Machine Learning Unveils System-Level Signatures of Integrated Motor–Non-Motor Dysfunction in Parkinson’s Disease during Telerehabilitation

In-home monitoring and rehabilitation are promising approaches for efficient and effective care of people with Parkinson’s Disease (PwPD). However, their adoption is still minimal, few evidences come from real in-home studies, and the integration of telemonitoring and telerehabilitation functions is absent. This study presents a novel system with remote smartwatch-based monitoring and rehabilitation integrated functions. Furthermore, an original machine learning approach is proposed to predict the clinical improvement, stability, or worsening of motor symptoms severity after a 3-month at-home rehabilitation period. Ninety-five PwPD were enrolled. An exploratory data analysis based on linear regression was applied to the collected smartwatch data to drive the extraction of meaningful features about the trend of physiological variables. Then, a time-series forest classification approach was applied to predict motor outcome from extracted features. The performance was evaluated against the clinical evaluation of the Unified Parkinson’s Disease Rating Scale before and after the 3-month period. Performance metrics reached 63% for accuracy, F1-score and recall, and 62% precision in nested cross-validation. Shapley analysis revealed an interesting agreement between model classification strategy and known clinical markers of overall health conditions and physical activity. Features related to heart rate and stress emerged as the most discriminative, showing on average decreased values of stress and heart rate in improved participants and increased values of the same features in worsened ones. Overall, these findings support the potential of combining remote monitoring and machine learning approaches to better understand the complex interaction between physiological signals, physical activity, and clinical status in PwPD, but the identified limitations suggest it should be considered a starting point for further research and improvements.