Scholarlink

<p>This study aimed to develop and validate an explainable machine learning model for predicting sarcopenia in older adults using nationally representative data. Using the 2022–2023 Korea National Health and Nutrition Examination Survey (KNHANES; n = 2,721, ≥ 65 y), we constructed a data-driven framework integrating demographic, anthropometric, biochemical, and bioelectrical impedance analysis (BIA) variables. Sarcopenia was defined by the 2019 Asian Working Group for Sarcopenia (AWGS) criteria, requiring both low appendicular skeletal muscle mass and reduced handgrip strength. Five supervised algorithms—logistic regression, random forest, XGBoost, LightGBM, and CatBoost—were trained using tenfold cross-validation, with sex-specific models to address biological heterogeneity. The optimized CatBoost model showed the highest and most balanced performance. For men, accuracy = 0.912, sensitivity = 0.833, specificity = 0.920, F1 = 0.645, and AUC = 0.930; for women, accuracy = 0.895, sensitivity = 0.567, specificity = 0.932, F1 = 0.523, and AUC = 0.928. Feature importance analysis based on total feature importance gain indicated that BIA-derived water indices—intracellular water (ICW), extracellular water (ECW), and total body water (TBW)—together with appendicular lean soft tissue mass accounted for 72.4% (men) and 70.8% (women) of the overall model importance. SHapley Additive exPlanations (SHAP) identified ICW and phase angle (PhA) as top predictors. In men, the mean absolute SHAP value for ICW was 0.660 and for PhA was 0.282, while in women, the corresponding values were 0.750 for ICW and 0.536 for PhA, confirming their major contribution to model output. In conclusion, this CatBoost-based, SHAP-interpretable model effectively integrates BIA-derived physiological parameters with clinical data, providing accurate and sex-specific sarcopenia prediction for scalable early screening.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Construction and evaluation of a CatBoost-based machine learning model for sarcopenia prediction using bioelectrical impedance analysis (BIA) and handgrip strength (HGS) from a nationally representative dataset of Korean older adults

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月1日

查看全文
  • Sungjong Shin,
  • Jung Min Cho

摘要

This study aimed to develop and validate an explainable machine learning model for predicting sarcopenia in older adults using nationally representative data. Using the 2022–2023 Korea National Health and Nutrition Examination Survey (KNHANES; n = 2,721, ≥ 65 y), we constructed a data-driven framework integrating demographic, anthropometric, biochemical, and bioelectrical impedance analysis (BIA) variables. Sarcopenia was defined by the 2019 Asian Working Group for Sarcopenia (AWGS) criteria, requiring both low appendicular skeletal muscle mass and reduced handgrip strength. Five supervised algorithms—logistic regression, random forest, XGBoost, LightGBM, and CatBoost—were trained using tenfold cross-validation, with sex-specific models to address biological heterogeneity. The optimized CatBoost model showed the highest and most balanced performance. For men, accuracy = 0.912, sensitivity = 0.833, specificity = 0.920, F1 = 0.645, and AUC = 0.930; for women, accuracy = 0.895, sensitivity = 0.567, specificity = 0.932, F1 = 0.523, and AUC = 0.928. Feature importance analysis based on total feature importance gain indicated that BIA-derived water indices—intracellular water (ICW), extracellular water (ECW), and total body water (TBW)—together with appendicular lean soft tissue mass accounted for 72.4% (men) and 70.8% (women) of the overall model importance. SHapley Additive exPlanations (SHAP) identified ICW and phase angle (PhA) as top predictors. In men, the mean absolute SHAP value for ICW was 0.660 and for PhA was 0.282, while in women, the corresponding values were 0.750 for ICW and 0.536 for PhA, confirming their major contribution to model output. In conclusion, this CatBoost-based, SHAP-interpretable model effectively integrates BIA-derived physiological parameters with clinical data, providing accurate and sex-specific sarcopenia prediction for scalable early screening.