Turning performance in neurodegenerative diseases: classification and longitudinal assessment

Neurodegenerative diseases and aging are often associated with difficulties in movement, such as impairments in turning, leading to functional disability and reduced quality of life. Turning outcomes have the potential to serve as biomarkers, facilitating diagnosis and disease severity assessment. In this context, a comprehensive analysis of turning performance in neurodegenerative diseases has been conducted. A large dataset, obtained by merging three studies conducted by Newcastle University (ICICLE-GAIT, GAITDEM) and the University of Bologna (InCHIANTI), has been analyzed. Turning-related digital mobility outcomes were extracted using a dedicated turning algorithm applied to inertial data from a sensor placed on the lower back during 180° turns in a controlled setting. Two analyses were conducted: a classification analysis aimed at distinguishing dementia subtypes (Alzheimer's Disease, AD, and Lewy Bodies Disease, LBD) from healthy controls (HC), and a longitudinal analysis to assess the differences in the physiological development of turning outcomes over time in relation to Parkinson's disease (PD) progression. The classification analysis was successfully completed using both statistical tools and machine learning models, achieving an F1 score higher than 80% in each task (HC vs AD, HC vs LBD, and AD vs LBD). The longitudinal analysis was performed using a Linear Mixed Effects Model (LMEM), observing how PD patients use different turning strategies compared to older adults (from the InCHIANTI study) and age-matched healthy participants (from the ICICLE-GAIT study). This wearable sensor and turning outcomes-oriented approach met the hypothesis of this work, correctly classifying neurodegenerative diseases and highlighting the differences between PD patients and healthy participants in the temporal evolution of turning performance.