Sarcopenia risk assessment among physically inactive middle-aged and older adults: interpretable machine-learning models in UK and US cohorts
Zhenhao Lin, Young-Je Sim, Chuang Zhang, Kunpeng Wu, Zhonghua Sun, Yuwen ShangGuan, Litao YanAbstract
Aim:
To develop and assess interpretable machine-learning models for sarcopenia risk assessment among physically inactive middle-aged and older adults using two large population-based datasets from the UK and the US.
Background:
Physical inactivity represents a major modifiable risk factor for sarcopenia in aging populations, yet prediction models specifically targeting this high-risk subgroup remain limited. This study developed and evaluated interpretable machine-learning models for sarcopenia risk stratification in physically inactive middle-aged and older adults using large-scale UK and US population-based data.
Methods:
We analyzed physically inactive participants from the English Longitudinal Study of Ageing (ELSA, 2012; n = 1,146) and the US National Health and Nutrition Examination Survey (NHANES, 1999–2006 and 2011–2018; n = 2,733). Sarcopenia and physical inactivity were defined using cohort-specific measurements and cutoffs. Within each cohort, six machine-learning algorithms were trained using 70/30 training–testing splits, Synthetic Minority Oversampling Technique to address class imbalance, and five-fold cross-validation for hyperparameter optimization. Model performance was evaluated using area under the curve, accuracy, precision, recall, and F1 scores. Shapley Additive Explanations quantified predictor contributions, and stratified analyses explored heterogeneity by age and body-composition strata.
Findings:
Random forest demonstrated optimal performance across both cohorts (area under the curve: 0.817 and 0.801; accuracy: 83.8% and 83.1%). Shapley Additive Explanations analysis revealed waist-to-height ratio as the dominant predictor, followed by age, frailty score, and poverty-income ratio. Stratified analyses showed heterogeneous risk patterns across age groups and body-composition categories.