Development and Validation of a Predictive Model for Sarcopenia Risk in Older Chinese Adults Based on Key Factors

doi:10.21203/rs.3.rs-4851355/v1

Download PDF

Research Article

Development and Validation of a Predictive Model for Sarcopenia Risk in Older Chinese Adults Based on Key Factors

https://doi.org/10.21203/rs.3.rs-4851355/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Background

Sarcopenia, characterized by progressive loss of skeletal muscle mass and function, poses a significant health risk to the aging population. This study aims to construct and validate a predictive model for sarcopenia in elderly Chinese individuals using data from the China Health and Retirement Longitudinal Study (CHARLS).

Methods

We observed participants aged 60 and above without a diagnosis of sarcopenia in 2011 and followed up in 2013 for the incidence of sarcopenia. After excluding participants with missing data, disabilities, cancer, and extreme values, a total of 2,197 individuals were included in the study. Sarcopenia was assessed based on the 2019 Asian Working Group for Sarcopenia (AWGS) criteria. The predictive factors analyzed included sociodemographic characteristics, health status, lifestyle habits, psychological status, pain-related information, and blood biochemical indicators. LASSO-logistic regression and XGBoost machine learning models were employed to identify key predictors and develop the predictive model.

Results

The study identified older age, lower BMI, female gender, memory-related diseases, arthritis or rheumatism, shorter night sleep duration, and lower education level as independent risk factors for sarcopenia. Both methods produced models with high predictive accuracy, though the XGBoost model had a slightly higher AUC than the logistic regression model (0.881 vs. 0.849). However, the difference in AUC between the two models was not statistically significant. The XGBoost model demonstrated higher sensitivity but lower specificity. Ultimately, the logistic regression model was considered the better choice for this study due to its interpretability and comparable performance.

Conclusion

This study identified key risk factors for sarcopenia using machine learning and traditional statistical methods, such as logistic regression, and developed robust predictive models. The findings provide valuable insights for early intervention and management of sarcopenia in the elderly Chinese population, highlighting the need for a multidisciplinary approach to improve health outcomes in this group.

Sarcopenia

Predictive Model

Elderly

XGBoost

The global population is aging rapidly, and China is no exception to this demographic trend. According to the 2020 Seventh National Population Census, over 260 million people in China are aged 60 and above, accounting for 18.7% of the total population, with more than 190 million individuals aged 65 and above, representing 13.5% of the population^[1]. The prevalence of sarcopenia among individuals over 60 years old globally ranges from 10% to 27%, influenced by age, nutritional intake, physical inactivity, and diseases. The World Health Organization estimates that over 50 million people worldwide are affected by sarcopenia, with this number expected to surpass 200 million by 2050^[2]. Since 2016, sarcopenia has been officially recognized by the WHO as a disease and remains a significant public health concern among the aging population^[3]. Sarcopenia is associated with numerous adverse outcomes, including comorbidities, disability, falls, increased hospitalization, functional decline, and higher mortality^[4]. Therefore, raising awareness, enhancing screening in high-risk populations, and implementing early assessments and targeted interventions are crucial to prevent adverse outcomes.

Existing predictive models for sarcopenia risk have several limitations, including small sample sizes, difficulty and time-consuming data collection, and the inability to comprehensively capture all risk factors. Lin et al.'s study using data from the West China Health and Aging Trend (WCHAT) cohort, which included individuals aged 60 and above, identified 87 cases of sarcopenia among 1042 non-sarcopenic individuals during a one-year follow-up period. The PRE-SARC model, developed using stepwise logistic regression, identified seven independent risk factors: age, gender, BMI, low physical activity level, risk of malnutrition, pain, and calf circumference. The AUC of this model was 0.87, but the data were derived from community samples in four provinces in western China, potentially limiting the generalizability of the findings due to regional population characteristics, lifestyle, and environmental factors^[5].

The CHARLS is a national large-scale social survey project aimed at collecting data on the health, economic status, and social life of middle-aged and elderly populations in China. Based on the 2015 CHARLS survey data, a cross-sectional study by Li et al. found a sarcopenia prevalence of 28.8% among elderly Chinese individuals^[6]. This study developed a sarcopenia risk prediction model based on factors such as gender, body mass index (BMI), mean systolic blood pressure , mean diastolic blood pressure and pain. The AUC of this model was 0.77 in the training set, indicating its potential for sarcopenia prediction, but with room for improvement. While incorporating more predictive factors might enhance model performance, it could also reduce its feasibility and practicality in real-world applications. In community and primary care settings, simple, easy-to-use, and low-cost models are more valuable.

With the rise of artificial intelligence, machine learning algorithms such as support vector machines, random forests, and extreme gradient boosting have demonstrated significant efficacy in disease prediction^[7]. This study applies the XGBoost machine learning model and traditional logistic regression to data analysis to develop a more suitable sarcopenia prediction model for the elderly Chinese population. Through this approach, we aim to improve the model's accuracy and practicality, providing more precise and reliable tools for health management in the elderly Chinese population, thus offering new perspectives and strategies for the prevention and management of sarcopenia in older adults.

2.1 Data Source

This study utilized data from the China Health and Retirement Longitudinal Study (CHARLS), publicly accessible at http://charls.pku.edu.cn. CHARLS has conducted biennial surveys since 2011, but the physical examination questionnaire was excluded after the 2015 survey. We observed that participants diagnosed with sarcopenia in 2013 showed some improvement in the 2015 survey. Therefore, we selected data from the 2011 and 2013 surveys to assess the incidence of sarcopenia in non-sarcopenic individuals over a two-year period and analyze the related risk factors. The CHARLS study was approved by the Ethics Committee of Peking University (IRB00001052-11015), and all participants provided written informed consent.

2.2 Data Extraction

In the 2011 baseline survey, 17,707 individuals participated. We excluded 15,510 individuals due to reasons such as missing data, being under 60 years of age, having physical disabilities, cancer, having been diagnosed with sarcopenia at baseline, and those who were lost to follow-up or lacked sarcopenia diagnostic information in the 2013 survey. To ensure data rigor and model stability, we excluded individuals with a height less than 130 cm or a weight over 100 kg and removed all individuals with missing data. Ultimately, 2,197 individuals were included in the study for data analysis, as detailed in Fig. 1.

2.3 Assessment of Symptoms of Sarcopenia

This study screened individuals for possible sarcopenia based on the 2019 Asian Working Group for Sarcopenia (AWGS) criteria. Assessments included muscle strength, physical function, and appendicular skeletal muscle mass (ASM). According to AWGS, the thresholds for low grip strength are < 28 kg for men and < 18 kg for women. Physical function decline was defined as a five-time chair stand time > 12 seconds or a 6-meter walk speed < 1 meter/second. ASM was measured using a validated anthropometric equation developed specifically for the Chinese population^[8]:

ASM = 0.193 × weight (kg) + 0.107 × height (cm) − 4.157 × sex − 0.037 × age (years) − 2.631

If an individual failed either the grip strength test or the physical function test, they were considered at risk for sarcopenia. If the relative skeletal muscle mass index (ASM/Ht²) showed < 7.0 kg/m² for men or < 5.4 kg/m² for women, and they also exhibited muscle strength or physical function decline, they were diagnosed with sarcopenia^[9]. Severe sarcopenia was diagnosed if all three criteria were met. For the purposes of this study, participants were classified into sarcopenia and non-sarcopenia groups for analysis.

2.4 Selection of Predictive Factors

This study included 49 predictive factors based on comprehensive consideration and data availability. These factors included: sociodemographic characteristics (age, gender, residence, and education level); health status (BMI, hypertension, diabetes, and heart disease); lifestyle habits (smoking, drinking, nap duration, and physical activity intensity); psychological status (depression and pain); pain information (head and neck pain, trunk pain, upper limb pain, lower limb pain, and stomach pain); and blood biochemical indicators (blood glucose, cholesterol, triglycerides, C-reactive protein, and Tyg index). As elderly individuals with impaired ADL (Activities of Daily Living) and IADL (Instrumental Activities of Daily Living) scores or physical function already exhibit significant sarcopenia symptoms, these scores were used as auxiliary indicators for sarcopenia diagnosis rather than predictive factors.

2.5 Assessment of Depression and Pain

This study used the 10-item Center for Epidemiologic Studies Depression Scale (CESD-10) to assess depressive symptoms (DS). The CESD-10 is widely used in large-scale population studies and has demonstrated good validity and reliability in the CHARLS study. Items 5 and 8 are reverse scored, with a total score range of 0–30. In this study, a CESD-10 score ≥ 12 was considered positive for DS; otherwise, it was negative. Higher scores indicate more severe depressive symptoms^[10]. Pain was reclassified based on survey records of pain in various body parts, including the head, shoulder, arm, wrist, fingers, chest, stomach, back, waist, buttocks, legs, knees, ankles, toes, and neck. Pain was categorized into head and neck pain, trunk pain, upper limb pain, lower limb pain, and stomach pain. Considering that waist and leg pain are common comorbid symptoms in elderly patients, waist pain was included in the lower limb pain category. Participants were also asked to assess the severity of pain, categorized as no pain, mild, moderate, and severe. Through detailed classification and data integration of pain information, we aim to more accurately assess and analyze the participants' pain conditions, providing more precise data support for the study.

2.6 Statistical Methods

2.6.1 Baseline Comparison

This study used data from the 2011 CHARLS database for baseline comparison analysis. Data were processed using SPSS 25.0 software. Continuous variables with non-normal distributions were expressed as medians and interquartile ranges, and group comparisons were performed using the Mann-Whitney U test. Categorical variables were expressed as frequencies and percentages, and group comparisons were conducted using the chi-square test (χ² test) or Fisher's exact test.

2.6.2 LASSO-Logistic Regression Model Development

We used R software (version 4.3.3), and the dataset was randomly divided into a training set (n = 1539) and a validation set (n = 658) in a 7:3 ratio. During this process, we set a random seed to ensure the reproducibility of the sampling process. First, we used the Least Absolute Shrinkage and Selection Operator (LASSO) regression analysis for variable selection. LASSO regression effectively handles high-dimensional datasets and multicollinearity, accurately selecting important variables and improving model interpretability. LASSO regression analysis was performed on the training set data, and the optimal tuning parameter λ was determined through 10-fold cross-validation, identifying significant feature variables. These features were then included in a multivariable logistic regression analysis to further select predictive factors with P-values < 0.05 to construct the predictive model.

2.6.3 XGBoost Model Development

We employed the Extreme Gradient Boosting (XGBoost) algorithm, with the following steps: data preprocessing, including cleaning, organizing, and standardizing the data; factor variable conversion, converting categorical variables to appropriate numerical forms; dividing the dataset into training and testing sets in an 8:2 ratio; performing 10-fold cross-validation to optimize parameters and find the best parameter combination; training the final model using the optimized parameters; making predictions and evaluating performance using ROC curves and AUC; and calculating feature importance to assess each feature's contribution to the prediction results.

2.6.4 Model Validation and Evaluation

We assessed the discrimination ability of the models by plotting receiver operating characteristic (ROC) curves and used the DeLong test to compare the differences in AUC between the traditional multivariable logistic regression model and the XGBoost prediction model. The Hosmer-Lemeshow test was used to evaluate model calibration, with a p-value > 0.05 indicating good consistency between predicted and observed outcomes. A Brier score less than 0.25 indicates good overall performance in discrimination and calibration. Decision curve analysis (DCA) was used to evaluate the clinical utility of the models at different probability thresholds. Finally, we used restricted cubic splines to analyze nonlinear relationships, which can flexibly capture complex relationships between variables and help understand the impact of each predictive factor on the outcome more accurately.

3.1 Baseline Characteristics Analysis

In this study, we followed 2,197 participants over two years, during which 226 developed sarcopenia. Baseline characteristics analysis revealed that participants in the sarcopenia group were older, had lower BMI, and a higher proportion were female. Lifestyle habits showed that the sarcopenia group had lower rates of alcohol consumption and shorter night sleep duration. Psychologically, this group had higher rates of depression. Health status analysis indicated that the sarcopenia group had lower incidences of hypertension and diabetes or hyperglycemia, but higher incidences of vision problems, hearing problems, memory-related diseases, and use of dentures. Blood biochemical indicators showed that the sarcopenia group had lower levels of Tyg index, glucose, triglycerides, LDL cholesterol, uric acid, and hematocrit, while HDL cholesterol levels were higher (all P-values < 0.05). Detailed baseline characteristics are presented in Table 1.

Table 1

Baseline characteristics of the study population
Variables	Total (n = 2197)	Non-sarcopenia(n = 1971)	Sarcopenia (n = 226)	P-value
Age(years)	65.00 [62.00;70.00]	65.00 [62.00;69.00]	69.00 [64.00;73.75]	< 0.001
BMI	23.58 [21.30;25.99]	23.84 [21.73;26.24]	20.77 [19.30;22.36]	< 0.001
Tyg_index	8.64 [8.26;9.09]	8.67 [8.28;9.12]	8.52 [8.14;8.89]	< 0.001
BUN(mg/dl)	15.63 [13.02;18.91]	15.57 [12.97;18.89]	15.87 [13.26;19.17]	0.526
Glucose(mg/dl)	104.04 [96.12;116.46]	104.40 [96.30;116.82]	102.33 [93.64;111.10]	0.001
CHO(mg/dl)	192.14 [168.17;217.27]	192.53 [168.94;217.66]	190.40 [161.21;215.82]	0.081
Triglycerides(mg/dl)	106.20 [76.11;157.53]	107.97 [77.00;160.18]	96.91 [72.57;135.18]	0.001
HDL(mg/dl)	48.71 [39.82;59.15]	48.33 [39.82;58.76]	51.42 [43.01;62.15]	0.001
LDL(mg/dl)	116.75 [95.49;139.95]	117.53 [95.68;140.34]	113.27 [93.94;133.09]	0.040
CRP(mg/l)	1.20 [0.64;2.40]	1.21 [0.65;2.40]	1.15 [0.55;2.46]	0.444
HbA1c (%)	5.20 [4.90;5.50]	5.20 [4.90;5.50]	5.20 [4.90;5.40]	0.051
Uric Acid(mg/dl)	4.50 [3.76;5.42]	4.56 [3.80;5.46]	4.24 [3.51;5.03]	< 0.001
Hematocrit(%)	41.60 [37.80;45.20]	41.80 [37.80;45.30]	40.30 [36.62;43.60]	< 0.001
Night sleep duration(h)	6.00 [5.00;8.00]	6.00 [5.00;8.00]	6.00 [4.00;7.00]	< 0.001
Male	1227 (55.85%)	1129 (57.28%)	98 (43.36%)
Female	970 (44.15%)	842 (42.72%)	128 (56.64%)
Residence(%)				0.0516
City/Town	2022 (92.03%)	1806 (91.63%)	216 (95.58%)
Village	175 (7.92%)	165 (8.37%)	10 (4.42%)
Depression(%)				0.002
NO	1517 (69.05%)	1382 (70.12%)	135 (59.73%)
YES	680 (30.95%)	589 (29.88%)	91 (40.27%)
Head and neck pain(%)				0.057
NO	1894 (86.21%)	1709 (86.71%)	185 (81.86%)
YES	303 (13.79%)	262 (13.29%)	41 (18.14%)
Trunk pain(%)				0.597
NO	1934 (88.03%)	1738 (88.18%)	196 (86.73%)
YES	263 (11.97%)	233 (11.82%)	30 (13.27%)
Upper limb pain(%)				0.083
NO	1843 (83.89%)	1663 (84.37%)	180 (79.65%)
YES	354 (16.11%)	308 (15.63%)	46 (20.35%)
Lower limb pain(%)				0.114
NO	1590 (72.37%)	1437 (72.91%)	153 (67.70%)
YES	607 (27.63%)	534 (27.09%)	73 (32.30%)
Gastric pain(%)				0.502
NO	2041 (92.90%)	1834 (93.05%)	207 (91.59%)
YES	156 (7.10%)	137 (6.95%)	19 (8.41%)
Brain Damage/Mental Retardation(%)				0.332
NO	2149 (97.82%)	1930 (97.92%)	219 (96.90%)
YES	48 (2.18%)	41 (2.08%)	7 (3.10%)
Vision Problem(%)				0.041
NO	2049 (93.26%)	1846 (93.66%)	203 (89.82%)
YES	148 (6.74%)	125 (6.34%)	23 (10.18%)
Hearing Problem(%)				0.011
NO	1953 (88.89%)	1764 (89.50%)	189 (83.63%)
YES	244 (11.11%)	207 (10.50%)	37 (16.37%)
Speech Impediment(%)				0.352
NO	2193 (99.82%)	1968 (99.85%)	225 (99.56%)
YES	4 (0.18%)	3 (0.15%)	1 (0.44%)
Hypertension(%)				0.001
NO	1495 (68.05%)	1318 (66.87%)	177 (78.32%)
YES	702 (31.95%)	653 (33.13%)	49 (21.68%)
Dyslipidemia(%)				0.016
NO	1951 (88.80%)	1739 (88.23%)	212 (93.81%)
YES	246 (11.20%)	232 (11.77%)	14 (6.19%)
Disabetes or High Blood Sugar(%)				0.035
NO	2039 (92.81%)	1821 (92.39%)	218 (96.46%)
YES	158 (7.19%)	150 (7.61%)	8 (3.54%)
Chronic Lung Diseases(%)				0.172
NO	1924 (87.57%)	1733 (87.92%)	191 (84.51%)
YES	273 (12.43%)	238 (12.08%)	35 (15.49%)
Liver Disease(%)				0.183
NO	2095 (95.36%)	1875 (95.13%)	220 (97.35%)
YES	102 (4.64%)	96 (4.87%)	6 (2.65%)
Heart Disease(%)				0.562
NO	1872 (85.21%)	1676 (85.03%)	196 (86.73%)
YES	325 (14.79%)	295 (14.97%)	30 (13.27%)
Stroke(%)				1.000
NO	2134 (97.13%)	1914 (97.11%)	220 (97.35%)
YES	63 (2.87%)	57 (2.89%)	6 (2.65%)
Kidney Disease(%)				0.512
NO	2053 (93.45%)	1839 (93.30%)	214 (94.69%)
YES	144 (6.55%)	132 (6.70%)	12 (5.31%)
Stomach or Other Digestive Disease(%)				0.070
NO	1722 (78.38%)	1556 (78.94%)	166 (73.45%)
YES	475 (21.62%)	415 (21.06%)	60 (26.55%)
Emotional, Nervous, or Psychiatric Problems(%)				1.000
NO	2166 (98.59%)	1943 (98.58%)	223 (98.67%)
YES	31 (1.41%)	28 (1.42%)	3 (1.33%)
Memory-related disease(%)				0.032
NO	2156 (98.13%)	1939 (98.38%)	217 (96.02%)
YES	41 (1.87%)	32 (1.62%)	9 (3.98%)
Arthritis or rheumatism(%)				0.105
NO	1377 (62.68%)	1247 (63.27%)	130 (57.52%)
YES	820 (37.32%)	724 (36.73%)	96 (42.48%)
Asthma(%)				0.239
NO	2091 (95.18%)	1880 (95.38%)	211 (93.36%)
YES	106 (4.82%)	91 (4.62%)	15 (6.64%)
Wearing Dentures,Fixed or Removable(%)				< 0.001
NO	1930 (87.85%)	1749 (88.74%)	181 (80.09%)
YES	267 (12.15%)	222 (11.26%)	45 (19.91%)
Distressed by pain(%)				0.193
NO	1470 (66.91%)	1328 (67.38%)	142 (62.83%)
YES	727 (33.09%)	643 (32.62%)	84 (37.17%)
Smoke(%)				0.258
NO	1467 (66.77%)	1308 (66.36%)	159 (70.35%)
YES	730 (33.23%)	663 (33.64%)	67 (29.65%)
Anemia(%)				0.059
NO	2100 (95.58%)	1890 (95.89%)	210 (92.92%)
YES	97 (4.42%)	81 (4.11%)	16 (7.08%)
Midday nap(%)				0.279
time < = 30,min	1313 (59.76%)	1167 (59.21%)	146 (64.60%)
30 < time < = 60,min	513 (23.35%)	468 (23.74%)	45 (19.91%)
60 < time,min	371 (16.89%)	336 (17.05%)	35 (15.49%)
Education(%)				< 0.001
Never went to school	643 (29.27%)	542 (27.50%)	101 (44.69%)
Elementary school	1091 (49.66%)	988 (50.13%)	103 (45.58%)
Middle school and above	463 (21.07%)	441 (22.37%)	22 (9.73%)
Drink(%)				0.050
Drink more than once a month	583 (26.54%)	531 (26.94%)	52 (23.01%)
Drink but less than once a month	153 (6.96%)	144 (7.31%)	9 (3.98%)
None of these	1461 (66.50%)	1296 (65.75%)	165 (73.01%)
Exercise intensity(%)				0.607
No exercising	1650 (75.10%)	1479 (75.04%)	171 (75.66%)
Walking	298 (13.56%)	268 (13.60%)	30 (13.27%)
Do Moderate Activities	187 (8.51%)	171 (8.68%)	16 (7.08%)
Do Vigorous Activities	62 (2.82%)	53 (2.69%)	9 (3.98%)
Pain intensity(%)				0.312
No pain	1473 (67.05%)	1330 (67.48%)	143 (63.27%)
Mild	181 (8.24%)	164 (8.32%)	17 (7.52%)
Moderate	265 (12.06%)	236 (11.97%)	29 (12.83%)
Severe	278 (12.65%)	241 (12.23%)	37 (16.37%)
Pain distribution(%)				0.089
No pain in any location	1475 (67.14%)	1332 (67.58%)	143 (63.27%)
Pain in one location	241 (10.97%)	218 (11.06%)	23 (10.18%)
Pain in two locations	201 (9.15%)	179 (9.08%)	22 (9.73%)
Pain in three locations	134 (6.10%)	119 (6.04%)	15 (6.64%)
Pain in four locations	92 (4.19%)	74 (3.75%)	18 (7.96%)
Pain in five locations	54 (2.46%)	49 (2.49%)	5 (2.21%)
Notes: Medians and interquartile ranges (25th and 75th percentiles) were calculated for continuous variables, and frequencies and percentages for categorical variables.Abbreviations: BMI: Body Mass Index; Tyg_index: Triglyceride-glucose index; BUN: Blood Urea Nitrogen; CHO: Total Cholesterol; HDL: High-Density Lipoprotein Cholesterol; LDL: Low-Density Lipoprotein Cholesterol; CRP: C-Reactive Protein; HbA1c: Glycated Hemoglobin.

3.2 LASSO Regression Model for Sarcopenia Risk Factors

Using the training set data, we applied the LASSO regression model with 10-fold cross-validation to identify the optimal λ value of 0.006689531. This process selected 18 significant predictors of sarcopenia, including blood urea nitrogen, glucose, total cholesterol, uric acid, hematocrit, age, stomach pain, education level, gender, hearing problems, BMI, Tyg index, night sleep duration, kidney disease, stomach or other digestive diseases, memory-related diseases, arthritis or rheumatism, and use of dentures.(Fig. 2A and Fig. 2B)

3.3 Multivariable Logistic Regression Analysis

The variables identified by the LASSO regression model were further analyzed using multivariable logistic regression. The results demonstrated that older age, female gender, memory-related diseases, and arthritis or rheumatism were independent risk factors for sarcopenia. Conversely, higher education level (junior high school and above), higher BMI, and longer night sleep duration were protective factors. All variables had variance inflation factor (VIF) values less than 1.2, indicating low multicollinearity and ensuring their independent impact on sarcopenia occurrence. Identifying these key factors provides a scientific basis for formulating targeted prevention and intervention strategies. The results of the multivariable analysis are shown in Table 2.

Table 2

Results of Multivariate Logistic Regression Analysis
Variables	B	SE	OR	CI	Z	P-value
Age	0.116	0.017	1.12	1.09–1.16	6.926	0
Sex	1.296	0.222	3.66	2.37–5.65	5.848	0
BMI	-0.394	0.038	0.67	0.63–0.73	-10.494	0
Education(Middle school and above)	-0.88	0.329	0.41	0.22–0.79	-2.672	0.008
Arthritis or rheumatism	0.448	0.196	1.57	1.07–2.3	2.281	0.023
Night sleep duration	-0.111	0.049	0.89	0.81–0.99	-2.247	0.025
Memory-related disease	1.128	0.568	3.09	1.02–9.41	1.986	0.047
Notes: B represents the regression coefficient, SE represents the standard error, OR represents the odds ratio, CI represents the confidence interval, Z represents the Z-value.

3.4 Validation of Predictive Models

To evaluate the performance of the predictive models, we used Receiver Operating Characteristic (ROC) curves. The Area Under the Curve (AUC) for the training set was 0.849 (95% CI: 0.821–0.878), indicating strong predictive performance. The validation set AUC was 0.835 (95% CI: 0.787–0.883), further confirming the model's robustness and reliability (Fig. 3A and Fig. 3B).

3.5 Model Calibration and Goodness-of-Fit Assessment

The Hosmer-Lemeshow test results showed a Brier score of 0.074 and a calibration slope of 1.0 for the training set, indicating high consistency between predicted and observed probabilities. The χ² value was 17.3778 with a p-value of 0.0431, slightly below 0.05, suggesting some differences that could be attributed to the large training sample size and potential overfitting. For the validation set, the Brier score was 0.075 with a calibration slope of 1.0, demonstrating high prediction consistency and accuracy across different datasets. The χ² value was 3.4886 with a p-value of 0.9417, indicating no significant difference between predicted and actual outcomes, thus confirming good model fit. Predicted probabilities of sarcopenia closely matched actual probabilities in both the training (Fig. 4A) and validation sets (Fig. 4B).

3.6 Clinical Validity Assessment

We utilized decision curve analysis (DCA) to assess the clinical validity of the models. The results indicated that the logistic regression predictive model had significant net benefits within the high-risk threshold range (particularly between 0.0 and 0.8) in both the training and validation sets, demonstrating the model's clinical practicality and predictive reliability. The decision curves showed that the net benefit of the model was significantly higher than the extreme strategies of treating all patients or treating none, further confirming the model's effectiveness and accuracy in clinical application. (Fig. 5A and Fig. 5B)

3.7 XGBoost Predictive Model and Feature Importance Analysis

In this study, we employed the XGBoost algorithm to predict the incidence of sarcopenia in Chinese elderly individuals over two years. Optimal hyperparameters were determined through ten-fold cross-validation, resulting in an AUC of 0.881 for the fifth fold. This high AUC value indicates strong predictive performance of the model (see Fig. 6). The overall accuracy was 92.27% (95% CI: 87.92%-95.43%), with a sensitivity of 99.01% and specificity of 16.67%, and a Kappa value of 0.2336. Although the AUC of the XGBoost model was slightly higher than that of the logistic regression model (0.881 vs. 0.849), the DeLong test results showed no statistically significant difference in AUC between the two models (test statistic D = -0.7968, p-value = 0.4262)(see Fig. 7). Thus, both models performed similarly in terms of discrimination ability. In the feature importance analysis, the XGBoost model identified the most influential predictors, with BMI and age being the most critical, having Gain values of 0.423 and 0.151, respectively. Gender (female), uric acid, night sleep duration, and LDL cholesterol also significantly impacted the prediction results, highlighting the essential roles of BMI and age in predicting sarcopenia risk among the elderly( Fig. 8).

The points labeled 0.087 (0.733, 0.944) for the XGBoost model and 0.088 (0.720, 0.830) for the Logistic Regression model represent specific cut-off values, with the numbers in parentheses indicating the specificity and sensitivity at those thresholds, respectively. These points provide insight into the models' performance at these specific thresholds.

The importance of each predictor is measured by the Gain metric, which reflects the contribution of a feature to the model's accuracy. The horizontal bars represent the relative importance of each predictor, with longer bars indicating higher importance. Predictors are ranked from top to bottom, with the most important features appearing at the top.

3.8 Nonlinear Relationship Analysis Between BMI, Night Sleep Duration and Sarcopenia Risk

We performed restricted cubic spline analysis on the training set to explore the nonlinear relationships between BMI, night sleep duration and sarcopenia risk.In the training set, a BMI range of approximately 21.285 to 26.013 was associated with a significant reduction in sarcopenia risk, with a log odds reduction of 2.43 (SE = 0.31708, 95% CI: -3.0514 to -1.8085), corresponding to an odds ratio (OR) of 0.088 (95% CI: 0.047 to 0.164). Beyond 26.013, the risk plateaued, indicating that a low BMI significantly increased sarcopenia risk( Fig. 9).

Increasing night sleep duration from 5 to 8 hours was associated with a gradual reduction in sarcopenia risk, with a log odds reduction of 0.395 (SE = 0.2247, 95% CI: -0.8359 to 0.0449), corresponding to an odds ratio (OR) of 0.673 (95% CI: 0.433 to 1.046). Although the confidence interval included 1, indicating statistical insignificance, the trend suggested that moderately increasing sleep duration might help reduce sarcopenia risk( Fig. 10).

This study demonstrates that the incidence of sarcopenia among the surveyed population in 2011 was 10.29% by 2013, with rates of 7.99% in men and 13.20% in women. A three-year follow-up of elderly residents in suburban Tianjin reported an incidence rate of 8.1% (9.6% for men and 6.8% for women)^[11]. Similarly, Lin et al. reported a one-year incidence of sarcopenia of approximately 8.35% in the Western China Health and Aging Trend cohort^[5]. Despite minor variations, these findings collectively indicate a high incidence of sarcopenia among the elderly in China, underscoring the need for public health interventions. Sarcopenia has emerged as a significant chronic disease affecting the elderly, warranting focused efforts on its prevention and management. Therefore, developing predictive models, identifying risk factors, and conducting targeted screenings and interventions for high-risk populations are crucial.

Our study employed LASSO regression for variable selection, followed by comparisons with multivariate logistic regression and the Extreme Gradient Boosting (XGBoost) machine learning method. ROC analysis revealed that both models exhibited comparable discriminatory capabilities. Although the XGBoost model demonstrated higher accuracy and sensitivity, it had relatively lower specificity. The high sensitivity resulted in a higher false-positive rate, indicating that more elderly individuals who were not at high risk of sarcopenia might be misclassified and subjected to unnecessary interventions. Furthermore, the XGBoost model identified 21 significant variables, with BMI, age, and gender ranking the highest, consistent with the multivariate logistic regression results. However, the interpretability of the XGBoost model was not ideal. Therefore, we concluded that the predictive model established using traditional logistic regression was more suitable for this study.

Our findings highlight that older women are at an independent risk for developing sarcopenia. Postmenopausal women experience a significant decline in estrogen levels, which directly impacts muscle tissue. Estrogen receptors are present on muscle cell membranes, cytoplasm, and nuclei, mediating protein synthesis through pathways such as IGF-1, Akt, and mTOR^[12]. Estrogen deficiency leads to decreased bone density and muscle mass, increasing the risk of sarcopenia ^[13]. Research has shown that postmenopausal women lose muscle mass and strength faster than men ^[14]. Women have relatively higher fat content and lower muscle mass, making them more susceptible to muscle loss with age. Numerous studies, including those by Chen et al., have shown higher incidence rates of sarcopenia in women compared to men^[15]. Cruz-Jentoft et al. conducted a systematic review revealing that women over 65 have a significantly higher incidence of sarcopenia than men^[16]. This finding aligns with our study, identifying women as an independent risk factor for sarcopenia. Physical activity is crucial in preventing sarcopenia, but women generally have lower activity levels, contributing to the higher incidence of sarcopenia. Some studies suggest hormone replacement therapy can improve muscle mass in postmenopausal women^[17], though its effectiveness remains controversial and requires further research.

This study did not use contextual memory and mental state scores as overall cognitive function assessment indicators due to data limitations and accessibility. Instead, we selected educational level and memory-related diseases as predictors to simplify the assessment. Higher education (middle school or above) was found to be a protective factor against sarcopenia. Individuals with higher education are more likely to access health information, adopt healthy lifestyles, and undergo regular physical examinations. Higher education is also associated with better cognitive reserve and higher physical activity levels, reducing the risk of sarcopenia. Memory-related diseases (such as Alzheimer's, brain atrophy, and Parkinson's disease) collectively affect cognitive function and the occurrence of sarcopenia through complex mechanisms. Alzheimer's disease is characterized by neuron loss and atrophy in the hippocampus and cerebral cortex, with β-amyloid plaques and tau protein tangles accompanied by chronic inflammation, leading to cognitive decline^[18]. Brain atrophy involves reduced brain tissue volume and neuron loss, affecting information processing and transmission. The main pathological changes in Parkinson's disease are the loss of dopaminergic neurons in the substantia nigra and Lewy body deposition, affecting motor and cognitive functions. Approximately 20–30% of patients develop Parkinson's disease dementia (PDD), characterized by memory loss, attention decline, slowed thinking, and impaired visual-spatial abilities^[19]. Most studies show a significant correlation between cognitive impairment and sarcopenia, with multiple mechanisms possibly causing a reciprocal relationship. These mechanisms include imbalances in actin secretion, vascular dysfunction, neuromuscular system damage, vitamin D deficiency, chronic inflammation, oxidative stress, and insulin resistance^[20].

Our study indicates that a higher BMI is a protective factor against sarcopenia. Most research shows that individuals with lower BMI have a higher risk of sarcopenia because those with higher BMI may have more muscle mass, helping maintain muscle strength and physical function^[21]. The obesity paradox suggests that in certain cases, obese individuals (especially those with mild to moderate obesity) may have better prognoses and lower mortality rates than normal or underweight individuals. This paradox has been observed in coronary heart disease patients, where obese individuals often show better short-term and long-term outcomes^[22]. However, excessive obesity may lead to decreased muscle mass and function, possibly due to metabolic disorders associated with adipose tissue, such as oxidative stress, inflammation, and insulin resistance^[23]. Thus, obese individuals face the risk of sarcopenia and may experience further muscle function decline due to the negative metabolic effects of adipose tissue. Our analysis of the nonlinear relationship between BMI and the risk of sarcopenia reveals that the risk significantly decreases when BMI ranges between approximately 21.285 and 26.013. Beyond 26.013, the risk stabilizes, indicating that higher BMI does not necessarily equate to better outcomes. Therefore, it is essential to consider both muscle mass and fat tissue ratios, maintaining an appropriate BMI (around 26) to benefit the elderly population in China and reduce the risk of sarcopenia.

Our study finds that arthritis or rheumatism is an independent risk factor for sarcopenia. Arthritis and rheumatism often accompany chronic systemic inflammation, leading to increased muscle catabolism and decreased anabolism, resulting in muscle mass and function decline. Additionally, these patients often reduce physical activity due to pain and limited mobility, further exacerbating muscle atrophy and strength loss. Although corticosteroids and other treatments effectively reduce inflammation and pain, long-term use may negatively impact muscle, increasing the risk of sarcopenia^[24]. Takeshi Mochizuki et al.found a significantly higher incidence of sarcopenia in elderly Japanese people with rheumatism than in the general population, emphasizing the importance of early diagnosis and management^[25]. These findings suggest that early sarcopenia screening and comprehensive intervention measures, including physical therapy and anti-inflammatory treatment, should be conducted in elderly arthritis patients to slow or prevent sarcopenia.

We found that longer nighttime sleep duration is a protective factor against sarcopenia in the elderly. Restricted cubic spline analysis shows that the risk of sarcopenia gradually decreases as nighttime sleep duration increases from 5 to 8 hours, reducing the log odds by 0.395. This trend suggests that moderately increasing sleep duration may help reduce the risk of sarcopenia in the elderly. This finding has important clinical and public health implications, providing a new perspective for preventing sarcopenia in the elderly. Adequate sleep promotes the secretion of growth hormone, which plays a crucial role in muscle growth and repair^[26]. Additionally, it can reduce inflammation levels in the body, protecting muscles by decreasing inflammatory responses. Insufficient sleep affects glucose metabolism and insulin sensitivity, increasing the risk of muscle loss^[27]. However, some studies indicate that excessively long sleep durations (over 8 hours) may be associated with increased health risks^[28]. Clinical observations show a high prevalence of sleep disorders in the elderly, such as insomnia and sleep-related breathing disorders, with a general prevalence of 30–48% and insomnia prevalence of 12–20%^[29]. These issues are often overlooked and may not receive early, effective treatment. Treatment measures include medication, cognitive-behavioral therapy (such as sleep restriction therapy and sleep hygiene education), and physical therapy (such as transcranial microcurrent stimulation). Future research should further explore sleep duration, comprehensively assess sleep characteristics, and investigate personalized interventions to provide more scientific evidence for the prevention and management of sarcopenia in the elderly.

This study only provided a simple classification of physical activity levels among the elderly (none, light, moderate, and vigorous) without detailed measurement of exercise duration, such as the number of exercise days per week and specific daily exercise times. Prior research has established that the absence of exercise habits is an independent risk factor for sarcopenia^[30]. Future studies should include comprehensive measurements of both the duration and intensity of physical activity to fully understand its impact on sarcopenia. Moreover, this study did not account for several significant predictive factors, such as dietary habits, nutritional status, and household income levels, potentially impacting the model's accuracy.

In our analysis, we excluded samples with missing values and did not employ multiple imputation methods, which could introduce estimation bias. During the predictive modeling process, participants with extreme height and weight values were also excluded. While this approach helps stabilize the model, it may restrict the generalizability of our findings. Additionally, we did not perform external validation of our model, limiting our ability to confirm its applicability across different regions and populations. Future research should address these limitations to enhance the robustness and applicability of sarcopenia predictive models.

This study provides a comprehensive analysis of longitudinal data from elderly populations in China, identifying and validating key independent risk factors for sarcopenia. These include advanced age in women, lower BMI, memory-related diseases, arthritis or rheumatism, shorter nighttime sleep duration, and lower educational attainment. The findings present a robust scientific foundation for early screening and targeted interventions, emphasizing the critical need for multidisciplinary management strategies in community healthcare settings.

Ethics Approval and Consent to Participate

This study was approved by the Ethics Committee of Peking University (IRB00001052-11015). Written informed consent was obtained from all participants.

Consent for Publication

Consent for publication was obtained from all participants involved in the study.

Availability of Data and Materials

The datasets generated and analyzed during the current study are available from the corresponding author on reasonable request.

Competing Interests

The authors declare that they have no competing interests.

Funding

No funding was received for this study.

Author Contributions:

Dr. Qianwei Sun: Conceptualization, Methodology, Writing - Original Draft, Project Administration

Dr. Lei Shen: Conceptualization, Data Collection, Writing - Review & Editing

Dr. Huamin Liu: Data Analysis

Dr. Zhangqun Lou: Resources, Supervision

Dr. Qi Kong: Supervision, Funding Acquisition, Review & Editing

Acknowledgements

We extend our gratitude to the China Health and Retirement Longitudinal Study (CHARLS) research team, the field team, and all respondents for their valuable time and efforts dedicated to the CHARLS project.

WHO.Ageing and health. World Health Organization. 2020. https://www.who.int/news-room/fact-sheets/detail/ageing-and-health
Petermann-Rocha F, Balntzi V, Gray SR, et al. Global prevalence of sarcopenia and severe sarcopenia: a systematic review and meta-analysis. J Cachexia Sarcopenia Muscle. 2022;13(1):86–99.
Chen LK, Woo J, Assantachai P, et al. Asian working group for sarcopenia: 2019 consensus Update on sarcopenia diagnosis and treatment. J Am Med Dir Assoc. 2020;21:300e307e302.
Sayer AA, Cruz-Jentoft A, et al. Sarcopenia definition, diagnosis and treatment: consensus is growing. Age Ageing. 2022;51(10):afac220.
Lin T, Liang R, Song Q, et al. Development and Validation of PRE-SARC (PREdiction of SARCopenia Risk in Community Older Adults) Sarcopenia Prediction Model. J Am Med Dir Assoc. 2024 Jul;5:105128.
Li Q, Cheng H, Cen W, et al. Development and validation of a predictive model for the risk of sarcopenia in the older adults in China. Eur J Med Res. 2024;29(1):278.
Seok M, Kim W, Kim J, et al. Machine Learning for Sarcopenia Prediction in the Elderly Using Socioeconomic, Infrastructure, and Quality-of-Life Data. Healthc (Basel). 2023;11(21):2881.
Wen X, Wang M, Jiang CM, et al. Anthropometric equation for estimation of appendicular skeletal muscle mass in Chinese adults. Asia Pac J Clin Nutr. 2011;20(4):551–6.
Sarcopenia prevalence and associated factors among older Chinese population. Findings from the China Health and Retirement Longitudinal Study. PLoS ONE. 2021;16(3):e0247617.
James C, Powell M, Seixas A, et al. Exploring the psychometric properties of the CES-D-10 and its practicality in detecting depressive symptomatology in 27 low- and middle-income countries. Int J Psychol. 2020;55(3):435–45.
Yu X, Hou L, Guo J, et al. Combined effect of osteoporosis and poor dynamic balance on the incidence of sarcopenia in elderly Chinese community suburban-dwelling individuals. J Nutr Health Aging. 2020;24:71–7.
White JP, Gao S, Puppa MJ, et al. Testosterone regulation of Akt/mTORC1/FoxO3a signaling in skeletal muscle. Mol Cell Endocrinol. 2013;365:174–86.
Chidi-Ogbolu N, Baar K, et al. Effect of Estrogen on Musculoskeletal Performance and Injury Risk. Front Physiol. 2019;9:1834.
Castro-Coronado J, Yasima-Vásquez G, Zapata-Lamana R, et al. Characteristics of resistance training-based programs in older adults with sarcopenia: Scoping review. Rev Esp Geriatr Gerontol. 2021;56:279–88.
Chen LK, Woo J, Assantachai P, et al. Asian working group for sarcopenia: 2019 consensus update on sarcopenia diagnosis and treatment. J Am Med Dir Assoc. 2020;21(3):300–e72.
Cruz-Jentoft AJ, Baeyens JP, Bauer JM, et al. Sarcopenia: European consensus on definition and diagnosis: Report of the European Working Group on Sarcopenia in Older People. Age Ageing. 2010;39(4):412–23.
Huang LT, Wang JH, et al. The Therapeutic Intervention of Sex Steroid Hormones for Sarcopenia. Front Med (Lausanne). 2021;8:739251.
Chen Y, Yu Y, et al. Tau and neuroinflammation in Alzheimer’s disease: interplay mechanisms and clinical translation. J Neuroinflammation. 2023;20:165.
Koeglsperger T, Rumpf SL, Schließer P, et al. Neuropathology of incidental Lewy body & prodromal Parkinson’s disease. Mol Neurodegeneration. 2023;18:32.
Yang J, Jiang F, Yang M, et al. Sarcopenia and nervous system disorders. J Neurol. 2022;269:5787–97.
Xu Hq S, Jp, Shen C, et al. Sarcopenia-related features and factors associated with low muscle mass, weak muscle strength, and reduced function in Chinese rural residents: a cross-sectional study. Arch Osteoporos. 2019;14:2.
Donini LM, Busetto L, Bischoff SC, et al. Definition and diagnostic criteria for sarcopenic obesity: ESPEN and EASO consensus statement. Clin Nutr. 2022;41(4):990–1000.
Bilski J, Pierzchalski P, Szczepanik M, et al. Multifactorial Mechanism of Sarcopenia and Sarcopenic Obesity. Role of Physical Exercise, Microbiota and Myokines. Cells. 2022;11(1):160.
Hein TR, Peterson L, Bartikoski BJ, et al. The effect of disease-modifying anti-rheumatic drugs on skeletal muscle mass in rheumatoid arthritis patients: a systematic review with meta-analysis. Arthritis Res Ther. 2022;24:171.
Mochizuki T, Yano K, Ikari K, et al. Sarcopenia in Japanese younger patients with rheumatoid arthritis: A cross-sectional study. Mod Rheumatol. 2021;31(2):504–5.
Liu K, Luo J, Chen Y, et al. Association between sarcopenia and sleep disorders: a cross-sectional population-based study. Front Nutr. 2024;11:1415743.
Chen L, Li Q, Huang X, et al. Association between sleep duration and possible sarcopenia in middle-aged and elderly Chinese individuals: evidence from the China health and retirement longitudinal study. BMC Geriatr. 2024;24:594.
Pourmotabbed A, Ghaedi E, Babaei A, et al. Sleep duration and sarcopenia risk: a systematic review and dose-response meta-analysis. Sleep Breath. 2020;24:1267–78.
Yue Z, Zhang Y, Cheng X, et al. Sleep quality among the elderly in 21st century Shandong province, China: a ten-year comparative study. Int J Environ Res Public Health. 2022;19(21):14296–308.
Hämäläinen O, Tirkkonen A, Savikangas T, et al. Low physical activity is a risk factor for sarcopenia: a cross-sectional analysis of two exercise trials on community-dwelling older adults. BMC Geriatr. 2024;24:212.

No competing interests reported.

Download PDF

Editorial decision: Revision requested
27 Oct, 2024
Reviews received at journal
13 Oct, 2024
Reviews received at journal
05 Oct, 2024
Reviewers agreed at journal
20 Sep, 2024
Reviews received at journal
17 Sep, 2024
Reviewers agreed at journal
17 Sep, 2024
Reviewers agreed at journal
16 Sep, 2024
Reviewers invited by journal
29 Aug, 2024
Editor invited by journal
09 Aug, 2024
Editor assigned by journal
09 Aug, 2024
Submission checks completed at journal
09 Aug, 2024
First submitted to journal
03 Aug, 2024

You are reading this latest preprint version

Development and Validation of a Predictive Model for Sarcopenia Risk in Older Chinese Adults Based on Key Factors

Status:

Version 1

Abstract

Background

Methods

Results

Conclusion

Figures

1. Introduction

2. Methods

2.1 Data Source

2.2 Data Extraction

2.3 Assessment of Symptoms of Sarcopenia

2.4 Selection of Predictive Factors

2.5 Assessment of Depression and Pain

2.6 Statistical Methods

2.6.1 Baseline Comparison

2.6.2 LASSO-Logistic Regression Model Development

2.6.3 XGBoost Model Development

2.6.4 Model Validation and Evaluation

3. Results

3.1 Baseline Characteristics Analysis

3.2 LASSO Regression Model for Sarcopenia Risk Factors

3.3 Multivariable Logistic Regression Analysis

3.4 Validation of Predictive Models

3.5 Model Calibration and Goodness-of-Fit Assessment

3.6 Clinical Validity Assessment

3.7 XGBoost Predictive Model and Feature Importance Analysis

3.8 Nonlinear Relationship Analysis Between BMI, Night Sleep Duration and Sarcopenia Risk

4. Discussion

5. Limitations

6. Conclusions

Declarations

References

Additional Declarations

Status:

Version 1