Abstract
Heterogenous clinical presentations of Parkinson's disease have aroused several attempts in its subtyping for the purpose of strategic implementation of treatment in order to maximise therapeutic effects. Apart from a priori classifications based purely on motor features, cluster analysis studies have achieved little success in receiving widespread adoption. A priori classifications demonstrate that their chosen factors, whether it be age or certain motor symptoms, do have an influence on subtypes. However, the cluster analysis approach is able to integrate these factors and other clinical features to produce subtypes. Differences in inclusion criteria from datasets, in variable selection and in methodology between cluster analysis studies have made it difficult to compare the subtypes. This has impeded such subtypes from clinical applications. This review analysed existing subtypes of Parkinson's disease, and suggested that future research should aim to discover subtypes that are robustly replicable across multiple datasets rather than focussing on one dataset at a time. Hopefully, through clinical applicable subtyping of Parkinson's disease would lead to translation of these subtypes into research and clinical use.