A Data-Driven Approach for Deriving Parkinson’s Disease Subtypes and Related Trajectories of Cognitive and Motor Function
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Content Notes
Abstract
Parkinson's disease (PD) is a progressive neurological illness that involves a variety of motor and non-motor features, with remarkable heterogeneity in clinical presentation and symptom progression. Previous studies have attempted to identify PD subtypes to better understand the clinical implications of this heterogeneity, though reliable subtypes remain unclear due to inconsistencies across studies, non-generalizable samples, and low subtype stability over time. This study aimed to identify PD subtypes based on motor, cognitive, psychiatric, and functional measures using K-means cluster analysis in a large PD sample (N=683) and explore subtype-specific prognosis. Results yielded a two-cluster solution (Cluster 1 N=457; 2 N=226) at baseline with adequate cluster quality. Cluster 1 exhibited more severe rest tremor but better functioning in all other measures (independence with activities of daily living, daytime sleepiness, depression, anxiety, general non-motor PD symptoms, and postural instability/gait difficulty). Ten half-sample replications of the cluster analysis model and Cohen's kappa revealed excellent model consistency in symptom trends and subjects' cluster membership. Comparison of additional baseline measures using analysis of covariance (ANCOVAs) found that Cluster 1 performed better in overall disease burden, quality of life, motor symptoms, PD medication side effects, mood, sleep dysfunction, compulsive behaviors, psychomotor processing speed, attention, and Aβ-42. ANCOVAs for annualized change showed Cluster 2 exhibiting greater disease burden over time according to a composite measure, with insignificant cluster differences for all other longitudinal analyses. Replication of the cluster analysis model at last visit revealed consistent cluster differences among model measures, but with suboptimal stability over time, as over 25% of subjects changed cluster membership from baseline. The current study utilized a large and heterogeneous PD sample and a statistically advanced subtyping approach to derive a two-cluster solution, with one group exhibiting relatively greater severity of rest tremor but better functioning in all other areas used to determine initial group differences, as noted above. Although several significant and clinically manful cluster differences were observed at baseline, longitudinal analyses revealed limited clinical and prognostic usefulness. Future PD subtyping efforts may consider increased variable diversity, novel statistical clustering methodologies, and examining clusters of symptom trajectories rather than baseline symptoms.