A machine learning pipeline identifies non-motor symptoms as key contributors to sarcopenia in Parkinson's disease.

OBJECTIVES

Sarcopenia, characterized by the loss of skeletal muscle mass and physical function, is more prevalent among patients with Parkinson's disease (PD) and is linked to poor clinical outcomes. This study aimed to investigate clinical features associated with sarcopenia in PD, which was further validated using a two-stage modeling pipeline.

METHODS

Clinical and demographic data were collected and compared between the sarcopenia and non-sarcopenia groups. The dataset was split into a training and a hold-out test set (70:30). Following an initial variable screening in the training set based on clinical judgment and inter-variable correlations, a least absolute shrinkage and selection operator regression was applied to identify six key predictors. These variables were then used to train a logistic regression model selected based on the highest Area Under the Receiver Operating Characteristic (AUROC) in stratified cross-validation. Finally, the model was evaluated on the independent hold-out test set.

RESULTS

Patients with PD and sarcopenia exhibited longer disease duration, higher Hoehn and Yahr stages, and a greater non-motor burden. The final logistic regression model revealed that body mass index, body fat ratio, and non-motor features including cognitive decline, autonomic dysfunction, fatigue, and mood disturbance were the most influential variables. The model achieved an AUROC of 0.793, recall of 0.500, and specificity of 0.902.

CONCLUSION

Patients with PD and sarcopenia exhibit distinct clinical characteristics defined by both anthropometric and non-motor features. These findings suggest that a comprehensive clinical assessment may be warranted for the early identification of sarcopenia in PD.