Reproducibility of data-driven Parkinson's disease subtypes for clinical research

Abstract

IntroductionPD subtype classification systems attempt to address heterogeneity in PD, a widely recognized feature of the disease with implications in prognosis and therapeutic development. There is no consensus on a valid PD subtype classification system, and its use in clinical research is sparse. Reproducibility has not been systematically assessed as a step for the validation of a PD subtype classification system. We aimed at assessing reproducibility of previously published data-driven PD subtype classification systems in a well-characterized cohort created for clinical research purposes, the Longitudinal and Biomarker Study in Parkinson's Disease (LABS-PD). MethodsWe identified all published studies of data-driven PD subtype classification systems and included those with variables that conceptually matched the variables available in LABS-PD. We reproduced the cluster analyses of the included studies in LABS-PD. Reproducibility was determined by a panel of experts using a modified Delphi consensus process. ResultsWe included eight studies of data-driven PD subtype classification systems and completed the replication in LABS-PD of the analyses conducted in each original study. After two iterations of the modified Delphi consensus process, no study was reproducible in LABS-PD. ConclusionsCurrently published data-driven PD subtype classification systems lack reproducibility in a well-characterized cohort of patients initially recruited for a clinical trial of a disease-modifying intervention. The results raise concerns about the utility of the widely-discussed concept of data-driven PD subtypes. This gap is a barrier for a meaningful use of PD subtypes and calls for the establishment of standards for the validation and use of these subtype classification systems.