Data-driven subtype classification of patients with early-stage multiple system atrophy

Abstract

Introduction: Patients with multiple system atrophy (MSA) are conventionally identified as having MSA-P (prominent parkinsonism) or MSA-C (prominent cerebellar ataxia) based on their predominant motor manifestations. The objective of the present study was to conduct latent class analysis (LCA) of various motor and nonmotor symptoms in early MSA to characterize data-driven subgroups.

Methods: Sixty-one probable or possible MSA patients with disease durations of 3 years or less were included prospectively. LCAs were performed to identify similar clinical subgroups giving even weights to a wide range of MSA motor and nonmotor features. We ran latent models of up to 6 class solutions; the overall model fit was evaluated based on the parsimony of the derived classes, the fit indices, and clinical interpretability.

Results: The LCA outcome supported categorization of at least three subgroups of patients with early MSA: the largest class 1, labeled "moderate parkinsonism + extensive dysautonomia", included approximately half of our study patients and showed marked autonomic dysfunction with a burden of parkinsonism. The two other classes, class 2 "predominant parkinsonism + limited dysautonomia" and class 3 "predominant cerebellar symptoms + limited dysautonomia", showed marked core motor features (parkinsonism or cerebellar symptoms) with generally mild dysautonomia.

Conclusions: To our knowledge, this is the first data-driven identification of disease subtypes covering various symptom constellations in early MSA (<3 years from motor symptom onset). The present LCA result did not replicate the conventional motor classification and supported the heterogeneity within MSA-P and MSA-C subtypes.

Keywords: Classification; Latent class analysis; Magnetic resonance imaging; Multiple system atrophy; Subtype.