Parkinson’s Disease Recognition based on Sleep Metrics from Actigraphy and Sleep Diaries

Abstract

Parkinson’s disease is accompanied by sleep disorders in most cases. Therefore patients with Parkinson’s disease could be identified according to proper sleep metrics. The study aims to train a classifier and identify proper sleep metrics, that could distinguish patients with Parkinson’s disease from subjects in control group based on data from actigraphy and sleep diaries. Study sample consisted of 23 patients with probable Parkinson’s disease and 71 control subjects resulting in 654 nights of actigraphy and sleep diary data, with 26 unique features per night. XGBoost classifier was trained to distinguish the groups, scoring 80% accuracy and 52% F1 on test data. Actigraphy based parameters targeted on wake analysis during sleep were marked as most important. The study provided classifier and obtained the most important parameters to identify patients with Parkinson’s disease based on actigraphy and sleep diary data.