Work

A machine learning-based risk prediction model for sarcopenia among community-dwelling older adults

Year: 2026

Type: article

Abstract: Against the backdrop of a globally aging population, sarcopenia has become an important public health concern among community-dwelling older adults. However, early identification of sarcopenia in primary care settings that lack specialized diagnostic equipment remains challenging. This cross-sectional study included 3,135 community-dwelling older adults (aged ≥60 years) from Xinjiang, China, enrolled between April 2023 and August 2024. Sarcopenia was diagnosed according to the AWGS-2019 criteria. LASSO regression was used for feature selection, and nine supervised learning algorithms—namely, random forest, LightGBM, XGBoost, logistic regression, K-nearest neighbors, support vector machine (SVM), artificial neural network, decision tree, and elastic net—were trained on the training set (70%, n = 2,195). Model performance was evaluated on the independent test set (30%, n = 940) using the area under the receiver operating characteristic curve (AUC), calibration curves, the Brier score, and the DeLong test. The SHAP method was used to interpret the model outputs. The prevalence of sarcopenia was 12.47% (391/3,135). LASSO regression selected 21 predictive variables; after excluding intracellular water, multicollinearity was eliminated (all variance inflation factors < 3). In the test set, elastic net and SVM achieved the best performance, each with an AUC of 0.929. SHAP analysis revealed that body mass index (BMI), percentage body fat (PBF), and age were the top three contributing predictors; higher BMI was a protective factor, whereas higher PBF and age increased the risk. The elastic net and support vector machine models based on routine clinical variables can accurately predict sarcopenia risk in community-dwelling older adults, demonstrating good discrimination and calibration. The models are easy to use, require no specialized equipment, and are suitable for resource-limited community settings, facilitating early identification and targeted intervention of high-risk individuals.

Source: Scientific Reports

DOI: 10.1038/s41598-026-59836-6