Development and multi-center cross-setting validation of an explainable prediction model for sarcopenic obesity: a machine learning approach based on readily available clinical features

Abstract

Objectives: Sarcopenic obesity (SO), characterized by the coexistence of obesity and sarcopenia, is an increasingly prevalent condition in aging populations, associated with numerous adverse health outcomes. We aimed to identify and validate an explainable prediction model of SO using easily available clinical characteristics. Setting and participants: A preliminary cohort of 1,431 participants from three community regions in Ziyang city, China, was used for model development and internal validation. For external validation, we utilized data from 832 residents of multi-center nursing homes. Measurements: The diagnosis of SO was based on the European Society for Clinical Nutrition and Metabolism (ESPEN) and the European Association for the Study of Obesity (EASO) criteria. Five machine learning models (support vector machine, logistic regression, random forest, light gradient boosting machine, and extreme gradient boosting) were used to predict SO. The performance of these models was assessed by the area under the receiver operating characteristic curve (AUC). The SHapley Additive exPlanations (SHAP) approach was used for model interpretation. Results: After feature reduction, an 8-feature model demonstrated good predictive ability. Among the five models tested, the support vector machine (SVM) model performed best in SO prediction in both internal (AUC = 0.862) and external (AUC = 0.785) validation sets. The eight key predictors identified were BMI, gender, neck circumference, waist circumference, thigh circumference, time to full tandem standing, time to five-times sit-to-stand, and age. SHAP analysis revealed BMI and gender as the most influential predictors. To facilitate the utilization of the SVM model in clinical setting, we developed a web application (https://svcpredictapp.streamlit.app/). Conclusions: We developed an explainable machine learning model to predict SO in aging community and nursing populations. This model offers a novel, accessible, and interpretable approach to SO prediction with potential to enhance early detection and intervention strategies. Further studies are warranted to validate our model in diverse populations and evaluate its impact on patient outcomes when integrated into comprehensive geriatric assessments.