Reference Values for Skeletal Muscle Mass – Current Concepts and Methodological Considerations

Assessment of a low skeletal muscle mass (SM) is important for diagnosis of ageing and disease-associated sarcopenia and is hindered by heterogeneous methods and terminologies that lead to differences in diagnostic criteria among studies and even among consensus definitions. The aim of this review was to analyze and summarize previously published cut-offs for SM applied in clinical and research settings and to facilitate comparison of results between studies. Multiple published reference values for discrepant parameters of SM were identified from 64 studies and the underlying methodological assumptions and limitations are compared including different concepts for normalization of SM for body size and fat mass (FM). Single computed tomography or magnetic resonance imaging images and appendicular lean soft tissue by dual X-ray absorptiometry (DXA) or bioelectrical impedance analysis (BIA) are taken as a valid substitute of total SM because they show a high correlation with results from whole body imaging in cross-sectional and longitudinal analyses. However, the random error of these methods limits the applicability of these substitutes in the assessment of individual cases and together with the systematic error limits the accurate detection of changes in SM. Adverse effects of obesity on muscle quality and function may lead to an underestimation of sarcopenia in obesity and may justify normalization of SM for FM. In conclusion, results for SM can only be compared with reference values using the same method, BIA- or DXA-device and an appropriate reference population. Limitations of proxies for total SM as well as normalization of SM for FM are important content-related issues that need to be considered in longitudinal studies, populations with obesity or older subjects.


Introduction
Beyond the well-established role of ageing associated loss in skeletal muscle mass (SM) (primary sarcopenia) as a risk factor of frailty, morbidity and mortality in older people, a low SM is observed as a result of diseases like malignant cancer, chronic obstructive pulmonary disease, heart failure and renal failure (secondary sarcopenia [1]) and is also an emerging prognostic marker in a number of diseases [2][3][4][5][6][7][8][9][10][11][12]. The etiology for sarcopenia as a risk factor might be partly explained by the correlation between SM and cardiac, respiratory or immune function but remains to be investigated further in order to understand the preventative and therapeutic potential of SM. Muscle not only functions as the major tissue for insulin-stimulated glucose uptake, amino acid storage and thermoregulation, but is also secreting a large number of myokines that regulate metabolism in muscle itself as well as in other tissues and organs including adipose tissue, the liver and the brain [13,14]. The recent popularity of SM outpaced the interest in fat mass (FM) that only has a limited and inconsistent impact on morbidity and mortality [15,16]. The assessment of SM by segmentation of continuous whole body magnetic resonance imaging (MRI) is considered as the gold standard [17]. However, this method is too cumbersome and expensive for clinical practice and is even rarely used in studies with larger sample sizes [17,18]. Instead, single slices at different reference sites measured by MRI or obtained from routine computed tomography (CT) examinations are taken as a proxy for the total tissue volume (e.g., L3 muscle cross-sectional area [17,19]). Most commonly, dual X-ray absorptiometry (DXA) is used to assess appendicular lean soft tissue (ASM, the sum of lean soft tissue from both arms and legs) or fat-free mass (FFM, total lean soft tissue plus bone mineral mass or body weight minus FM) as a proxy for SM. More simple and even non-invasive, the output of bioelectrical impedance analysis (BIA) depends on the reference method used to generate the BIA algorithm and can be FFM [20], ASM, e.g., [21][22][23] or even SM, e.g., [24][25][26][27].
To facilitate comparison between studies and to evaluate individual results for SM in patients, it is important to understand the differences between parameters and cut-offs for SM. These differences are not only method inherent but also depend on characteristics of the study population (e.g., ethnicity, age and disease). Device-specific characteristics by different manufacturers determine the validity and precision of parameters for SM. In addition, the available reference values differ with respect to parametric normalization (linear regression or indexing) to account for body size. Further complexity to the definition of a normal SM is derived from the concept of sarcopenic obesity [28]. Since high levels of FM may adversely affect the quality and function of SM [29,30], a normal SM may also depend on the amount of FM.
Different professional associations have published definitions of sarcopenia based on an estimate of SM and impaired muscle strength and/or physical performance [31][32][33][34][35][36][37], but no consensus definition has yet been reached. The aim of this review is not to provide an optimal diagnosis of sarcopenia but to compare current definitions of a low SM considering the impact of the underlying methodological assumptions, limitations and normalization of SM parameters for height, weight, body mass index (BMI) or FM.

Methods
In order to identify reference values for SM, seven consensus reports were reviewed [31][32][33][34][35][36][37]. Further studies were identified through reference lists and a search for relevant articles based on the keywords "sarcopenia", "low muscle mass", "cut-off sarcopenia", "reference value sarcopenia", "sarcopenic obesity". Only parameters of SM normalized for height, weight, BMI or FM were considered. To be included in this article, studies were required to contain the following information: method of SM assessment (device), cut-off points for SM and description of the reference population including geographical location, sample size, distribution between sexes and age (range and/or standard deviation (SD) ± mean). Only English language articles were considered. Therefore, 64 studies were identified that met the inclusion criteria. Main reasons for the exclusion of articles were duplicate analyses conducted on the same reference population (only the first published paper was included), a missing normalization of reference values, a sample size <200 subjects (sample size <200 subjects will not be representative for both sexes, all ages and BMI-groups), the use of anthropometric measures to determine a low SM and the adoption of previously published cut-offs regarding SM and obesity. 24.6 ± 3.8 24.1 ± 5.4 → 2 SDs below the sex-specific means of young, healthy adults ASMI, appendicular skeletal muscle mass index; BMI, body mass index; DXA, dual X-ray absorptiometry; NA, not available; SD, standard deviation; SM, skeletal muscle mass; SMI, skeletal muscle mass index. Table 2. Cut-off values and diagnostic criteria of a low muscle mass using bioelectrical impedance analysis (BIA).

Reference
Device  28.1 ± 5.5 → Class I sarcopenia: 1 SD below the means for young, healthy adults → Class II sarcopenia: 2 SDs below the means for young, healthy adults  26.6 ± 4.3 27.0 ± 5.5 → receiver operating characteristics ASMI, appendicular skeletal muscle mass index; BIA, bioelectrical impedance analysis; BMI, body mass index; FFM, fat-free mass; NA, not available; SD, standard deviation; SM, skeletal muscle mass; SMI, skeletal muscle mass index. Table 3. Cut-off values and diagnostic criteria of a low muscle mass using computed tomography (CT).

Reference
Device   Table 4 shows reference values of 34 publications for a low SM in combination with different measures of obesity. Cut-offs for a low SM were mostly determined by DXA or BIA, whereas only a few studies reported CT-defined cut-offs in combination with obesity criteria. SM parameters were commonly normalized for height squared or given as % of body weight. In addition, two studies adjusted ASM for BMI [66,67]. Alternative parameters were FM/FFM ratio [68], visceral fat area/thigh muscle area ratio (VFA/TMA) [69] and fat mass index (FMI) in combination with fat-free mass index (FFMI) [70].

Combination of Measures for Muscle mass and Obesity
Prado et al. [71] published CT-derived SMI cut-offs determined in a population of obese (BMI ≥ 30 kg/m 2 ) Canadians with tumors of the respiratory or gastrointestinal tract. In 2013, this CT database was extended by Martin et al. [72] and low SM reference values were reported for subjects with normal weight and overweight according to BMI classifications. In both studies, optimal stratification was used to determine the threshold of mortality. Many studies adopted the criteria proposed by Prado et al. [71] and Martin et al. [72] (e.g., [73][74][75]). Only one further study developed BMI-dependent reference values for SM [76]. Although some studies referenced the cut-offs by Prado et al. [71], reported thresholds differ from the original work (e.g., [77,78]). These reported values were then cited in further studies [79].
Other criteria include %FM [50,81,[96][97][98][99][100][101], visceral fat area [73] or fat-muscle ratios like visceral fat area (VFA) to total abdominal muscle area (TAMA) [74]. Table 5 displays cut-offs and average values for body composition stratified into groups of subjects with underweight, normal weight, overweight and obesity. Cut-offs for FMI DXA were released by the National Health and Nutrition Examination Survey (NHANES; [102]) and respective BMI-dependent normal values for FFMI DXA were calculated as BMI minus FMI. For each given BMI displayed in Table 5, corresponding normal value for SMI MRI were calculated using a stepwise regression analysis (SMI MRI , men = 0.479 × FFMI DXA −0.017 × age + 0.683 and SMI MRI , women = 0.348 × FFMI DXA − 0.011 × age + 1.971) in a healthy Caucasian population. In addition, respective values for SMI BIA validated against MRI were generated based on a young and healthy Caucasian population using linear regression analysis (SMI BIA , men = 0.168 × BMI + 5.49 (R 2 = 0.53, standard error of estimate (SEE) = 0.514) and SMI BIA , women = 0.159 × BMI + 3.72 (R 2 = 0.61, SEE = 0.465)). Adjacent to the average SMI BIA (median) for each BMI, cut-offs with two SDs below the sex-specific mean of the young and healthy population were shown. 24.1 ± 2.5 → Class I sarcopenia: 1 SD below the sex-specific means of young adults → Class II sarcopenia: 2 SDs below the sex-specific means of young adults 30.9 ± 7.9 BMI (kg/m 2 ) 35.1 ± 4.6 → Class I sarcopenia: 1 SD below the sex-specific means of young adults → Class II sarcopenia: 2 SDs below the sex-specific means of young adults     24.6 ± 3.8 24.1 ± 5.4 (a) → 2 SDs below the sex-specific means of young, healthy adults (b) → >sex-specific median

Discussion
SM has evolved as the most promising body composition parameter associated with health risk in ageing and many chronic diseases [1]. Evaluation of SM is complicated by a variety of available methods that provide different outcome parameters as a proxy for total body SM. Therefore, it is important to have accurate reference values that apply to the patient or population under study as well as to the respective body composition method. In this review, we identified multiple published reference values for discrepant parameters of SM (Tables 1-4), discussed the differences in the underlying assumptions and limitations as well as different concepts for normalization of SM parameters for height, weight, BMI or FM.
Imaging technologies are thought to provide the best assessment of SM. Briefly, segmentation of transversal images by special software (e.g., SliceOmatic Tomovision, version 4.3; Montreal, Québec, Canada) results in muscle areas that are multiplied by the correspondent slice thickness to calculate muscle volume [27] that is transformed to SM by assuming a constant density (1.04 kg/L) of adipose tissue-free SM [103]. Muscles at the head, hands and feet are commonly neglected in this approach. The precision of whole body SM MRI is high (intra-observer coefficient of variation = 1.8% [104]). Reference data for total SM based on the gold standard whole body MRI (Table 5) are scarce due to high costs and cumbersome image-segmentation [17,18]. However, whole body MRI was integrated in the assessment of current large and representative national databases like the UK biobank [105] or the national cohort (NAKO) in Germany [106]. Future evaluation of these databases will provide the basis of statistically derived normal values whereas prospective investigation of mortality or correlation with frailty, fracture risk, glucose or amino acid metabolism would allow to establish even more meaningful disease-specific cut-offs.
Instead of whole body imaging, reference values for L3 single slices are frequently published (Tables 3 and 4), especially in patients where CT images are routinely applied for cancer staging. The use of these cut-offs may be specific for the population studied and transferability of the results to other patient groups needs to be investigated. Radiation exposure is a major limitation that confines the application of CT to individual transversal images or the secondary analysis of routine clinical measurements. As a further drawback, clinical CT protocols for L3 are not standardized across hospital sites. SMA at L1, L2, L4, L5, and the thoracic vertebra T12, T11, and T10 were reported to be suitable alternatives to SMA measured at L3 [58]. Nonetheless, there are also advantages of CT images with a high resolution and precision of the measurement. Most studies report the precision of single slice CT scan analysis to range between 1% and 2% [107]. Thus, automated segmentation is facilitated by using a characteristic range of Hounsfield units for fat-free muscle tissue [107,108]. CT can also differentiate individual muscle or muscle groups and can thus for example investigate the impact of pectoralis muscle area for survival at the Intensive Care Unit [12] because respiratory musculature may determine weaning from mechanical ventilation. On the other hand, characteristic changes in the Hounsfield distribution of muscle can reveal qualitative changes of the tissue (e.g., fatty infiltration or edema) that have been found to be of prognostic value [71].
DXA is the most commonly used method for assessment of SM (Table 1). Lean soft tissue at the arms and legs (ASM) is highly correlated with muscle volume derived from imaging studies (correlation coefficients ranging from 0.77 to 0.97 for both, whole body and regional scans [51,[109][110][111][112][113][114][115]). However, only 44% of total lean soft tissue is derived from extremities (unpublished results) and only part of total lean soft tissue is SM. Therefore, SM measured by DXA is considerably higher when compared with muscle volume measured by imaging technologies [27,116]. Precision errors for total ASM are reported to be low (1-3%), device specific and depend on population characteristics like age or prevalence of obesity [117].
BIA can assess SM, ASM or FFM, depending on the reference method used to generate the BIA-algorithm. The choice of the BIA-algorithm not only depends on the desired target-parameter but also on the agreement between the BIA-device or reference population used to generate the BIA-algorithm and the BIA-device and patient characteristics to be evaluated [118]. However, in two studies, the equation by Janssen et al. [56] that is not suitable for Asians was used to predict SM in Asian populations [53,55] with only one study providing a validation in 41 Taiwanese people (age: 20-99 years; BMI: 17.6-34.6 kg/m 2 ) [55]. Except for the study by Masanés et al. [26], all other studies used different BIA devices than Janssen et al. [56] (Table 2). Validity and precision of BIA results differ between manufacturers and depend on the hardware as well as the appropriate validation of the BIA-algorithm [119]. Discrepancies in the assumptions of the homogeneous bioelectrical model that lead to a higher measurement error occur with changes in hydration (e.g., edema) and with differences in body shape that are associated with aging (decreasing limb relative to trunk diameter), obesity (apple and pear shape of body fat distribution) and ethnicity (trunk to leg length, regional adiposity and muscularity). Therefore, segmental BIA that can measure the relative contribution of trunk and extremities to total body conductivity may help to reduce assumptions on body shape leading to an improved prediction compared with conventional wrist-ankle measurements [27]. The accuracy of phase-sensitive segmental BIA compared with MRI as a reference is clinically acceptable when whole body SM was assessed (two SDs: 11-12% for different ethnicities) but it was low when small compartments of the body were assessed (e.g., two SDs: 20-29% for the arms) [27].

Limitations of Proxies for Total Skeletal Muscle
Single SMA at L3 level turned out to be the best compromise site to assess volumes of total SM together with visceral adipose tissue (VAT) and subcutaneous adipose tissue (SAT) (r = 0.832-0.986; p < 0.01 [17]). Furthermore, SMA at L3 is considered as a valid proxy for whole body FFM (r = 0.940; p < 0.001 [120]). Other authors reported high correlations between single abdominal SMA at L4-L5 intervertebral space and total SM (r = 0.710-0.920 [121]), whereas the use of PMI to determine whole body SM is controversial because psoas is a relatively small muscle. A good correlation between PMI and SMI measured by BIA in healthy 35 Asian liver donors (r = 0.737; p < 0.001) and a moderate correlation in 137 living donor liver transplantation recipients (r = 0.682; p < 0.001) were found [63]. Other authors argue that L3 PMA is not representative of total SM [122,123]. Despite acceptable correlations, the accuracy of single images is limited in individual cases. Likewise, it is well established that the correlation between BMI and FM is fairly good at the population level whereas at the individual level BMI is only a poor indicator of adiposity [124]. In addition, validity of the assessment of changes in SM during follow-up is limited by the use of individual images from L3 or mid-thigh. These images cannot be used as pars pro toto because of regional differences in changes of muscle volume with age or obesity (e.g., the contribution of SM MRI at the arms and legs to ASM tended to decrease at higher adiposity in both genders [104]).
Similarly, ASM has limitations to assess the change in total SM with ageing or overweight and obesity. Since lean soft tissue from the extremities also contains lean compartments from connective tissue (e.g., skin and adipose tissue), SM accounts for only about 50% of FFM in obesity [116]. ASM was therefore shown to overestimate appendicular SM assessed by MRI with increasing BMI [27]. In line with this finding, DXA was also shown to underestimate the age-related loss of thigh muscle mass in comparison with MRI [125]. Furthermore, DXA measures of change in lean mass before and 10-week after resistance training were only modestly associated with MRI measures of change in muscle volume [126].
In summary, the random error of single images or ASM as a proxy for total SM limits the applicability of these substitutes in individual cases and together with the systematic error limit the accurate detection of changes in SM.

Normalization of Skeletal Muscle Mass for Body Size and Obesity
Normalization of lean mass for weight is inappropriate because two people with the same %FFM who differ in height have a different nutritional status, with the taller person having a lower muscularity [127]. FFM has been shown to scale to height with a power of around two in different ethnicities, ranging from 1.86 in non-Hispanic white women to 2.32 in non-Hispanic black men [128]. Consequently, appropriate normalization of total SM, SM-area, ASM and FFM is performed for height 2 .
In addition to the physiologic increase in SM with height, there is also an increase in SM with weight gain that depends on the initial amount of FM [129]. The evaluation of SM may thus also depend on the amount of FM. With increasing obesity, adverse effects on myocyte metabolism, muscle tissue composition and peak force generation can be mediated via paracrine signaling of proinflammatory immune cells in intermuscular adipose tissue [30]. The same SM at a higher FM may also lead to a limitation of strength and increased disability because at the same work load, energy expenditure and muscle force are higher for a person with obesity [130]. In line with these mechanisms, patients with a low SM and a concomitant high FM were shown to have a higher morbidity and mortality when compared to patients with a high FM only (for review see [131]). However, it remains unclear whether the risk of a low SM and a high FM is additive or if the risk of a high FM is disproportionally higher at a concomitantly low SM.
Published definitions of sarcopenic obesity use BMI to assess overweight and obesity in combination with fixed cut-offs for a low SM that are derived from subjects with normal weight and/or overweight [72,76]. To the best of our knowledge, all current definitions disregard the relationship between fat and lean mass that can be investigated by applying the Forbes rule (energy partitioning, i.e., the fraction of energy lost or gained as protein, is a nonlinear function of FM [129]) or the Hattori chart (two dimensional plot of FMI vs. FFMI [132]). Table 5 provides novel BMI-dependent SMI cut-offs.
The combination of FFMI with FMI [133], %FM [6,8] or BMI [134] facilitate to investigate the proportional contribution of fat and lean compartments to health risk as well as their presumable interaction. An attractive alternative to the simultaneous use of two indices is integration of information on fat and lean compartments in one index as FM/FFM 2 . This index was proposed by Wells and Victoria who determined the appropriate power by which to raise the denominator from regressing FM on FFM [135]. The usefulness of this index needs to be investigated in future studies because it depends on a linear correlation between FM and FFM 2 , as well as on absence of heteroscedasticity.
Moreover, suitable reference values require an appropriate sample size ideally comprised of healthy or "normal" subjects (normative approach) or derive cut-offs from an older population or a group of patients (stratification approach). In addition, reference values can be reported using parametric methods, like Z-scores or 2 SDs below the mean, that rely on normal distribution of the data, on the absence of residual associations, and on constant variance of the normalized measurements throughout the entire sample (absence of heteroscedasticity, logarithmic transformation of the dependent variables or weighted regression models). In Tables 1-4, most studies used cut-off thresholds for low SM on the basis of young healthy adults' reference groups according to the recommendations proposed by the European Working Group on Sarcopenia in Older People [32]. The majority of these studies used two SDs below the means of healthy young subjects as a cut-off, e.g., [21,39,40,44,45,50] whereas other studies defined a low SM as one SD below the mean, e.g., [85,90,94,95]. Six articles stratified the cut-offs according to severity of a low SM [22,44,46,49,76,80]. One SM threshold was based on the fifth percentile [59] or on the 20th percentile [92] or on the 50th percentile [89]. Other studies used the sex-specific lowest quintiles [43], quartiles [47,62], tertiles [84], the lower two quintiles of the study population [98,100] or the lowest 20% of the distribution [38,42,48]. In one study, receiver operating characteristics analysis was used to develop SM cut-offs associated with physical disability [24]. In four studies, optimal stratification was used to determine the SM threshold of mortality risk in cancer patients [64,65,71,72]. Further diagnostic criteria applied classification and regression tree analysis [66,67].

Conclusions and Recommendations
In summary, published reference values for SM differ widely dependent on the outcome parameter and reference population. Results should consider the limitation of all proxies for total SM with respect to application in individual cases as well as for measurement of changes in SM. To facilitate comparison between results of different studies, authors should use a unified nomenclature for outcome parameters and indicate the device and software version of the body composition analyzer. In addition, the choice of body composition method should depend on the aim of the study. For assessment of changes in SM and evaluation of individual patients, a high precision is required that is, for instance, not fulfilled when segmental bioelectrical impedance is used to assess limb SM. The adverse effects of obesity on muscle quality and function may lead to an underestimation of sarcopenia in obesity and therefore requires normalization of SM for FM.