The improved values of sensitivity, specificity, positive predictive value, and negative predictive value would be applicable to epidemiological studies and clinical settings to a level comparable to that of DXA, unlike previous studies. As stated previously, such improvements would be based on the electrophysiological principle behind the usage of ZI within the 2 MHz–3 MHz range that allows sufficient conductivity in skeletal muscles [26, 32, 35]. It is recommended that future studies develop BIA prediction equations for appendicular skeletal muscle using ZI within the high frequency range of 2 MHz–3 MHz.
This study developed a new appendicular lean-specific high-frequency BIA prediction equation for the prediction of ALM based on 195 healthy older people aged 70 years and over in Korea. The relatively narrow range of age and BMI, living environment, the health status of the participants, and the sample size of the population under study, could affect the generalizability of our results. This study used DXA as a reference method for the ASM measurement. There are several limitations to using DXA to measure the ALM. ALM measurements were obtained by calculating the total muscle mass by subtracting the mineral content from the lean arm and leg mass. DXA does not separate skeletal muscle from the skin, connective tissue, or blood vessels [36, 37]. Thus, DXA may overestimate ALM [31, 32] compared to the gold standard (i.e., CT or MRI) to quantify body composition. Although there is a high correlation between DXA and the gold standard, DXA has been considered as the reference standard for the BIA model [3, 10, 38,39,40], fat infiltration that occurs in aging upon the replacement of water and connective tissue within muscle tissue with fat can have some influence on the accuracy of skeletal muscle mass measured using DXA [36, 41].
This study only investigated the elderly in the Seoul area of South Korea, so caution is required when generalizing its results to other populations. In addition, this study only measured the ALM of the elderly to diagnose sarcopenia, which has certain limitations in assessing sarcopenia in the elderly. Therefore, future research could add tests to diagnose severe sarcopenia (including muscle strength and performance), and also expand to different groups (children, adolescents and adults) and increase the sample size. Regarding the level of physical activity (PA), we recruited elcerly people who were apparently healthy and able to do activityes of daily living as the research participants. Participants' PA levels may also have an impact on BIA results. Future research should take morbidity/co-morbidities into consideration.