A Machine Learning-Based Visualized Prediction System for 2-Year Incident Sarcopenia Risk in Older Adults with Multimorbidity: A Nationally Representative Cohort Study (Preprint)

BACKGROUND Sarcopenia is closely associated with multimorbidity in older adults. However, the risk factors for sarcopenia in this population remain unclear. Additionally, current screening tools lack both sensitivity and dynamic risk evaluation. These limitations highlight the need for the development of accurate prediction models. OBJECTIVE This study aimed to develop and validate a machine learning (ML)-based visualized prediction system for 2-year incident sarcopenia risk in older adults with multimorbidity, addressing the limitations of current screening methods. METHODS Utilizing longitudinal data from the China Health and Retirement Longitudinal Study (CHARLS), we enrolled 1336 participants aged ≥60 years with multimorbidity and no baseline sarcopenia from the 2011-2013 cohort, randomly dividing them into training (70%) and testing (30%) sets. An additional 1678 participants from the 2013-2015 cohort served as an independent validation set. Sarcopenia was diagnosed according to the Asian Working Group for Sarcopenia (AWGS) 2019 criteria. Through comprehensive literature review, we initially selected 14 candidate predictors, subsequently identifying key predictive factors via LASSO regression. Eight ML models were trained and evaluated using the following metrics: Receiver Operating Characteristic-Area Under the Curve (ROC-AUC), Precision-Recall (PR) curves, calibration curves, and Decision Curve Analysis (DCA). SHapley Additive exPlanations (SHAP) values were employed to enhance model interpretability. Finally, we developed a web-based interactive prediction system for clinical implementation. RESULTS The Light Gradient Boosting Machine (LightGBM) model demonstrated superior performance, achieving ROC-AUC values of 0.997 (training), 0.744 (testing), and 0.782 (validation), with sensitivity ranging from 0.934 to 0.985. SHAP analysis identified BMI<23 kg/m², age>65 years, fall history, female, and activities of daily living (ADL) impairment as key predictors. The visualized prediction system provided dynamic risk assessment with intuitive visualizations, enabling personalized clinical decision-making. CONCLUSIONS This ML-based system provides accurate, interpretable, and dynamic sarcopenia risk prediction for older adults with multimorbidity, addressing limitations of current screening methods. Further validation in diverse populations is warranted to enhance clinical utility.