Objective:

To explore heterogeneity and identify data-driven subtypes in early Parkinson’s disease (PD) derived from patients’ perspectives using self-reported most bothersome symptoms.

Background:

PD is a heterogeneous disorder with diverse motor and non-motor features across individuals. Traditional subtyping based on clinician-rated or biomarker measures may overlook aspects most meaningful to people with PD (PwP). The PD patient report of problems (PD-PROP), a large-scale collection of verbatim patient reports from the Fox Insight study, was curated using a human-in-the-loop natural language processing and machine-learning framework, enabling exploration of PD subtypes grounded in PwP’s own experiences.

Design/Methods:

Data were obtained from the Fox Insight longitudinal study via its online platform, the Fox DEN (Data Exploration Network), in July 2025, consisting of the PD-PROP, baseline demographics, and self-reported outcome variables namely the PDQ-8, UPDRS-II, NMS-QUEST, PDAQ-15, EQ-5D-5L, and a composite health score. We included 10,464 individuals (mean age 65.0±10.1 years; 54.5% male) with self-reported PD, PD diagnosis for <3 years and valid baseline PD-PROP. We classified 67 curated symptoms into 30 motor and non-motor domains by consensus. For exploratory subtyping, we conducted K-means clustering. 

Results:

Among 10,464 participants with PD <3 years, a four-cluster solution emerged based on the silhouette statistic: Cluster 1 (excessive daytime sleepiness/fatigue), Cluster 2 (urinary, cognitive, and fluctuation symptoms), Cluster 3 (motor-dominant), and Cluster 4 (pain/dystonia). Cluster 3 comprised most participants (67.9%), Cluster 4 members were significantly younger, and Cluster 2 showed the highest rate of probable REM sleep behavior disorder (RBD) (62.2%) and lowest self-perceived overall health, highest PDQ-8 (poorest quality of life), highest UPDRS-II (highest motor burden), and lowest PDAQ-15 (worst cognitive-functional abilities).

Conclusions:

Patient-reported symptom profiles from the PD-PROP identified four distinct PD subtypes, highlighting the value of patient-reported data to evaluate clinical heterogeneity early in clinical disease.