Table 2.

Comparison of methodology of recent PD subtyping studies using cluster analysis.

Methodological steps	Liu 2011	Van Rooden 2011	Fereshtehnejad 2015	Erro 2016	Fereshtehnejad 2017	Mu 2017
Data pre-processing	Standardized scores	z scores	z scores	NS	Normative values	Standardised scores
Clustering algorithm	K-means	Model-based	2-step	K-means	Hierarchical	K-means & Hierarchical
Basis of the determination of the number of clusters	NS	NS	Bayesian information criterion	Calinski-Harabasz pseudo-F value	Estimate, Hartigan’s rule	Various e.g. Gap Statistic and the 1-standard-error method
Cluster validation on independent sample	No	Yes	No	No	No	No
Evaluation of discriminative variables	No	Discriminant analysis	No	No	Principal component analysis	No
Follow up period, years± mean	N/A	N/A	4.5	N/A	2.73 ± 0.78	No
Post hoc analysis of variables not included in the cluster analysis	Yes, motor phenotype consistency	No	Yes, disease progression	Yes, 123[I]-FP-CIT binding values	Yes, CSF and imaging biomarkers, and disease progression	No