Opportunistic Measurement of Skeletal Muscle Size and Muscle Attenuation on Computed Tomography Predicts 1-Year Mortality in Medicare Patients

Abstract

Background

Opportunistic assessment of sarcopenia on CT examinations is becoming increasingly common. This study aimed to determine relationships between CT-measured skeletal muscle size and attenuation with 1-year risk of mortality in older adults enrolled in a Medicare Shared Savings Program (MSSP).

Methods

Relationships between skeletal muscle metrics and all-cause mortality were determined in 436 participants (52% women, mean age 75 years) who had abdominopelvic CT examinations. On CT images, skeletal muscles were segmented at the level of L3 using two methods: (a) all muscles with a threshold of −29 to +150 Hounsfield units (HU), using a dedicated segmentation software, (b) left psoas muscle using a free-hand region of interest tool on a clinical workstation. Muscle cross-sectional area (CSA) and muscle attenuation were measured. Cox regression models were fit to determine the associations between muscle metrics and mortality, adjusting for age, sex, race, smoking status, cancer diagnosis, and Charlson comorbidity index.

Results

Within 1 year of follow-up, 20.6% (90/436) participants died. In the fully-adjusted model, higher muscle index and muscle attenuation were associated with lower risk of mortality. A one-unit standard deviation (SD) increase was associated with a HR = 0.69 (95% CI = 0.49, 0.96; p = .03) for total muscle index, HR = 0.67 (95% CI = 0.49, 0.90; p < .01) for psoas muscle index, HR = 0.54 (95% CI = 0.40, 0.74; p < .01) for total muscle attenuation, and HR = 0.79 (95% CI = 0.66, 0.95; p = .01) for psoas muscle attenuation.

Conclusion

In older adults, higher skeletal muscle index and muscle attenuation on abdominopelvic CT examinations were associated with better survival, after adjusting for multiple risk factors.

Sarcopenia is a common medical condition defined by decreased muscle mass and muscle function (1). Although usually diagnosed using a combination of functional measures and dual x-ray absorptiometry (DXA)-derived appendicular lean mass, sarcopenia can also be evaluated using computed tomography (CT) (1–4). Unlike DXA, which only measures muscle quantity, CT measures muscle quantity (eg, muscle cross-sectional area [CSA] or muscle index) and muscle quality [eg, muscle attenuation or radiodensity (3,4)]. Lower CT attenuation of muscle is usually due to increased deposition of adipose tissue in muscle and the accumulation of lipid within muscle, known as “myosteotosis” (1). There is increasing interest in developing clinically relevant approaches to the evaluation of sarcopenia using CT, but more data is needed on the relationship between CT-derived muscle metrics and relevant health outcomes, including mortality.

In many high-risk populations, CT-derived muscle metrics have been shown to predict mortality (5–16). Excess mortality has been variably associated with CT measures of skeletal muscle index and muscle attenuation (5–16). Most of these studies have used abdominopelvic CTs to measure either all of the visualized muscles at the level of L3 or just the psoas muscles (5–16). No prior study has investigated these CT metrics in an opportunistic setting, without selecting a study cohort already associated with high mortality.

Patients enrolled in a Medicare Shared Savings Plan (MSSP) serve as a relevant cohort in which to evaluate the opportunistic screening for sarcopenia using CT. The MSSP patients are enrolled in a value-based contract where Medicare adjusts reimbursement depending on metrics that evaluate how well the patients are managed. Ideally, opportunistic CT measurements of muscle would be combined with the electronic health record (EHR) to help determine patient prognosis. The purpose of our study was to determine if routine CT examinations in patients from the MSSP could be used to predict 1-year mortality, after adjustment for common covariates known to impact mortality readily available in the EHR. Our hypothesis was that higher muscle quantity and quality would be associated with improved survival.

Materials and Methods

This retrospective cohort study was conducted among the MSSP-ACO population at Wake Forest Baptist Medical Center. “Further information about MSSP can be found on: https://www.cms.gov/Medicare/Medicare-Fee-for-Service-Payment/sharedsavingsprogram/about.html. As of December 1, 2016, there were 12,798 individuals attributed to our health system’s MSSP-ACO population. For this study, inclusion criteria were: age 65 years or older, North Carolina Residency, and the presence of a noncontrast abdominopelvic CT examination (CPT code: 74176) within the span of 6 months (October 1, 2014 to April 1, 2015). The follow-up period for mortality was defined as 365 days after the date of CT. Cases were identified, and data were extracted from Clarity (EPIC electronic data warehouse). The study was approved by the Wake Forest Institutional Review Board.

CT Evaluation

Various abdominopelvic multidetector CT protocols were used. Acquisition parameters were: 100–140 kVp, variable mAs (depending on patient size), 2.5–3.00 mm detector collimation. CT scans were acquired using seven different General Electric and Siemens systems.

On CT images, skeletal muscles were segmented at the level of L3 vertebra using two methods. The first method used dedicated segmentation software (Mimics version 19.0; Materialise, Leuven, Belgium). On a single CT image, all visualized muscles with a threshold of −29 to +150 HU were segmented, including rectus abdominis, internal and external obliques, transversus abdominis, quadratus lumborum, psoas major and minor, erector spinae, and latissimus dorsi muscles, as previously validated (10). The second method used a clinical Picture Archiving and Communication System (PACS) workstation (iSite version 3.6, Philips Healthcare). On a single CT image, left psoas muscle was segmented using a free-hand region of interest tool, without applying muscle thresholds, as previously validated (17).

Using both segmentation methods, muscle CSA and muscle attenuation were recorded. Muscle index variables were derived from CSA (cm²) scaled to patient height (m²). Height was not statistically associated with mortality and did not act as a confounder for index predictors.

The CT-based muscle analysis was standardized by following the Standard Operating Procedures manual (17). All CT images with measurements were archived and validated for measurement accuracy by two musculoskeletal radiologists (LL and RB), blinded to patient outcomes. CT images with asymmetry between psoas muscles (due to scoliosis, degenerative diseases, or prior surgery) as well as scans with other internal or external artifacts were excluded (n = 35). Using the same methodology, our interclass correlation coefficients were 0.84–0.9 (17,18).

Mortality

Mortality data from the EHR was supplemented with the North Carolina State Center for Health Statistics death index, which is deterministically matched to data from EHR on a monthly basis. For patients confirmed as deceased, length of follow-up was determined in days from date of initial CT visit to date of death. All other participants were right-censored at 365 days post CT visit.

Statistical Analysis

For demographic and outcome measures, summary statistics were determined as counts and percentages for categorical variables and as medians and interquartile ranges for continuous variables. Four predictor variables were considered: (a) total muscle index, (b) total muscle attenuation, (c) psoas muscle index, and (d) psoas muscle attenuation. Correlations between the four predictor variables were analyzed by Pearson’s correlation.

Cox proportional hazard regression models were fit to the CT metrics with adjustment for covariates known to be associated with mortality including age, sex (female vs male), race (Caucasian vs African American or Other), smoking history, diagnosis of cancer, and Charlson Comorbidity Index (CCI) (categorized: 0, 1–4, and ≥5) in the EHR. The diagnosis of cancer and CCI were determined by International Classification of Diseases (ICD-9) codes before or on the date of the CT. ICD-9 codes for cancer were 140.x-209.3x, 239.x, and V10.x.

Splines with three knots were fit to the continuous variable, age, to relax normality assumptions. Three models were used to examine the relationships between CT muscle metrics and mortality: (a) unadjusted, (b) adjusted for age, sex, and race, and (c) adjusted for age, sex, race, smoking, cancer, and CCI.

The variance inflation factor was used to assess multicollinearity. For all models with covariates, the variance inflation factor was <2 for all variables indicating that multicollinearity was not an issue. Martingale residuals and Shoenfeld residuals were examined to confirm linear relationships and to verify the proportional hazard assumption, respectively. No nonlinear relationships were apparent on examination of Martingale residual plots, and Shoenfeld residual plots with respective chi-square test indicated that the global models and each of its components met the assumption of proportional hazards.

Hazard ratios with 95% confidence intervals were calculated for continuous variables by comparing the risk of mortality per standard deviation of each muscle metric. For each muscle metric, Kaplan–Meier survival plots were plotted and stratified into standardized score categories: Z ≤ −1; Z > −1 and Z ≤ 1; Z > 1.

Data analyses were performed using R version 3.4.3.

Results

Five hundred eleven participants in the MSSP had a noncontrast abdominopelvic CT examination within the 6-month timeframe. Of these, 40 were excluded due to out-of-state or indeterminate residency status and 35 were excluded due to suboptimal CT image quality. The remaining 436 participants met the study inclusion criteria.

Table 1 presents demographic characteristics, muscle metrics, and survival data. The mean age was 75 years (SD: 7.8). The study cohort included 51.6% women and 24.8% African Americans and other race categories. After 1 year of follow-up, 90 of 436 (20.6%) patients died. Compared with the survivor group, the deceased group had a significantly higher CCI and a larger proportion of cancer diagnoses. Compared with the survivor group, the deceased group had a significantly lower muscle index and muscle attenuation.

Table 1.

Participant Characteristics

Characteristic	All (n = 436)	Survivors (n = 346)	Deceased (n = 90)	p-Value
Age (years)	75.8 (7.8)	75.3 (7.5)	78 (8.5)	.006
Female	225 (51.6)	179 (51.7)	46 (51.1)	.916
BMI (kg/m2)	28 (6.1)	28.3 (6.1)	26.9 (6.3)	.063
Black/African American or Other	108 (24.8)	86 (24.9)	22 (24.4)	.936
Height (m2)	2.9 (0.4)	2.9 (0.4)	2.8 (0.4)	.897
Caucasian	328 (75.2)	260 (75.1)	68 (75.6)	.936
Past or current smoker	58 (13.3)	45 (13.0)	13 (14.4)	.720
Cancer diagnosis	217 (49.8)	162 (46.8)	55 (61.1)	.028
CCI	4.5 (3)	4.2 (3)	5.6 (3)	<.001
Total muscle index (cm2/m2)	43 (10.8)	43.6 (10.8)	40.4 (10.5)	.011
Psoas muscle index (cm2/m2)	2.7 (0.9)	2.8 (0.8)	2.5 (0.9)	.004
Total muscle attenuation (HU)	25 (8.6)	25.8 (8.7)	21.9 (7.6)	<.001
Psoas muscle attenuation (HU)	32 (11.7)	33 (11.2)	28.4 (12.9)	.003

Mean, standard deviations (SD), t-test reported for continuous variables; Number, percent (%), and chi-squared test reported for categorical variables. BMI = body mass index; CCI = Charlson Comorbidity Index; HU = Hounsfield units.

Muscle index measures were moderately correlated (r = .52, p < .01) as were muscle attenuation measures (r = .64, p < .01). In contrast, only weak correlation was present between total muscle index and total muscle attenuation (r = .18, p < .01) as well as between psoas muscle index and psoas muscle attenuation (r = .14, p < . 01).

Relationships between muscle indices and mortality within 1 year are shown in Table 2. Increased muscle index was associated with decreased mortality in the models adjusted for age, sex, race, smoking status, cancer diagnosis, and CCI. A one-unit SD increase in total muscle index was associated with a 31% decrease in risk of mortality. A one-unit SD increase in psoas muscle index was associated with a 33% decrease in risk of mortality.

Table 2.

Association Between Muscle Index and All-Cause Mortality

CT Measurement	HR Per SD (95% CI)	p-Value
Total muscle index
Unadjusted	0.687 (0.515–0.915)	.0103
Adjusted for adjusted for age, sex, and race	0.693 (0.498–0.965)	.0299
Adjusted for age, sex, race, past or current smoking status, cancer diagnosis, and CCI	0.685 (0.491–0.956)	.0259
Psoas muscle index
Unadjusted	0.642 (0.483–0.853)	.0023
Adjusted for adjusted for age, sex, and race	0.645 (0.475–0.875)	.0048
Adjusted for age, sex, race, past or current smoking status, cancer diagnosis, and CCI	0.665 (0.492–0.898)	.0079

CCI = Charlson Comorbidity Index; CI = confidence interval; HR = hazard ratio; SD = standard deviation.

Relationships between muscle attenuation and 1-year mortality are shown in Table 3. Increased muscle attenuation was associated with decreased mortality in the models adjusted for age, sex, race, current smoking, cancer diagnosis, and CCI. A one-unit SD increase in total muscle attenuation was associated with a 46% decrease in mortality. A one-unit SD increase in psoas muscle attenuation was associated with a 21% decrease in mortality.

Table 3.

Association Between Muscle Attenuation and All-Cause Mortality

CT Measurement	HR Per SD (95% CI)	p-Value
Total muscle attenuation
Unadjusted	0.576 (0.436–0.762)	.0001
Adjusted for adjusted for age, sex, and race	0.549 (0.406–0.742)	.0001
Adjusted for age, sex, race, past or current smoking status, cancer diagnosis, and CCI	0.542 (0.396–0.742)	.0001
Psoas muscle attenuation
Unadjusted	0.750 (0.629–0.895)	.0014
Adjusted for adjusted for age, sex, and race	0.765 (0.637–0.918)	.0040
Adjusted for age, sex, race, past or current smoking status, cancer diagnosis, and CCI	0.791 (0.658–0.952)	.0131

CCI = Charlson Comorbidity Index; CI = confidence interval; HR = hazard ratio; SD = standard deviation.

Table 4 presents muscle metrics stratified into standardized Z-score categories, based on the cohort median muscle metrics. Kaplan–Meier survival analysis for total muscle index and psoas muscle index are shown in Figures 1 and 2. Comparing patients at least 1 SD below the cohort median muscle index, the patients with more muscle (higher muscle index) had more favorable overall survival at each time point.

Table 4.

Muscle Metrics Stratified into Standardized Z-Score Categories

CT Measurement	Median	Z ≤ −1	−1 < Z ≤ 1	Z > 1
Total muscle index (cm2/m2)	41.0	20.23 to 30.33	30.39 to 51.90	51.93 to 99.91
Psoas muscle index (cm2/m2)	2.6	0.80 to 1.76	1.79 to 3.50	3.51 to 6.91
Total muscle attenuation (HU)	25.6	−1.62 to 17.02	17.04 to 34.24	34.30 to 49.51
Psoas muscle attenuation (HU)	34.0	−46.07 to 22.00	22.38 to 45.67	45.76 to 56.88

Figure 1.

Kaplan–Meier plots for patients stratified by total muscle index. Patient survival is compared among three groups, ranging from at least 1 SD below cohort median (Z ≤ −1) to more than 1 SD above cohort median (Z > 1).

Figure 2.

Kaplan–Meier plots for patients stratified by psoas muscle index. Patient survival is compared among three groups, ranging from at least 1 SD below cohort median (Z ≤ −1) to more than 1 SD above cohort median (Z > 1).

Kaplan–Meier survival analysis for total muscle attenuation and psoas muscle attenuation are shown in Figures 3 and 4. Comparing patients at least 1 SD below the cohort median muscle attenuation, the patients with higher muscle attenuation (lower myosteotosis) had more favorable overall survival at each time point.

Figure 3.

Kaplan–Meier plots for patients stratified by total muscle attenuation. Patient survival is compared among three groups, ranging from at least 1 SD below cohort median (Z ≤ −1) to more than 1 SD above cohort median (Z > 1).

Figure 4.

Kaplan–Meier plots for patients stratified by psoas muscle attenuation. Patient survival is compared among three groups, ranging from at least 1 SD below cohort median (Z ≤ −1) to more than 1 SD above cohort median (Z > 1).

Discussion

While the majority of prior CT studies of sarcopenia and mortality have focused on high-risk populations, this is the first truly opportunistic study of older adults, showing a robust association of CT muscle metrics with all-cause mortality.

CT examinations of the abdomen and pelvis are commonly used to evaluate a wide range of medical conditions. In addition to providing data related to the specific reasons that the CT examinations were obtained, the CT images can be “opportunistically” evaluated to provide quantitative measurements related to a number of chronic conditions, including sarcopenia (19). In the era of increasing use of informatics in clinical medicine, these secondary analyses of CT images can be easily linked to other data elements in the EHR. The goal of this study was to assess the relationship between CT-derived quantitative muscle metrics and the 1-year risk of mortality in Medicare patients to determine whether this information might provide additional prognostic information.

The novel finding in this study of older adults is that measures of sarcopenia (eg, lower muscle index and muscle attenuation) on abdominopelvic CT were associated with increased risk of all-cause mortality, thereby confirming our initial hypothesis. Importantly, the association of muscle metrics with mortality persisted in all of our models, even after adjusting for the diagnosis of cancer and CCI.

The majority of prior studies investigating the relationship of CT-derived muscle metrics and mortality have been in oncology patients, where abdominopelvic CTs were for tumor diagnosis, staging, or surgical planning (6,7). In a meta-analysis of 38 studies containing 7,843 participants with nonhematologic solid tumors, Shachar and colleagues (7) reported that a higher CT-derived skeletal muscle index was associated with improved overall survival (HR = 0.69; 95% CI = 0.64–0.72; p < .001). Because cancer patients have a higher prevalence of muscle depletion than community-dwelling older adults (20,21), generalizing from studies of cancer patients warrants caution, which is why we adjusted our analyses for cancer diagnosis.

In nononcology patients, CT muscle metrics have been shown to predict mortality in observational cohorts of patients with medical conditions (eg, pneumonia, sepsis, chronic liver disease, chronic renal disease, diabetes, trauma, intensive care unit admissions) and undergoing surgery [general, vascular, transplant, and orthopedic surgery (9–15,22–26)]. In a study of 408 trauma patients, aged 65 years and older, admitted to the intensive care unit, CT-measured sarcopenia (total muscle at L3 level) was associated with increased 1-year mortality [HR = 10.3; 95% CI = 1.3–78.8; p = .03 (9)]. Like studies of oncology patients, these studies were conducted in higher-risk hospitalized patients rather than using opportunistic screening of both outpatients and inpatients.

One of the obstacles to wider use of CT for sarcopenia screening has been that the approach for measuring sarcopenia on CT is not standardized. Although this is certainly the case, two approaches are dominant in the CT literature. The most common approach uses specialized software to threshold muscle (usually at −29 to +150 HU) and evaluates total skeletal muscle at the L3 level. The second most common approach uses more widely available clinical PACS software to measure psoas muscle only, without the use of thresholding. The second approach is faster, more convenient, and does not require additional cost of post-processing with specialized software on a separate workstation. Importantly, our study uses both of these approaches, shows moderate correlation between the muscle metrics derived from the two approaches, and shows that muscle metrics from both approaches predict mortality.

Another inconsistency in the CT literature on sarcopenia is regarding the use of muscle metrics related to muscle size (CSA or index) or muscle attenuation (radiodensity). Whereas muscle size measured with CT is associated with lean mass measured by DXA, muscle attenuation on CT is a measure of muscle quality (myosteotosis), not captured by DXA. Significantly, our study uses both of these metrics and shows that each is able to predict mortality. Even after adjusting for muscle attenuation, muscle size indices remain predictive of mortality.

Many mechanisms have been proposed to explain increased mortality in older adults with sarcopenia. Decreased physical activity, poor nutrition, and metabolic abnormalities all play a role (27,28). In older adults, a decline in muscle function corresponds to increased fatty infiltration of muscle [lower attenuation on CT (27,28)]. In patients with cancer, excess mortality is associated with decrease in muscle mass [lower muscle index on CT (29,30)]. Because age-related and cancer-induced sarcopenia may have different pathogenic mechanisms, it is important to determine the relationship of CT muscle metrics and mortality not just in cancer patients, but also in community-dwelling individuals without cancer.

Clinical Relevance

If future studies confirm our results and show that CT-derived muscle measurements are independently associated with mortality, institutions could take advantage of opportunistic screening to identify at-risk patients. At many medical centers, teams of care managers are available to assist with the coordination of care. Based on opportunistic CT screening, at-risk patients could be prioritized for care management referrals.

Future research should examine the association CT muscle metrics with additional health outcomes such as hospital readmissions and explore how the incorporation of these measures into prediction models improve stratification of risk. Research should also examine the generalizability of these findings at other institutions. Similar to CT, there may also be value of opportunistically assessing muscle metrics on MRI exams obtained during routine patient care (31,32).

Strengths and Limitations

Our study has several limitations. We did not collect physical function measurements or lean mass measurements using DXA as these are not commonly available in the electronic health record. This is a common limitation of all studies that use “opportunistic” CT. We did not use diagnostic cut-points to diagnose sarcopenia on CT and instead used muscle metrics as a continuous variable. This is because any diagnostic cut-point based on CT attenuation would need to be adjusted for specific muscle groups (different for total muscle compared with psoas muscle). Finally, owing to sample size considerations, we did not differentiate between causes of mortality.

Our study also has several important strengths. This is one of the largest studies using opportunistic CT measures of muscle in older adults. The CT examinations were acquired using CT scanners from different manufacturers and using different scanner protocols, likely generalizable to other health care systems. The methodology appears suitable for future pragmatic trials.

Conclusion

This is the first study of CT muscle metrics that evaluated all adults aged more than 65 years who undergo a noncontrast abdominopelvic CT, rather than limiting the study group to high-risk cancer, surgery, or trauma patients. Our results lay the groundwork for large-scale case finding screening strategy for sarcopenia that does not require additional examination cost or radiation exposure to patients.

Funding

This work was supported by the National Institutes of Health Pepper Center (P30 AG021332) and Clinical and Translational Science Institute (UL1TR001420).

Conflict of Interest

None reported.