Abstract
BACKGROUND: Progressive supranuclear palsy (PSP) is a rare neurodegenerative disorder characterized by parkinsonism and impairments in balance, language, and cognition. As an atypical parkinsonism, PSP progresses rapidly, lacks effective treatments, and poses significant caregiving burdens. While prior studies have identified risk factors, they often fail to capture the complex temporal dynamics of co-occurring conditions. OBJECTIVES: To identify common PSP disease trajectories using a computational framework combining machine learning and network analysis. METHODS: We analyzed ICD-10 codes from the University of California Health Data Warehouse to identify significant temporal risk factors and determine diagnostic trajectories. Multi-step trajectories were clustered using dynamic time warping and clustering. Network analysis identified common trajectory clusters, which were compared for patient demographics, symptoms, and PSP progression. Causal relationships were inferred through association tests and compared to control groups. RESULTS: From an initial cohort of 1205 PSP patients, 247 with multi-step trajectories were included. The final dataset yielded 258 unique multi-step trajectories. A total of 258 unique trajectories were identified, grouped into three clusters: binocular movement disorder (N = 29), Parkinson's disease (N = 72), and other neurodegenerative diseases (N = 168). Demographics, symptoms, and disease progression differed significantly across clusters. Causal analysis revealed that 36.8% of trajectory connections were significant. Patients with multi-step risk factor trajectories exhibited higher PSP risk compared to those with single or no risk factors. CONCLUSIONS: Distinct PSP trajectories were identified, offering insights into temporal and causal relationships. These findings can improve PSP risk assessment, early diagnosis, and personalized management, enhancing patient care and guiding future research.