Construction and validation of a predictive model for the risk of sarcopenia in middle-aged and elderly patients with knee osteoarthritis based on machine learning
10.3760/cma.j.cn211501-20250208-00296
- VernacularTitle:基于机器学习的中老年膝骨关节炎患者肌少症风险预测模型的构建与验证
- Author:
Guangyuan DONG
1
;
Jihua LI
;
Yun LU
;
Nanyan LI
;
Qingzhao LIANG
;
Lei SHI
Author Information
1. 南方医科大学第三附属医院护理部，广州 510515
- Publication Type:Journal Article
- Keywords:
Knee osteoarthritis;
Sarcopenia;
Machine learning;
Risk prediction model
- From:
Chinese Journal of Practical Nursing
2025;41(26):2023-2032
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To construct a prediction model for the risk of sarcopenia in middle-aged and elderly patients with knee osteoarthritis (KOA) based on machine learning, and to provide a basis for carrying out the prevention of sarcopenia in patients with KOA.Methods:Clinical data of KOA patients from three tertiary hospitals in Guangdong Province were collected between December 2023 and September 2024 using a convenience sampling method. The data were randomly split into training and test sets at an 8:2 ratio, with the occurrence of sarcopenia as the outcome variable. Risk prediction models for sarcopenia were constructed using eight machine learning algorithms: logistic regression, K-nearest neighbors, support vector machine, decision tree, neural network, random forest, gradient boosting machine (GBM), and eXtreme gradient boosting. Model performance was evaluated based on metrics including the area under the receiver operating characteristic curve (AUC), accuracy, precision, sensitivity, specificity, and F1 score. The optimal model was selected, and feature importance was visualized using the Shapley Additive exPlanations (SHAP) method.Results:Data from 640 KOA patients were analyzed, 143 males and 497 females, (67.51± 7.72) years, with 136 cases (21.25%) developing sarcopenia. All eight prediction models showed high AUC values, with the GBM model demonstrating the best performance. Its metrics included an AUC of 0.926 (95% CI 0.874 - 0.965), accuracy of 0.852, precision of 0.611, sensitivity of 0.815, specificity of 0.861, and F1 score of 0.698. SHAP analysis identified body mass index, calf circumference, body fat percentage, WOMAC score, and age as the most important predictive features. Conclusions:The GBM-based risk prediction model for sarcopenia in middle- aged and elderly KOA patients demonstrated optimal performance, enabling healthcare professionals to accurately and promptly identify high-risk groups among these patients and to develop effective, evidence-based intervention strategies.
