Original Research
Musculoskeletal Imaging
October 25, 2023

Skeletal Muscle Area on CT: Determination of an Optimal Height Scaling Power and Testing for Mortality Risk Prediction

Please see the Editorial Comment by Linda C. Chu discussing this article.
To listen to the podcast associated with this article, please select one of the following:
iTunes, Spotify, Google, or direct download.

Abstract

BACKGROUND. Sarcopenia is commonly assessed on CT by use of the skeletal muscle index (SMI), which is calculated as the skeletal muscle area (SMA) at L3 divided by patient height squared (i.e., a height scaling power of 2).
OBJECTIVE. The purpose of this study was to determine the optimal height scaling power for SMA measurements on CT and to test the influence of the derived optimal scaling power on the utility of SMI in predicting all-cause mortality.
METHODS. This retrospective study included 16,575 patients (6985 men, 9590 women; mean age, 56.4 years) who underwent abdominal CT from December 2012 through October 2018. The SMA at L3 was determined using automated software. The sample was stratified into two groups: 5459 patients without major medical conditions (based on ICD-9 and ICD-10 codes) who were included in the analysis for determining the optimal height scaling power and 11,116 patients with major medical conditions who were included for the purpose of testing this power. The optimal scaling power was determined by allometric analysis (whereby regression coefficients were fitted to log-linear sex-specific models relating height to SMA) and by analysis of statistical independence of SMI from height across scaling powers. Cox proportional hazards models were used to test the influence of the derived optimal scaling power on the utility of SMI in predicting all-cause mortality.
RESULTS. In allometric analysis, the regression coefficient of log(height) in patients 40 years old and younger was 1.02 in men and 1.08 in women, and in patients older than 40 years old, it was 1.07 in men and 1.10 in women (all p < .05 vs regression coefficient of 2). In analyses for statistical independence of SMI from height, the optimal height scaling power (i.e., those yielding correlations closest to 0) was, in patients 40 years old and younger, 0.97 in men and 1.08 in women, whereas in patients older than 40 years old, it was 1.03 in men and 1.09 in women. In the Cox model used for testing, SMI predicted all-cause mortality with a higher concordance index using of a height scaling power of 1 rather than 2 in men (0.675 vs 0.663, p < .001) and in women (0.664 vs 0.653, p < .001).
CONCLUSION. The findings support a height scaling power of 1, rather than a conventional power of 2, for SMI computation.
CLINICAL IMPACT. A revised height scaling power for SMI could impact the utility of CT-based sarcopenia diagnoses in risk assessment.

Highlights

Key Finding
In the test set of 11,116 patients, Cox proportional hazard models incorporating SMI and age predicted all-cause mortality with a higher concordance index using a derived optimal height scaling power of 1 than the conventional scaling power of 2 in men (0.675 vs 0.663, p < .001) and women (0.664 vs 0.653, p < .001).
Importance
The conventional SMA height scaling power of 2 is not supported by allometric modeling and may contribute to overdiagnosis of sarcopenia in tall patients.
Sarcopenia is a condition characterized by depletion of normal skeletal muscle accompanied by low strength and physical performance [1]. Sarcopenia is now recognized as an independent predictor of numerous adverse health outcomes, including health-related quality of life [2], number of days of hospitalization [3], major postoperative complications [4], poor outcomes after cancer treatment [5], and mortality [68].
The study of sarcopenia has evolved rapidly and has helped clarify three important issues. First, although sarcopenia is associated with aging, it is not a normal part of aging; in fact, muscle depletion is increasingly recognized to occur in young patients, including those with cancer, trauma, or malnutrition [9, 10]. Second, although sarcopenia was previously viewed as an inherently chronic process, recognition is increasing that stressor events (e.g., surgery or hospitalization) can result in rapid-onset muscle insufficiency described as “acute” sarcopenia [11]. Third, evidence-based sarcopenia treatments are now available [1219], and recent cost-effectiveness analyses of sarcopenia management interventions have shown favorable results [20, 21]. These advances highlight an unmet need for effective diagnosis of sarcopenia in the general population.
Opportunistic CT has emerged as a value-added method of screening for imaging features of sarcopenia using clinically indicated CT examinations, without additional cost or radiation exposure [22]. When opportunistic CT evaluation for sarcopenia is performed, the skeletal muscle area (SMA) is typically measured (in square centimeters) on an axial image at the L3 level. This cross-sectional area measurement is then adjusted (or indexed) to the patient's height using a scaling power of 2 (i.e., adjusted to the square of the patient's height), yielding a skeletal muscle index (SMI; expressed as square centimeters per square meter) [23, 24].
SMI using a height scaling power of 2 was initially adopted in 2008 for CT muscle area measurements, analogous to the height scaling power used for calculations of BMI (weight in kilograms divided by the square of height in meters) and the dual-energy x-ray absorptiometry (DEXA)-based ratio of appendicular lean mass (ALM) to height (with ALM expressed as kilograms divided by the square of height in meters) [25]. SMI is the imaging metric most frequently used to define sarcopenia and to study sarcopenia-related clinical outcomes using widely adopted dichotomous cutoffs [25, 26]. Interest has also increased in the use of age- and sex-specific SMI reference values for diagnosing sarcopenia [2730].
Two recent studies have questioned the current convention of defining SMI on CT examinations as SMA scaled by height to a power of 2 [31, 32]. Based on allometric analyses (i.e., analyses of the size of a body part relative to the size of the entire body), those studies concluded that the height scaling power for SMA should be less than 2. Moreover, those studies suggested that the optimal scaling power could differ between men and women.
A revision to the current convention for defining SMI using an SMA height scaling power of 2 would greatly impact the CT diagnosis of sarcopenia and thus also impact prognostic models using sarcopenia. Indeed, the validity of an adjusted definition of SMI could be assessed by exploring the prognostic utility of SMI when the adjusted definition is used. Moreover, any evaluation of the optimal SMA height scaling power should be performed in large samples of both men and women of varying ages.
The aim of this study was to determine the optimal height scaling power for SMA measurements on axial CT images at the L3 level and to test the influence of the derived optimal scaling power on the utility of SMI in predicting all-cause mortality.

Methods

Patients

Institutional review board approval was obtained for this HIPAA-compliant retrospective study, with the requirement for written, informed consent waived. The Stanford University Medical Center radiology database was searched for patients who underwent abdominal CT examinations from December 2012 through October 2018, and 19,713 patients were identified. In patients who underwent multiple abdominal CT examinations during the study period, one such CT examination was selected at random for further consideration for study inclusion. A total of 3138 patients were then excluded for the following reasons: missing data in the patient's EHR entry (n = 2341), age younger than 18 years (n = 226), outlier value for height or weight (as described later in this paragraph) (n = 432), outlier value for SMA measurement (as described later in the Methods section) (n = 21), and lack of follow-up data in the EHR for a patient in the test set (as described later in the Methods section) (n = 118). These exclusions resulted in a final study sample of 16,575 patients (6985 men, 9590 women; mean age, 56.4 ± 18.3 [SD] years). Height was considered an outlier value when less than 125 cm (n = 65) or greater than 196 cm (n = 45), and weight was considered an outlier value when less than 30 kg (n = 32) or greater than 136 kg (n = 359), to conform to the population from the National Health and Nutrition Examination Survey (NHANES), consistent with the approach by Heymsfield et al. [33]. Figure 1 shows the flow of patient selection.
Fig. 1 —Flowchart shows patient selection. SMA = skeletal muscle area.
The patient sample was stratified into two subsets: a group without major medical conditions that was included for the purpose of determining the optimal height scaling power and a group with major medical conditions that was used to test the influence of the derived optimal height scaling power on the utility of SMI as a predictor of all-cause mortality. To define these two groups, the EHR was searched for ICD-9 and ICD-10 codes corresponding to major medical conditions associated with muscle depletion [6, 22, 23] (Table S1): metastatic cancer, malnutrition, nutritional deficiency, obesity, cerebral vascular disease, paralysis, amputation, chronic kidney disease, diabetes, or ischemic heart disease. Patients were considered to have the given medical condition if the code was entered into their record any time before the date of the CT examination or up to 1 month after the CT examination. A total of 5459 patients did not have any of these major medical conditions. Thus, a final sample of 5459 patients (2268 men, 3191 women; mean age, 48.3 ± 17.5 years) without a major medical condition was included in the analysis for determining the optimal height scaling power. The remaining 11,116 patients (4717 men, 6399 women; mean age, 60.4 ± 17.3 years) who had a major medical condition and available follow-up data in the EHR were included in the analysis for testing the influence of the derived optimal height scaling power on the utility of SMI as a predictor of all-cause mortality.

CT Examinations and Image Analysis

CT examinations were performed for a spectrum of indications on one of 17 MDCT scanners from four manufacturers. All scanners underwent daily calibration using a quality assurance phantom, in accordance with the specifications of the American College of Radiology. CT protocol parameters included a tube voltage of 120 kV (range, 70–140 kV), slice thickness of 1.25 mm (range, 0.5–5 mm), rotation time of 0.5 second, effective tube current setting based on BMI, and variable utilization of contrast material. Based on prior studies, the effect of radiation dose, slice thickness, and IV contrast material on SMA measurements was considered to be small and clinically insignificant [3436]. All examinations covered the L3 level.
To evaluate the CT examinations, a slice at the L3 level was first selected using the open-source tool developed by Kanavati et al. [37]. Then, automated deep learning open-source software (Comp2Comp) was used to segment skeletal muscle at the selected slice [38]. This software used a model based on the 2D UNet architecture [39] and returned an SMA measurement. A prior study that evaluated the tool in a holdout internal test set of 40 CT examinations observed a mean Dice score of 0.97 ± 0.03 (SD) for muscle segmentation at the L3 level compared with reference manual segmentations, and a mean percentage error in SMA of less than 2% [38]. Examples of the automated image segmentation outputs are presented in Figure 2.
Fig. 2A —CT images show automated segmentation of skeletal muscle.
A, Axial CT images obtained at L3 level in 21-year-old woman (A) and 47-year-old man (B) show automated segmentation of skeletal muscle (coral), which was used to determine skeletal muscle area (SMA). SMA measured 114 cm2 (A) and 194 cm2 (B). Green indicates automated segmentation of visceral adipose tissue (VAT), orange denotes automated segmentation of subcutaneous adipose tissue (SAT), and blue represents automated segmentation of intermuscular adipose tissue (IMAT).
Fig. 2B —CT images show automated segmentation of skeletal muscle.
B, Axial CT images obtained at L3 level in 21-year-old woman (A) and 47-year-old man (B) show automated segmentation of skeletal muscle (coral), which was used to determine skeletal muscle area (SMA). SMA measured 114 cm2 (A) and 194 cm2 (B). Green indicates automated segmentation of visceral adipose tissue (VAT), orange denotes automated segmentation of subcutaneous adipose tissue (SAT), and blue represents automated segmentation of intermuscular adipose tissue (IMAT).
Automated segmentation was considered to have failed if the SMA value was less than 20 cm2. This threshold was selected on the basis of the investigators' earlier experience, which suggested that outputted SMA values below this threshold are likely to result in underestimation of the true SMA. Two radiologists (L.Y. and L.L., with 31 and 27 years of posttraining experience, respectively) independently performed a qualitative assessment of the adequacy of automated SMA segmentation in 100 randomly selected examinations from the group of patients without major medical conditions. For patients in whom the automated segmentation was deemed inadequate, the radiologist recorded whether the automated segmentation underestimated or overestimated true SMA. Examinations in which the automated segmentation was considered inadequate were not excluded from further analyses.

Prediction of All-Cause Mortality

In the subset of patients with major medical conditions, the EHR was reviewed for evidence of death from any cause after the date of the CT examination, as a measure of all-cause mortality. Patients without evidence of death were censored at the time of the last documented clinical encounter. For each patient, the duration of follow-up was recorded as the time between the date that CT was performed and the date of either death or censoring. Patients for whom the EHR contained no follow-up data after the date of the CT examination were excluded.

Statistical Analysis

Determination of the optimal skeletal muscle area height scaling power—Two methods were used to identify the optimal height scaling power for SMA in the subgroup of patients without major medical conditions: allometric analysis and analysis for statistical independence of SMI from height across scaling powers. In accordance with the methods of Derstine et al. [32], both analyses were performed separately in men and women. For patients of each sex, separate analyses were in turn performed for patients 40 years old and younger (reflecting a previously applied age threshold for defining a reference population of adults without age-related sarcopenia [32, 40, 41]), patients older than 40 years old (reflecting a group in whom physiologic age-related muscle loss may become evident), and patients of all ages.
The allometric analysis was similar to the method described by Heymsfield et al. [33], which explored associations between SMA measurements and patient height. The allometric model included an age adjustment in patients older than 40 years old as well as in patients of all ages: SMA = a × ageb × heightc, where a, b, and c are fitted parameters. In patients 40 years old and younger, the allometric model did not use an age correction: SMA = a × heightc. Coefficients for log(height) and the CIs of the coefficients were estimated from linear regression analyses applied to the log-linear forms of the allometric equations. Coefficients for log(height) were compared with values of 0 and 2 using one-sample t tests. Coefficients were compared between men and women using two-sample t tests. Coefficients for log(age) and the CIs of the coefficients were similarly estimated from linear regression analyses applied to the log-linear forms of the allometric equations. The coefficients for age were compared with a value of 0 using one-sample t tests. These coefficients for height and age were evaluated as possible alternate scaling powers in subsequent analyses.
Scatterplots of SMA measurements versus patient height were generated. A plot was constructed for each combination of patient sex and age group. Each plot also displayed predicted SMA based on the conventional height scaling power of 2 and the optimal height scaling power derived from the allometric model for a 40-year-old patient. Additional plots were generated that depicted the differences (i.e., residuals) between observed SMA values and SMA values predicted by the allometric model. These plots were assessed qualitatively for homogeneity of residuals across the ranges of height and age.
Given that SMI should be unbiased between tall versus short individuals, SMI was assessed for statistical independence from height, with varying height scaling powers for SMI applied [31]. To perform this analysis, SMI was calculated separately for men and women using the following equation: SMI = SMA / (ageb × heightc). The exponent b was fixed at an age scaling power derived from the allometric analyses in patients older than 40 years old and was fixed at 0 in patients 40 years old and younger. The exponent c, representing the height scaling power, was varied from 0 to 2 in increments of 0.1. At each value of c, the Pearson correlation coefficient was computed between height and the corresponding computed SMI. For each age group, the optimal height scaling power was defined as the scaling power where the correlation between these variably adjusted SMIs and the original height measurements was closest to 0.
On the basis of the allometric analysis and the analysis for statistical independence from height, a single integer value was selected in men and in women as the optimal height scaling power for SMA.
Testing the influence of the derived optimal height scaling factor on mortality risk prediction using the skeletal muscle index—In the group of patients with major medical conditions, Cox proportional hazards models were studied separately in men and women to assess the utility of SMI for predicting all-cause mortality. This analysis reflected the increasing data showing a role of metrics of sarcopenia as predictors of mortality [68]. In each group, SMI was calculated using both the optimal height scaling power derived in the prior analysis of patients without major medical conditions and using the conventional height scaling power of 2. Model predictors were SMI and patient age. Model performance was assessed using concordance scores; these scores assessed the ability of models to correctly rank-order patients by survival times, with higher concordance scores indicating better model discrimination. HRs for prediction of all-cause mortality were also computed. SMI was transformed to z scores, such that the HR expressed the impact of a 1-SD change in the SMI on mortality risk. A total of 1000 bootstrapped samples of the data were studied to estimate variances for concordance scores for models using optimal and conventional height scaling powers for SMI and also to estimate CIs for the HRs. One-sample t tests were applied to the bootstrap results to assess for statistically significant differences in concordance scores between models.
Software packages—p values of less than .05 were considered statistically significant. All analyses were performed using Python 3.9.12. Python statsmodels 0.14.0 was used to perform the allometry regression analyses. Python lifelines 0.27.7 was used to fit the Cox proportional hazards regression models.

Results

Derivation of Optimal Height Scaling Power

Table 1 summarizes the characteristics of the 5459 patients without major medical conditions who were included in the analysis for determining the optimal height scaling power. Within this group, men and women differed significantly in terms of mean weight (80.7 ± 14.3 [SD] vs 66.7 ± 14.3 kg), height (176.2 ± 8.3 [SD] vs 162.0 ± 7.5 cm), BMI (26.0 ± 4.1 [SD] vs 25.4 ± 5.2), and SMA (161.8 ± 29.8 [SD] vs 108.9 ± 20.3 cm2) (all p < .001). Both radiologists deemed the automated segmentation of skeletal muscle at the L3 level to be adequate in 97 of 100 randomly selected examinations in this group. The two radiologists deemed the automated segmentation to be inadequate in the same three examinations; in these three examinations, the automated segmentation was considered to underestimate the true SMA.
TABLE 1: Characteristics of 5459 Patients Without Major Medical Conditions Who Were Included in Analysis for Determining Optimal Height Scaling Power
VariableAll Patients (n = 5459)Men (n = 2268)Women (n = 3191)p
Age (y)48.3 ± 17.548.4 ± 17.448.2 ± 17.6.79
Weight (kg)72.5 ± 15.980.7 ± 14.366.7 ± 14.3< .001
Height (cm)167.9 ± 10.5176.2 ± 8.3162.0 ± 7.5< .001
BMI25.7 ± 4.826.0 ± 4.125.4 ± 5.2< .001
SMA (cm2)130.9 ± 35.9161.8 ± 29.8108.9 ± 20.3< .001

Note—Data are expressed as mean ± SD. Men and women were compared using t test. SMA = skeletal muscle area.

Allometric analysis—Table 2 summarizes the results of the allometric analysis of SMA at L3 in the 5459 patients without major medical conditions. The regression coefficient for log(height) was as follows: in patients 40 years old or younger, 1.02 in men and 1.08 in women; in patients older than 40 years old, 1.07 in men and 1.10 in women; and in patients of all ages, 1.15 in men and 1.18 in women. All of these coefficients for log(height) were significantly different from 0 and 2 (all p < .05). The regression coefficient was not significantly different between men and women in patients 40 years old and younger, patients older than 40 years old, or patients of all ages (all p < .05). In patients older than 40 years old and in patients of all ages, the regression coefficient for log(age) was negative (range, –0.14 to –0.43) and significantly different from 0 (all p < .05).
TABLE 2: Results of Allometric Analysis in 5459 Patients Without Major Medical Conditions
Age Group and SexaScaling PowerR2
Heightp vs Power of 2p vs Power of 0Agep vs Power of 0
40 y and younger (p = .70)      
Men (n = 831)1.02 ± 0.11< .001< .0010.09
Women (n = 1175)1.08 ± 0.10< .001< .0010.09
Older than 40 y (p = .80)      
Men (n = 1437)1.07 ± 0.10< .001< .001−0.43 ± 0.02< .0010.28
Women (n = 2016)1.10 ± 0.08< .001< .001−0.34 ± 0.02< .0010.25
All ages (p = .73)      
Men (n = 2268)1.15 ± 0.08< .001< .001−0.14 ± 0.01< .0010.17
Women (n = 3191)1.18 ± 0.07< .001< .001−0.16 ± 0.01< .0010.21

Note—Except where otherwise indicated, data are mean ± standard error.

a
p values in this column indicate comparison of height scaling power between men and women for given age group.
Scatterplots of SMA versus height in men and women are shown in Figures 3 and 4, respectively. Figures S1 and S2 show plots of the residuals of observed SMA versus predicted SMA based on the derived optimal scaling powers from the allometric analysis as a function of height in men and women, respectively; the residuals did not show qualitative increases toward the extremes of the height distribution. Figures S3 and S4 show plots of the residuals of observed versus predicted SMA based on the derived optimal scaling powers from the allometric analysis as a function of age in men and women, respectively; the residuals did not show qualitative increases toward the outer edges of the age distribution.
Fig. 3A —Scatterplots show observed skeletal muscle area (SMA) as function of height in 2268 men. Each scatterplot also contains lines reflecting predicted SMA for 40-year-old man as based on allometric model using conventional height scaling power of 2 versus optimal height scaling power derived from allometric model.
A, Scatterplot shows data for men 40 years old and younger.
Fig. 3B —Scatterplots show observed skeletal muscle area (SMA) as function of height in 2268 men. Each scatterplot also contains lines reflecting predicted SMA for 40-year-old man as based on allometric model using conventional height scaling power of 2 versus optimal height scaling power derived from allometric model.
B, Scatterplot shows data for men older than 40 years old (with age as covariate).
Fig. 3C —Scatterplots show observed skeletal muscle area (SMA) as function of height in 2268 men. Each scatterplot also contains lines reflecting predicted SMA for 40-year-old man as based on allometric model using conventional height scaling power of 2 versus optimal height scaling power derived from allometric model.
C, Scatterplot shows data for men of all ages (with age as covariate).
Fig. 4A —Scatterplots show skeletal muscle area (SMA) as function of height in 3191 women. Each scatterplot also contains lines reflecting predicted SMA for 40-year-old woman as based on allometric model using conventional height scaling power of 2 versus optimal height scaling power derived from allometric model.
A, Scatterplot shows data for women 40 years old and younger.
Fig. 4B —Scatterplots show skeletal muscle area (SMA) as function of height in 3191 women. Each scatterplot also contains lines reflecting predicted SMA for 40-year-old woman as based on allometric model using conventional height scaling power of 2 versus optimal height scaling power derived from allometric model.
B, Scatterplot shows data for women older than 40 years old (with age as covariate).
Fig. 4C —Scatterplots show skeletal muscle area (SMA) as function of height in 3191 women. Each scatterplot also contains lines reflecting predicted SMA for 40-year-old woman as based on allometric model using conventional height scaling power of 2 versus optimal height scaling power derived from allometric model.
C, Scatterplot shows data for women of all ages (with age as covariate).
Analysis for statistical independence of skeletal muscle index from height across height scaling powers—Figure 5 shows the results of the analysis for statistical independence of SMI from height across height scaling powers. For this analysis, SMI was calculated using an age scaling factor of −0.43 for men and −0.34 for women in patients older than 40 years old, on the basis of the allometric analyses, and an age scaling factor of 0 in patients 40 years old and younger. The optimal height scaling power, as defined by the scaling power where the correlation between SMI and height was closest to 0, was 0.97 in patients 40 years old and younger, 1.03 in men with older than 40 years old, and 1.10 in men of all ages; it was 1.08 in women 40 years old and younger, 1.09 in women older than 40 years old, and 1.1 in women of all ages. SMI determined using a height scaling power of 2 did not show statistical independence from height in patients 40 years old and younger, as indicated by a correlation of −0.32 in men 40 years old and younger, −0.27 in men older than 40 years old, and −0.25 in men of all ages; the correlation was −0.26 in women 40 years old and younger, −0.25 in women older than 40 years old, and −0.23 in women of all ages.
Fig. 5A —Analyses for statistical independence of skeletal muscle index (SMI) from height across height scaling powers in 5459 patients. Each point represents correlation between height and SMI determined using height scaling power corresponding with x-axis value. For each analysis, height scaling power corresponding with correlation closest to 0 represents optimal power.
A, Graph shows data for patients 40 years old and younger. Optimal height scaling power is 0.97 for men and 1.08 for women.
Fig. 5B —Analyses for statistical independence of skeletal muscle index (SMI) from height across height scaling powers in 5459 patients. Each point represents correlation between height and SMI determined using height scaling power corresponding with x-axis value. For each analysis, height scaling power corresponding with correlation closest to 0 represents optimal power.
B, Graph shows data for patients older than 40 years old. Optimal height scaling power is 1.03 for men and 1.09 for women.
Fig. 5C —Analyses for statistical independence of skeletal muscle index (SMI) from height across height scaling powers in 5459 patients. Each point represents correlation between height and SMI determined using height scaling power corresponding with x-axis value. For each analysis, height scaling power corresponding with correlation closest to 0 represents optimal power.
C, Graph shows data for patients of all ages. Optimal height scaling power is 1.10 for men and 1.17 for women.
Based on the results from the allometric analysis and the analysis for statistical independence of height from scaling power, an optimal height scaling power of 1 was selected for further testing.

Testing of Influence of Optimal Height Scaling Power on Role of Skeletal Muscle Index in Predicting All-Cause Mortality

Table 3 summarizes the characteristics of the 11,116 patients with major medical conditions who were included in the analysis for testing the optimal height scaling power as a predictor of all-cause mortality. In this group, men and women differed significantly in terms of mean age (61.0 ± 16.8 vs 60.0 ± 17.6 years), weight (83.9 ± 18.0 vs 73.6 ± 20.4 kg), height (174.6 ± 8.6 vs 160.9 ± 7.7 cm), BMI (27.4 ± 5.2 vs 28.4 ± 7.5), and SMA (143.0 ± 34.3 vs 103.9 ± 24.1 cm2) (all p < .05).
TABLE 3: Characteristics of 11,116 Patients With Major Medical Conditions Who Were Included in Analysis for Testing Optimal Height Scaling Power
VariableAll Patients (n = 11,116)Men (n = 4717)Women (n = 6399)p
Age (y)60.4 ± 17.361.0 ± 16.860.0 ± 17.6.004
Weight (kg)77.9 ± 20.083.9 ± 18.073.6 ± 20.4< .001
Height (cm)166.7 ± 10.5174.6 ± 8.6160.9 ± 7.7< .001
BMI28.0 ± 6.627.4 ± 5.228.4 ± 7.5< .001
Skeletal muscle area (cm2)120.5 ± 34.7143.0 ± 34.3103.9 ± 24.1< .001

Note—Except where otherwise indicated, data are mean ± SD. Men and women were compared using a t test.

A total of 675 patients (338 men and 337 women) died, with a median duration of 0.4 years (IQR, 0.1–2.3 years) from the date CT was performed to the date of death. Patients who survived had a median follow-up interval of 4.3 years (IQR, 2.7–5.9 years) after the date CT was performed.
Table 4 shows the results of the Cox proportional hazard models for predicting all-cause mortality based on SMI and age in patients with major medical conditions. In men, the concordance index was significantly higher for SMI using the derived optimal height scaling power of 1 than using the conventional height scaling power of 2 (0.675 vs 0.663, p < .001). In men, SMI had HRs of 0.56 (95% CI, 0.50–0.60) for predicting all-cause mortality using a height scaling power of 1 and 0.58 (95% CI, 0.52–0.64) using a height scaling power of 2. In women, the concordance index was significantly higher for SMI using the derived optimal scaling power of 1 than using the conventional height scaling power of 2 (0.664 vs 0.653, p < .001). In women, SMI had HRs of 0.57 (95% CI, 0.50–0.65) for predicting all-cause mortality using a height scaling power of 1 and 0.60 (95% CI, 0.52–0.69) using a height scaling power of 2.
TABLE 4: Cox Proportional Hazards Survival Analysis for Predicting All-Cause Mortality Based on SMI and Age in 11,116 Patients With Major Medical Conditions
Sex and Height Scaling PowerConcordance IndexpaHR for SMIHR for Age
Men (n = 4717) < .001  
10.675 0.56 (0.50–0.60)1.11 (1.00–1.23)
20.663 0.58 (0.52–0.64)1.13 (1.01–1.25)
Women (n = 6399) < .001  
10.664 0.57 (0.50–0.65)1.09 (0.96–1.24)
20.653 0.60 (0.52–0.69)1.13 (0.99–1.28)

Note—Data in parentheses represent 95% CI. p values and HR CIs were computed from 1000 bootstrapped samples. SMI = skeletal muscle index.

a
For comparison of concordance index between height scaling powers of 1 and 2 for given sex.

Discussion

SMI, a marker of possible sarcopenia, is conventionally derived by normalizing SMA, measured on an individual axial CT image, by the square of patient height. The results of the current study challenge this convention. Based on two different analytic methods, the derived optimal height scaling power, rounded to the nearest integer, was 1 in both men and women. This scaling power yielded better performance than the conventional scaling power of 2 when SMI was used to predict all-cause mortality in a test set of patients with major medical conditions.
The current convention of normalizing SMA using a height scaling power of 2 may result in a bias toward overdiagnosis of sarcopenia in tall patients. To illustrate how the SMA scaling power affects SMI, a woman with a height of 1.63 meters (5 feet 4 inches) and an SMA of 86 cm2 would be classified differently by SMI calculated using a conventional height scaling power of 2 (SMI, 32.4) than by SMI calculated using a scaling power of 1.0 (SMI, 52.8), considering that at present the most commonly used cutoff for low muscle mass in women is an SMI of 38.5 [23, 25]. Current SMI cutoffs were derived using a conventional height scaling power of 2, and new cutoffs based on the optimal scaling power derived in the current study will need to be defined.
The present study results align with the findings of smaller studies. Brown et al. [31] studied single-slice axial CT images at the L3 level in 2036 older patients (mean age, 64 ± 11 years) who had colorectal cancer, and they also found that SMA scaled with height by powers less than 2 and that these powers differed between men (mean, 1.06 ± 0.12) and women (mean, 0.80 ± 0.12). The allometric analysis by Brown et al. included a correction factor for age. Derstine et al. [32], in a study of CT examinations of 1849 healthy adults who were kidney donor candidates (mean age, approximately 31 years), observed that SMA at L3 scales with height to a power of 1.02 for men and 1.33 for women. In their study, the allometric analysis was performed only in adults under 40 years old aand did not include a correction factor for age. In addition, Derstine et al. observed that the ratio of SMA to height using a height scaling power of 1, had a negative association with age in patients older than 40 years old and no significant association with age in patients under 40 years old. In the current study's allometric analysis, in patients older than 40 years old, age likewise showed a negative coefficient that was significantly different from 0, indicating decreases in SMA as age increased within this patient group.
The relevance of this study is heightened by the current impetus to integrate CT-based muscle metrics into imaging workflows to automate the diagnosis of sarcopenia. This vision of wider implementation of opportunistic imaging diagnosis has gained feasibility from advances in artificial intelligence tools and the availability of robust population reference standards for CT muscle metrics [2730]. Clinical adoption of these methods requires optimal validated biomarker definitions that facilitate effective patient risk stratification. The findings regarding the use of SMI for prediction of all-cause mortality further support the potential clinical utility of this muscle metric. SMI, as an imaging marker of muscle quantity, could help identify individuals at risk for various adverse outcomes and prompt earlier interventions, such as patient education, nutritionist consultation (e.g., for optimization of dietary protein and nutritional supplementation), and physical therapy (e.g., high-intensity resistance exercise, blood flow restriction training) [1219]. Progressive resistance training can also potentially reverse sarcopenia, including in older individuals [42].
Age adjustment is a deviation from traditional allometric analysis and addresses the known physiologic loss of muscle mass that occurs in otherwise healthy older adults. The method of age adjustment influences the estimated optimal height scaling power for SMA, and different recommendations for height scaling power may be expected in younger versus older patient groups. In theory, a recommended normative height scaling power for SMA would be derived from a healthy young adult population, analogous to the use of T scores when fracture risk is estimated from DEXA. In this study, allometric analysis of patients 40 years old and younger, with no age correction was applied, suggested optimal scaling powers close to 1 in both men and women, contributing to the recommendation of an optimal height scaling power of 1 for defining SMI.
Although this study examined axial CT images, analogous considerations could apply to normalization of similar metrics of sarcopenia based on 2D cross-sectional images from MRI or ultrasound. Those modalities are both increasingly used for muscle evaluation [43, 44]. In contrast, 3D or volumetric metrics, such as BMI and DEXA-derived ALM, have been shown to scale with height to a power of approximately 2.
This study had limitations. First, the optimal height scaling power for SMA will vary by anatomic location; this analysis measured SMA at the L3 level, a level that is now commonly used to assess for sarcopenia. Second, the indications for the CT examinations were not considered, and it is possible that patients had medical conditions not captured by the screened ICD-9 and ICD-10 codes that also affect muscle health. Third, the analysis in the test set was not stratified by individual major medical conditions. Fourth, the derived optimal height scaling power was tested internally in examinations from the same center that was used for deriving the optimal scaling power; external validation was not performed. Fifth, reasons for patient death in the test set were not identified. Sixth, several of the reviewed automated segmentations were qualitatively judged to have underestimated true SMA; these segmentations were not excluded from further analyses. Last, this study did not examine various thresholds for defining sarcopenia based on SMA or SMI.
In conclusion, the results of this large single-center study support an optimal height scaling power of 1 for SMA at the L3 level in both men and women. The findings call for a reconsideration and possible revision of the current convention of using a height scaling power of 2 in defining SMI, a standardized metric of muscle quantity for purposes of sarcopenia diagnosis.

Footnotes

Provenance and review: Not solicited; externally peer reviewed.
Peer reviewers: All reviewers chose not to disclose their identities.

Supplemental Content

File (23_29889_suppl.pdf)

References

1.
Cruz-Jentoft AJ, Bahat G, Bauer J, et al.; Writing Group for the European Working Group on Sarcopenia in Older People 2 (EWGSOP2); the Extended Group for EWGSOP2. Sarcopenia: revised European consensus on definition and diagnosis. Age Ageing 2019; 48:16–31
2.
Beaudart C, Demonceau C, Reginster JY, et al. Sarcopenia and health-related quality of life: a systematic review and meta-analysis. J Cachexia Sarcopenia Muscle 2023; 14:1228–1243
3.
Jiang T, Lin T, Shu X, et al. Prevalence and prognostic value of preexisting sarcopenia in patients with mechanical ventilation: a systematic review and meta-analysis. Crit Care 2022; 26:140
4.
Jones K, Gordon-Weeks A, Coleman C, Silva M. Radiologically determined sarcopenia predicts morbidity and mortality following abdominal surgery: a systematic review and meta-analysis. World J Surg 2017; 41:2266–2279
5.
Couderc AL, Liuu E, Boudou-Rouquette P, et al. Pre-therapeutic sarcopenia among cancer patients: an up-to-date meta-analysis of prevalence and predictive value during cancer treatment. Nutrients 2023; 15:1193
6.
Zhou HH, Liao Y, Peng Z, Liu F, Wang Q, Yang W. Association of muscle wasting with mortality risk among adults: a systematic review and meta-analysis of prospective studies. J Cachexia Sarcopenia Muscle 2023; 14:1596–1612
7.
Westbury LD, Beaudart C, Bruyère O, et al.; International Musculoskeletal Ageing Network. Recent sarcopenia definitions: prevalence, agreement and mortality associations among men—findings from population-based cohorts. J Cachexia Sarcopenia Muscle 2023; 14:565–575
8.
Xu J, Wan CS, Ktoris K, Reijnierse EM, Maier AB. Sarcopenia is associated with mortality in adults: a systematic review and meta-analysis. Gerontology 2022; 68:361–376
9.
Somasundaram E, Castiglione JA, Brady SL, Trout AT. Defining normal ranges of skeletal muscle area and skeletal muscle index in children on CT using an automated deep learning pipeline: implications for sarcopenia diagnosis. AJR 2022; 219:326–336
10.
Kudo W, Terui K, Hattori S, et al. Establishment and validation of reference values for abdominal skeletal muscle compartments in children. Clin Nutr 2023; 42:653–660
11.
Montero-Errasquín B, Cruz-Jentoft AJ. Acute sarcopenia. Gerontology 2023; 69:519–525
12.
Negm AM, Lee J, Hamidian R, Jones CA, Khadaroo RG. Management of sarcopenia: a network meta-analysis of randomized controlled trials. J Am Med Dir Assoc 2022; 23:707–714
13.
Shen Y, Shi Q, Nong K, et al. Exercise for sarcopenia in older people: a systematic review and network meta-analysis. J Cachexia Sarcopenia Muscle 2023; 14:1199–1211
14.
Shi Y, Tang Y, Stanmore E, McGarrigle L, Todd C. Non-pharmacological interventions for community-dwelling older adults with possible sarcopenia or sarcopenia: a scoping review. Arch Gerontol Geriatr 2023; 112:105022
15.
Chen LK, Arai H, Assantachai P, et al. Roles of nutrition in muscle health of community-dwelling older adults: evidence-based expert consensus from Asian Working Group for Sarcopenia. J Cachexia Sarcopenia Muscle 2022; 13:1653–1672
16.
Gielen E, Beckwée D, Delaere A, De Breucker S, Vandewoude M, Bautmans I; Sarcopenia Guidelines Development Group of the Belgian Society of Gerontology and Geriatrics (BSGG). Nutritional interventions to improve muscle mass, muscle strength, and physical performance in older people: an umbrella review of systematic reviews and meta-analyses. Nutr Rev 2021; 79:121–147
17.
Beckwée D, Delaere A, Aelbrecht S, et al. Exercise interventions for the prevention and treatment of sarcopenia: a systematic umbrella review. J Nutr Health Aging 2019; 23:494–502
18.
De Spiegeleer A, Beckwée D, Bautmans I, Petrovic M; Sarcopenia Guidelines Development Group of the Belgian Society of Gerontology and Geriatrics (BSGG). Pharmacological interventions to improve muscle mass, muscle strength and physical performance in older people: an umbrella review of systematic reviews and meta-analyses. Drugs Aging 2018; 35:719–734
19.
Hernández-Lepe MA, Miranda-Gil MI, Valbuena-Gregorio E, Olivas-Aguirre FJ. Exercise programs combined with diet supplementation improve body composition and physical function in older adults with sarcopenia: a systematic review. Nutrients 2023; 15:1998
20.
Pickhardt PJ, Correale L, Hassan C. AI-based opportunistic CT screening of incidental cardiovascular disease, osteoporosis, and sarcopenia: cost-effectiveness analysis. Abdom Radiol (NY) 2023; 48:1181–1198
21.
Darvishi A, Shafiee G, Balajam NZ, Hemami MR, Ostovar N, Heshmat R. Cost-effectiveness analysis of sarcopenia management interventions in Iran. BMC Public Health 2023; 23:819
22.
Boutin RD, Lenchik L. Value-added opportunistic CT: insights into osteoporosis and sarcopenia. AJR 2020; 215:582–594
23.
Amini B, Boyle SP, Boutin RD, Lenchik L. Approaches to assessment of muscle mass and myosteatosis on computed tomography: a systematic review. J Gerontol A Biol Sci Med Sci 2019; 74:1671–1678
24.
Hopkins JJ, Skubleny D, Bigam DL, Baracos VE, Eurich DT, Sawyer MB. Barriers to the interpretation of body composition in colorectal cancer: a review of the methodological inconsistency and complexity of the CT-defined body habitus. Ann Surg Oncol 2018; 25:1381–1394
25.
Prado CM, Lieffers JR, McCargar LJ, et al. Prevalence and clinical implications of sarcopenic obesity in patients with solid tumours of the respiratory and gastrointestinal tracts: a population-based study. Lancet Oncol 2008; 9:629–635
26.
Martin L, Birdsell L, Macdonald N, et al. Cancer cachexia in the age of obesity: skeletal muscle depletion is a powerful prognostic factor, independent of body mass index. J Clin Oncol 2013; 31:1539–1547
27.
van der Werf A, Langius JAE, de van der Schueren MAE, et al. Percentiles for skeletal muscle index, area and radiation attenuation based on computed tomography imaging in a healthy Caucasian population. Eur J Clin Nutr 2018; 72:288–296
28.
Derstine BA, Holcombe SA, Ross BE, Wang NC, Su GL, Wang SC. Skeletal muscle cutoff values for sarcopenia diagnosis using T10 to L5 measurements in a healthy US population. Sci Rep 2018; 8:11369
29.
Kim EH, Kim KW, Shin Y, et al. Reference data and T-scores of lumbar skeletal muscle area and its skeletal muscle indices measured by CT scan in a healthy Korean population. J Gerontol A Biol Sci Med Sci 2021; 76:265–271
30.
Kong M, Geng N, Zhou Y, et al. Defining reference values for low skeletal muscle index at the L3 vertebra level based on computed tomography in healthy adults: a multicentre study. Clin Nutr 2022; 41:396–404
31.
Brown JC, Heymsfield SB, Caan BJ. Scaling of computed tomography body composition to height: relevance of height-normalized indices in patients with colorectal cancer. J Cachexia Sarcopenia Muscle 2022; 13:203–209
32.
Derstine BA, Holcombe SA, Ross BE, Wang NC, Su GL, Wang SC. Optimal body size adjustment of L3 CT skeletal muscle area for sarcopenia assessment. Sci Rep 2021; 11:279
33.
Heymsfield SB, Heo M, Thomas D, Pietrobelli A. Scaling of body composition to height: relevance to height-normalized indexes. Am J Clin Nutr 2011; 93:736–740
34.
Lortie J, Gage G, Rush B, Heymsfield SB, Szczykutowicz TP, Kuchnia AJ. The effect of computed tomography parameters on sarcopenia and myosteatosis assessment: a scoping review. J Cachexia Sarcopenia Muscle 2022; 13:2807–2819
35.
Morsbach F, Zhang YH, Martin L, Lindqvist C, Brismar T. Body composition evaluation with computed tomography: contrast media and slice thickness cause methodological errors. Nutrition 2019; 59:50–55
36.
Fuchs G, Chretien YR, Mario J, et al. Quantifying the effect of slice thickness, intravenous contrast and tube current on muscle segmentation: implications for body composition analysis. Eur Radiol 2018; 28:2455–2463
37.
Kanavati F, Islam S, Arain Z, Aboagye E, Rockall A. Fully-automated deep learning slice-based muscle estimation from CT images for sarcopenia assessment. arXiv website. arxiv.org/abs/2006.06432. Published Jun 10 2020. Accessed Jul 1, 2023
38.
Blankemeier L, Desai A, Chaves JM, et al. Comp2Comp: open-source body composition assessment on computed tomography. arXiv website. arxiv. org/abs/2302.06568. Published Feb 23, 2023. Accessed Jul 1, 2023
39.
Ronneberger O, Fischer P, Brox T. U-net: convolutional networks for biomedical image segmentation. In: Navab N, Hornegger J, Wells WM, Frangi AE, eds. Medical image computing and computer-assisted intervention: MICCAI 2015. Springer, 2015; 18:234–241
40.
Moon SW, Lee SH, Woo A, et al. Reference values of skeletal muscle area for diagnosis of sarcopenia using chest computed tomography in Asian general population. J Cachexia Sarcopenia Muscle 2022; 13:955–965
41.
Baumgartner RN, Koehler KM, Gallagher D, et al. Epidemiology of sarcopenia among the elderly in New Mexico. Am J Epidemiol 1998; 147:755–763
42.
Mende E, Moeinnia N, Schaller N, et al. Progressive machine-based resistance training for prevention and treatment of sarcopenia in the oldest old: a systematic review and meta-analysis. Exp Gerontol 2022; 163:111767
43.
Boutin RD, Houston DK, Chaudhari AS, Willis MH, Fausett CL, Lenchik L. Imaging of sarcopenia. Radiol Clin North Am 2022; 60:575–582
44.
Lenchik L, Mazzoli V, Cawthon PM, Hepple RT, Boutin RD. Muscle steatosis and fibrosis in older adults, from the AJR special series on imaging of fibrosis. AJR 2024; 222:e2329742

Information & Authors

Information

Published In

American Journal of Roentgenology
PubMed: 37877596

History

Submitted: July 2, 2023
Revision requested: July 19, 2023
Revision received: September 6, 2023
Accepted: October 13, 2023
Version of record online: October 25, 2023

Keywords

  1. artificial intelligence
  2. CT
  3. mortality risk prognostication
  4. sarcopenia
  5. skeletal muscle area

Authors

Affiliations

Louis Blankemeier, MS
Department of Electrical Engineering, Stanford University, Stanford, CA.
Lawrence Yao, MD
Radiology and Imaging Sciences, NIH Clinical Center, Bethesda, MD.
Jin Long, PhD
Center for Artificial Intelligence in Medicine & Imaging, Stanford University, Palo Alto, CA.
Eduardo P. Reis, MD
Department of Radiology, Stanford University School of Medicine, 300 Pasteur Dr, MC-5105, Stanford, CA 94305.
Leon Lenchik, MD
Department of Radiology, Wake Forest University School of Medicine, Winston-Salem, NC.
Akshay S. Chaudhari, PhD
Department of Radiology and of Biomedical Data Science, Stanford University School of Medicine, Stanford, CA.
Robert D. Boutin, MD [email protected]
Department of Radiology, Stanford University School of Medicine, 300 Pasteur Dr, MC-5105, Stanford, CA 94305.

Notes

Address correspondence to R. D. Boutin ([email protected]).
First published online: Oct 25, 2023
Version of record: Jan 10, 2024
A. S. Chaudhari receives research support from GE Healthcare, Philips, Amazon, Microsoft, Google, and Stability.ai; has provided consulting services to Subtle Medical; and serves on the scientific advisory board of Chondrometrics GmbH and Brain Key. R. D. Boutin receives research support from GE Healthcare. The remaining authors declare that there are no other disclosures relevant to the subject matter of this article.

Funding Information

Supported by NIH grants R01 AR077604, R01 EB002524, R01 AR079431, K24 AR062068, and P41 EB027060; the Stanford Precision Health and Integrated Diagnostics Seed Grant; and the Stanford Human-Centered AI and Artificial Intelligence in Medicine and Image Seed Grant.

Metrics & Citations

Metrics

Citations

Export Citations

To download the citation to this article, select your reference manager software.

Articles citing this article

Media

Figures

Other

Tables

Share

Share

Copy the content Link

Share on social media