# Cluster Analysis of Motor Symptoms in Early-Diagnosed Parkinson’s Disease Patients

## TL;DR

This study uses cluster analysis to identify subtypes of Parkinson’s disease patients based on motor symptoms, revealing seven distinct groups.

## Contribution

A decision tree cluster analysis method was applied to categorical symptom data, automatically determining the number of clusters.

## Key findings

- Seven PD patient subtypes were identified based on motor symptom presence.
- The largest cluster contained half of the patients and showed three common motor symptoms: bradykinesia, rigidity, and tremors.
- The method reduced estimation errors by automatically determining cluster numbers.

## Abstract

Parkinson’s disease (PD) is a common movement disorder affecting adults. People diagnosed with PD can have a multitude of physical (motor) symptoms, including tremors, and rigidness, and psychological (non-motor) symptoms, including anxiety and depression. These symptoms dramatically affect daily living activities, including dressing oneself, preparing meals, and speaking and writing. Background/Objectives: To determine the symptom similarities and differences among PD patients, a method referred to as cluster analysis can be applied to patient data. This method can separate patients who differ by symptom presence while grouping patients with disease similarities. Previous PD cluster analysis studies provided patient groups that were defined by their age and disease duration—both numerical values—and excluded categorical values, such as patient gender, family history of the disease, and symptom presence. In addition, patient age and disease duration were limited in range in previous studies, providing a patient group that was too similar to divide into distinct clusters. Methods: This study utilized a decision tree cluster analysis method applied to categorical symptom data from PD patients. The applied cluster method automatically determines the number of clusters, reducing estimation errors, as many cluster analysis methods require the end user to estimate the number of clusters prior to applying cluster analysis. A post analysis of additional categorical and numerical variables was conducted, and this provided a means to describe the PD patient clusters in terms of gender, family history of PD, median age, disease duration, and symptom presence. The patient dataset utilized was accessed from the Parkinson’s Progression Markers Initiative (PPMI) website. Results and Conclusions: The cluster analysis results provided a means to describe seven PD patient subtypes based on motor symptom presence, with the largest PD patient cluster containing half of the patient sample, and these individuals had three of the motor symptoms present: bradykinesia, rigidity, and tremors.