Background: Within Parkinson's there is a spectrum of clinical features at presentation which may represent sub-types of the disease. However there is no widely accepted consensus of how best to group patients.
Objective: Use a data-driven approach to unravel any heterogeneity in the Parkinson's phenotype in a well-characterised, population-based incidence cohort.
Methods: 769 consecutive patients, with mean disease duration of 1.3 years, were assessed using a broad range of motor, cognitive and non-motor metrics. Multiple imputation was carried out using the chained equations approach to deal with missing data. We used an exploratory and then a confirmatory factor analysis to determine suitable domains to include within our cluster analysis. K-means cluster analysis of the factor scores and all the variables not loading into a factor was used to determine phenotypic subgroups.
Results: Our factor analysis found three important factors that were characterised by: psychological well-being features; non-tremor motor features, such as posture and rigidity; and cognitive features. Our subsequent five cluster model identified groups characterised by (1) mild motor and non-motor disease (25.4%), (2) poor posture and cognition (23.3%), (3) severe tremor (20.8%), (4) poor psychological well-being, RBD and sleep (18.9%), and (5) severe motor and non-motor disease with poor psychological well-being (11.7%).
Conclusion: Our approach identified several Parkinson's phenotypic sub-groups driven by largely dopaminergic-resistant features (RBD, impaired cognition and posture, poor psychological well-being) that, in addition to dopaminergic-responsive motor features may be important for studying the aetiology, progression, and medication response of early Parkinson's.
Keywords: Cohort studies; Parkinson’s disease; cluster analysis; factor analysis.