Sleep subtypes in Parkinson's disease: A polysomnography-based two-step cluster analysis

Introduction: Circadian and sleep disturbances are prevalent in Parkinson's disease (PD), with documented alterations in sleep architecture. While cluster analysis and person-centred statistical approaches have been utilized to identify clinical subtypes across various sleep disorders and PD presentations, research on polysomnography (PSG)-based sleep parameters' heterogeneity in PD is limited. This study aimed to explore PSG-derived sleep profiles in a cohort of PD patients. Method: Ninety-seven patients with idiopathic PD (mean age: 67.1 ± 7.9; 23.7% female) from the PD Outpatient Unit of the University Hospital of Rome Tor Vergata underwent ambulatory video-PSG. Sleep parameters, including total sleep time (TST), sleep efficiency (SE), sleep latency (SL), percentage of sleep time in non-REM (N1, N2, N3) and REM stages, REM latency, apnea-hypopnea index (AHI), periodic limb movement index (PLMI), and REM sleep without atonia (RSWA) were recorded. Disease severity was evaluated using the Hoehn and Yahr stage (H&Y), and motor symptoms were assessed with the Unified Parkinson's Disease Rating Scale-part III (UPDRS-III). Disease duration and levodopa daily equivalent dose (LEDD) were recorded. Two-step cluster analysis was performed considering sleep parameters as clustering variables. Differences in clinical variables among clusters were examined. Results: Three clusters emerged based on Bayesian Information Criteria: a fragmented sleep cluster (FragS; n = 27) with reduced SE, low REM, and high PLMI and AHI; a good sleep cluster (GS; n = 47) with high SE, low SL, high N2, and high REM; a high SL with preserved deep sleep cluster (SL + DS; n = 19) characterized by prolonged SL, high AHI, reduced REM, and high N3. Four participants remained unclassified, and a higher proportion of females was found in the SL + DS cluster. No differences in RSWA prevalence were observed across clusters. Analysis of variance showed higher UPDRS-III in SL + DS compared to GS, higher LEDD in FragS compared to GS and SL + DS, and longer disease duration in FragS and SL + DS compared to GS. No H&Y differences were found. Conclusion: Cluster analysis revealed three distinct sleep profiles in PD patients, each exhibiting different clinical characteristics possibly related to disease course and sleep alterations. Further research should explore the clinical utility of statistical clustering in classifying sleep disruptions in PD.