Walking Speed and Muscle Mass Estimated by the D₃-Creatine Dilution Method Are Important Components of Sarcopenia Associated With Incident Mobility Disability in Older Men: A Classification and Regression Tree Analysis

Abstract

Objectives:

It is unknown whether muscle mass measured by the D₃-creatine dilution method is a superior predictor of incident mobility disability than traditional components of sarcopenia definitions (including grip strength, walking speed, appendicular lean mass). The objective of this study was to determine the relative importance of strength; physical performance; and lean, fat, and muscle mass in predicting incident mobility disability in older men.

Design:

Longitudinal cohort study of participants in the Osteoporotic Fractures in Men (MrOS) study.

Setting and Participants:

Muscle mass was assessed by D₃-creatine dilution in 1098 men (aged 83.7 ± 3.7 years). Participants also completed anthropomorphic measures, 6-m walking speed (m/s), grip strength (kg), chair stands (seconds), and dual x-ray absorptiometry appendicular lean mass (ALM), and total body fat percentage. Men self-reported incident mobility disability defined by the development of an inability to complete at least one of walking 2–3 blocks, climbing 10 steps, or carrying 10 lb over 2.2 ± 0.3 years.

Methods:

Classification and regression tree analysis was conducted to determine relative variable importance and algorithm cutpoints for predicting incident mobility disability. Given the proximity of walking speed with the primary outcome (mobility disability), analyses were conducted with and without walking speed.

Results:

Walking speed followed by D₃Cr muscle mass/weight were the most important variables (variable importance: 53% and 12%, respectively) in the prediction of self-reported incident mobility disability. D₃Cr muscle mass was the most important variable in predicting incident mobility disability when walking speed was excluded, followed by chair stands (variable importance: 35% and 33%, respectively). Body fat percentage, ALM, and grip strength were not selected as nodes in either analysis and had low or negligible variable importance.

Conclusions and Implications:

This study has provided valuable insights into the importance of different variables in predicting incident mobility disability in older men. Muscle mass by D₃Cr may be a key tool for predicting the risk of negative outcomes in older adults in the future, particularly in post-acute and long-term care settings.

Sarcopenia has been defined as a condition of low muscle strength, physical performance, and muscle mass.^1,2 The are many definitions of sarcopenia,^3–6 and universal consensus has proven elusive.^7,8 The optimal means of assessing muscle strength, mass, and physical performance remains a source of debate.⁷ Current operational definitions of sarcopenia comprise different diagnostic methods, adjustments and cutpoints for components of body composition, including lean body mass, fat-free mass, and muscle mass.^1–6 Most definitions of sarcopenia include walking speed, grip strength, and appendicular lean mass (ALM).^1–4,6

A novel and accurate method of measuring muscle mass, the D₃-creatine dilution (D₃Cr) method, has strong associations with negative functional and mobility disability outcomes.⁹ Furthermore, D₃Cr-measured muscle mass (D₃Cr muscle mass) is moderately correlated with commonly used strength and physical performance measures.¹⁰ The relative importance of D₃Cr muscle mass in predicting negative outcomes in older adults, such as incident mobility disability, compared with other variables used in current sarcopenia definitions [eg, grip strength, walking speed, and dual x-ray absorptiometry (DXA)ederived lean body mass] is unknown. We sought to answer this question by conducting a classification and regression tree analysis (CART) using data from the ongoing Osteoporotic Fractures in Men (MrOS) study—a large, well-characterized, prospective multicenter study of community-dwelling older men.

Methods

MrOS Study

Between 2000 and 2002, 5994 ambulatory community-dwelling men aged ≥65 years without bilateral hip replacements were enrolled in MrOS, a multicenter cohort study of aging and osteoporosis (mrosdata.sfcc-cpmc.net).^11,12 All men provided written informed consent. The study was approved by the Institutional Review Board at each center. In 2014–2016, 2786 survivors were contacted to participate in visit 4 (year 14) clinic visit. Of these, 362 refused participation, 583 completed questionnaires only, and 1841 completed questionnaires and at least part of the clinic visit (Supplementary Material 1).⁹

Measures of Muscle Mass

The D₃Cr dilution method was used to estimate muscle mass. Participants ingested 30 mg of stable isotope-labeled creatine (D₃-creatine). Fasting urine samples were collected 72 to 144 hours following ingestion. High-performance liquid chromatography and tandem mass spectroscopy was used to measure D₃-creatinine, unlabeled creatinine, and creatine. With these measures, a validated algorithm was used to determine total body creatine pool size and skeletal muscle mass.¹³

Measures of lean mass and body fat appendicular lean mass (ALM) and body fat were determined by whole-body DXA scans (Hologic 4500 scanners, Waltham, MA) in a previously described method.¹⁴

Measures of Physical Performance and Strength

Walking speed was measured over a 6-m course using the average of 2 trials (m/s).¹⁵ Time and ability to complete 5 repeated chair stands was assessed.¹⁶ The number of chair stands in 10 seconds was calculated. Those unable to commence the task scored zero. Grip strength (kg) was assessed using Jamar handheld dynamometers. Two tests were taken from each hand, and the maximum value obtained across all tests was included in the analysis.

Anthropometric Measures

Height (m) was measured at each visit using a Harpenden stadiometer. Weight was measured using a balance beam scale at all sites except Portland, which used a digital scale.

Incident Mobility Disability

Participants completed activities of daily living questionnaires on mobility, including walking 2 to 3 blocks on level ground, climbing 10 steps without resting, and carrying or lifting 10 lb. Men classified the degree of difficulty (none, some, much, or unable). These questions were asked at both visit 4 (year 14 visit) and a follow-up mailed questionnaire 2.2 ± 0.3 years later (year 16). We defined incident mobility disability as any new self-reported inability to perform 1 or more of the mobility measures. Those with disability at the initial time point (visit 4) were excluded from the analysis.

Participant Characteristics

Participants answered questions at the baseline visit on race and education level. Questions regarding alcohol use, comorbidities, smoking status, and the Physical Activity Scale for the Elderly were answered at visit 4.¹⁷ Cognitive performance was assessed by the Teng Modified Mini-Mental State Exam (3MS) at visit 4.¹⁸

Study Sample

All 1841 with a year 14 clinic visit were invited to complete the D₃Cr dilution protocol; no inclusion or exclusion criteria were applied. A total of 1641 agreed to participate and 187 were excluded, for reasons which have been previously described.⁹ A total of 1425 men had valid measures of D₃Cr muscle mass/weight, of which 327 were missing outcome or covariate data, or had mobility disability at the year 14 visit. A total of 1098 were included in the CART analysis (Supplementary Material 1).

Statistical Analysis

Participant characteristics were summarized comparing those with and without incident mobility disability using means and standard deviations for continuous variables and counts and percentages for categorical data. Groups were compared using the paired t test and Kruskal-Wallis tests. CART was selected as the analytical method to understand which variable or combination of variables best predicts incident mobility disability. The analysis was conducted with 10-fold cross-validation.^3,19 Eleven key variables related to sarcopenia^1–6 were included as potential predictors of mobility disability [height (m), weight (kg), total percentage body fat, ALM (kg)/weight (kg), ALM (kg)/height² (m), ALM (kg)/body mass index (BMI), muscle mass/weight (kg) by D₃Cr, walking speed (m/s), chair stands in 10 seconds, grip strength (kg)]. An additional CART model was conducted excluding the walking speed variable in light of its proximity with the outcome, mobility disability.

The CART analysis assigns variable importance reported in percentage integer values to numerically denote a variable’s “importance” in predicting the outcome. Variables with less than 1% importance are not presented. Importance is derived from averages of both primary and surrogate divisions of the trees generated by the CART model. Subsequent pruning of parsimonious splits of the tree were required for both models. We required that both CART models had no greater than 3 generations of nodes, which allowed for meaningful interpretation and prevented overfitting or divisions that insignificantly altered the models’ predictive ability. Analyses were conducted using SAS, version 9.4, and the rpart package in R software, version 3.4.0.

Results

There were 1098 participants in the MrOS study with all variable data required for CART analysis. Of these, 104 (9%) developed mobility disability during follow-up. The mean age of participants was 83.7 ± 3.7 years, with BMI 26.8 ± 3.5 and walking speed 1.12 ± 0.22 m/s; 8.9% were nonwhite. Participant baseline characteristics in the study are described in Table 1.

Table 1.

Characteristics of Men Who Participated in Visit 4 of the MrOS Study

Anthropometrics and Demographics	Men (n = 1098)	Incident Mobility Disability		P Value
		Yes (n = 104)	No (n = 994)
Age	83.7 ± 3.7	85.5 ± 4.2	83.5 ± 3.6	<.001
White	990 (90.2)	99 (95.2)	891 (89.6)	.07
Height, m	1.73 ± 0.07	1.72 ± 0.07	1.73 ± 0.06	.72
Weight, kg	79.8 ± 12.1	82.7 ± 13.2	79.5 ± 11.9	.012
BMI	26.8 ± 3.5	27.7 ± 3.6	26.7 ± 3.5	.003
DXA ALM/height2	7.57 ± 0.86	7.52 ± 0.92	7.58 ± 0.85	.48
DXA ALM/weight	0.29 ± 0.03	0.27 ± 0.03	0.29 ± 0.03	<.001
DXA ALM/BMI, m2	0.85 ± 0.11	0.81 ± 0.10	0.86 ± 0.11	<.001
Percentage fat mass, %	27.6 ± 5.6	29.6 ± 5.5	27.4 ± 5.6	<.001
D3O muscle mass, kg	24.7 ± 4.0	23.1 ± 3.9	24.8 ± 3.9	<.001
D3Cr muscle mass/weight	0.31 ± 0.05	0.28 ± 0.04	0.32 ± 0.05	<.001
Maximum grip strength, kg	36.5 ± 7.5	32.6 ± 7.3	36.9 ± 7.4	<.001
Walking speed, m/s	1.12 ± 0.22	0.94 ± 0.21	1.14 ± 0.21	<.001
Chair stands in 10 seconds	4.0 ± 1.6	2.8 ± 1.7	4.2 ± 1.5	<.001
Number of medical comorbidities, n (%)*
0	745 (67.9)	68 (65.4)	677 (68.1)	.13
1	272 (24.8)	22 (21.2)	250 (25.2)
2	66 (6.0)	12 (11.5)	54 (5.4)
3	14(1.3)	2 (1.9)	12 (1.2)
4	1 (0.1)	0	1 (0.1)
Education
<High school	30 (2.7)	9 (8.7)	21 (2.1)	<.001
High school graduate	154 (14.0)	19(18.3)	135 (13.6)
>High school	914 (83.2)	76 (73.1)	838 (84.3)
Alcohol use
< 1 drink/wk	513 (46.7)	53 (51.0)	460 (46.3)	.15
1–13 drinks/wk	533 (48.5)	43 (41.4)	490 (49.3)
14+ drinks/wk	52 (4.7)	8 (7.7)	44 (4.4)
Smoking status
Never	434 (39.5)	42 (40.4)	392 (39.4)	.95
Current or past	616 (56.1)	58 (55.8)	558 (56.1)
Missing smoking data	48 (4.4)	4 (3.9)	44 (4.4)
Physical activity score (PASE)	127.1 ± 62.5	92.6 ± 55.0	130.7 ± 62.1	<.001
Modified Mini-Mental	93.0 ± 6.3	91.4 ± 6.2	93.2 ± 6.3	<.001
State (3MS)
Exam score

PASE, Physical Activity Scale for the Elderly; SD, standard deviation.

Data shown as mean SD or n (%).

^*Medical comorbidities included congestive heart failure, chronic obstructive pulmonary disease, diabetes mellitus, and ischemic heart disease.

For the outcome of incident mobility disability with walking speed included in the analysis, CART models identified walking speed (variable importance: 53%), followed by D₃Cr muscle mass/weight (variable importance: 12%), as the most important predictors (Table 2). CART models identified walking speed as the primary node (Figure 1A). Those with a walking speed less than 0.88 m/s were more than 5 times more likely to develop incident mobility disability than those with walking speed equal to or greater than 0.88 m/s. The model identified the secondary node as D₃Cr muscle mass/weight of 0.32, among those with slow walking speed (<0.88 m/s). Walking speed of 0.73 m/s was identified as the tertiary node. Height, weight, BMI, total percentage body fat, ALM/weight, ALM/height², ALM/BMI, chair stands, and grip strength were not selected as nodes and had low variable importance.

Table 2.

Variable Importance for Incident Mobility Disability by Classification and Regression Tree Analysis in the MrOS Study

Variable	Variable Importance, %	Variables of Negligible Importance, <1%
Walking speed included in analysis Walking speed, m/s	53	Grip strength, kg DXA ALM/height2
D3Cr muscle mass/weight	12
Height, m	7
Weight, kg	6
DXA ALM/BMI	5
Chair stands in 10 s	5
BMI	4
DXA ALM/weight	4
Percentage fat mass, %	3
Walking speed excluded from analysis D3Cr muscle mass/weight	35	Grip strength, kg DXA ALM/height2
Chair stands in 10s	33
BMI	7
DXA ALM/weight	7
Percentage fat mass, %	6
DXA ALM/BMI	6
Weight, kg	4
Height, m	2

Fig. 1.

Classification and regression tree for incident mobility disability of sample: (A) including walking speed and (B) excluding walking speed. CS, chair stands; D3Cr MM/wgt, D3Cr-measured muscle mass/weight; IMD, incident mobility disability.

When walking speed was excluded from the analysis, D₃Cr muscle mass/weight (variable importance: 35%) followed by chair stands in 10 seconds (variable importance: 33%) were the most important variables predicting incident mobility disability. Three nodes were identified. Chair stands was the primary node and D₃Cr muscle mass/weight were the secondary and tertiary nodes (Figure 1B). Those with less than 2.5 chair stands in 10 seconds were more than 4 times more likely to have incident mobility disability than those with 2.5 chair stands or greater in 10 seconds. D₃Cr muscle mass/weight <0.3 was a strong predictor of mobility disability. No participants with mobility disability had chair stands <2.5 and D₃Cr muscle mass/weight ≥0.3, whereas 41.3% of participants developed mobility disability when chair stands <2.5 and D₃Cr muscle mass/weight <0.3. Height, weight, total percentage body fat, ALM/weight, ALM/height², ALM/BMI, chair stands, and grip strength were not selected as nodes and had low variable importance in the model without walking speed.

Discussion

This novel study in older, community-dwelling men found that, of the included variables, walking speed was the strongest predictor of incident mobility disability, followed by D₃Cr muscle mass/weight. Excluding walking speed from the analyses, given its clinical contiguity with the outcome, found that D₃Cr muscle mass/weight followed by chair stands were the most important predictors. Grip strength, ALM, total body fat percentage, and anthropometric measures were relatively unimportant predictors of incident mobility disability.

The purpose of these exploratory analyses was to allow key variables to compete for an important clinical outcome in older persons—incident mobility disability.²⁰ Incident mobility disability was selected as the outcome as it is independently associated with loss of independence, institutionalization, mortality,²¹ and health care costs.²²

In contrast to other studies where grip strength has been identified as a predictor of poor outcomes in older adults^23,24 and a strong discriminator of slow walking speed in cross-sectional analyses,³ our CART analysis showed that grip strength was of negligible importance in predicting incident mobility disability. A possible explanation for this finding is that D₃Cr muscle mass/ weight directly competed with grip strength (given their moderate correlation¹⁰) and was the stronger of the two in predicting the outcome. A recent factor analysis of the same study cohort showed that grip strength and D₃Cr muscle mass/weight co-segregated into the same factor (defined as the muscle mass, strength, and performance factor, which also included walking speed and chair stands in 10 seconds). Together, this factor was strongly protective against relative risk of mobility disability.¹⁰ In both CART models, ALM (with standardization to body size) had low variable importance and was not selected as a node in the tree models. This is an important finding that may clarify the mixed results observed in other studies regarding the association of ALM with negative functional and mobility outcomes in older adults.^9,25–28 One explanation for the lack of association of DXA-derived ALM with outcomes may be that ALM is an inaccurate surrogate of muscle mass.²⁹ Given its lack of importance in our models, the value of including ALM and its adjustments should be considered in future definitions of sarcopenia, despite the fact that a recent international consensus guideline recommended ALM as the preferred tool for assessing muscle mass.³⁰ Our findings are consistent with a recent Position Statement from the Sarcopenia Diagnostic and Outcomes Consortium (SDOC) that “lean mass measured by DXA should not be included in the definition of sarcopenia.”⁵

Perhaps the most significant finding in our study was that D₃Cr muscle mass/weight had strong predictive capacity at the secondary node in both models. In the model excluding walking speed, there were no participants with incident mobility disability if chair stands were less than 2.5 seconds and D₃Cr muscle mass/weight ≥0.3, whereas 41.3% had incident mobility disability if D₃Cr muscle mass/weight was <0.3. This suggests that either a higher D₃Cr muscle mass/weight is strongly protective for the outcome, or that the CART models produced overfitted trees for the data, or a combination of the two. Our study was strengthened by the large sample size, prospective design with men not selected for any specific health outcome, and the use of an important incident outcome allowing us to establish temporal order. In addition, CART is insensitive to outliers and its agnostic nature permits equitable and unbiased competition among highly correlated predictors without prespecification of interaction terms. The cutpoints derived in our CART analyses are not proposed for use as diagnostic criteria, principally because of the relatively limited generalizability of our sample. Although informative, our findings cannot be directly compared with previous CART models used to identify sarcopenia diagnostic cut-points, as both the outcome and the competing variables differ.^3,17 Furthermore, our analyses are limited to older, mostly white, community-dwelling men and whether similar results would be found in women, other racial and ethnic groups, or population-based samples is not known.

The D₃Cr dilution method is not currently approved for clinical use, which affects the practical applications of our findings. A significant limitation of CART analysis is that it does not consider important issues that affect clinical decision making, including patient factors (eg, those who cannot perform the chair-sit-to-stand because of osteoarthritis), resources (eg, test availability), and public health costs (eg, cost-benefit ratios). Nevertheless, our results generally support the use of walking speed as a component of existing sarcopenia algorithms given the importance of walking speed shown in our analyses.

Conclusions and Implications

This study has provided valuable insights into the relative importance of different measures in predicting incident mobility disability in older men. Future research should investigate the D₃Cr dilution method’s clinical utility in predicting risk of negative outcomes in older adults. Given the time required to undertake the D₃Cr dilution method (between labeled creatine ingestion and urine measurement), one may consider this novel method’s potential role in the care of older adults in post-acute or long-term care settings.

Supplementary Material

Supplementary Flow Chart

Supplementary Material 1. Participant flow in the Osteoporotic Fractures in Men (MrOS) study. CART, classification and regression tree; D₃Cr, D₃-creatine dilution method.