Objective: To develop a classification and staging system for multiple system atrophy (MSA) using data-driven disease progression modeling.

Background: MSA exhibits varying degrees of degeneration in striatonigral and olivopontocerebellar systems. Progression of the MSA pathology remains poorly understood. Recent advances in data-driven disease progression modeling have enabled probabilistic reconstruction of continuous disease progression patterns that have yet to be applied to MSA.

Method: In 72 autopsy-confirmed MSA cases from the Mayo Clinic brain bank, we semi-quantitatively assessed neuronal loss (NL) in the putamen, substantia nigra, pontine nuclei, inferior olivary nucleus, and cerebellar Purkinje cells using a 4-point scale. We applied an unsupervised learning algorithm for disease classification and staging (Subtype and Stage Inference – SuStaIn) on NL scores and correlated the findings with clinical information.

Results: In the two-subtyping task, 46 cases were assigned to subtype 1 and 24 to subtype 2. Most patients (89%) with SuStaIn subtype 1 had parkinsonism-predominant MSA, while most patients (63%) with SuStaIn subtype 2 had cerebellar-predominant MSA (P < 0.0001). When cases were grouped into three subtypes, 37 cases were classified as subtype 1 (S1), 24 as subtype 2 (S2), and 9 as subtype 3 (S3). NL started in the striatonigral system in S1 and in the olivopontocerebellar system in S2, while S3 had NL early in both systems. Predominant symptoms were parkinsonism in 95% of S1 and 78% of S3, and cerebellar symptoms in 67% of S2 (P < 0.0001). S3 had more frequent autonomic dysfunction as an initial symptom (S1 – 19%, S2 – 42%, S3 – 67%) compared to S1 (P = 0.03), less cognitive impairment (41%, 58%, and 0% respectively) compared to both S1 and S2 (P = 0.04 and P=0.01), and a trend of more frequent early falls (41%, 58%, and 89% respectively) compared to S1 (P = 0.07). S3 also had a shorter disease duration compared to S1 (5±4 vs. 8±4 years; P = 0.009). Additionally, a positive correlation was observed between the SuStaIn stage and disease duration (r = 0.32, P = 0.006).

Conclusion: The present study identified a unique subtype of MSA that had both striatonigral and olivopontocerebellar pathology earlier and had a poor prognosis. The SuStaIn algorithm may contribute to novel classifications of MSA subtypes.

« Back to 2024 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/unsupervised-learning-approach-for-pathology-based-subtyping-and-staging-multiple-system-atrophy/