Objective: To identify Progressive Supranuclear Palsy (PSP) disease trajectories, defined as the pattern of diagnoses over time, prior to PSP diagnosis in electronic health records.

Background: PSP is an atypical Parkinsonian disorder characterized by gaze palsy, loss of balance, Parkinsonism and cognitive impairment. While prior studies have explored PSP risk factors, they have not evaluated the complex temporal relationships of these risk factors. Understanding these disease trajectories could improve diagnosis and subtypes of PSP. To address this gap, we developed a computational framework using machine learning and network analysis to identify common PSP trajectories.

Method: We analyzed patient data from the UC Health Data Warehouse’s, coded with ICD-10 diagnoses. Using the Fine-Gray subdistribution hazard model, we identified significant temporal risk factors and refined patient diagnostic trajectories by eliminating non-significantly associated temporal associations. Trajectories were clustered using Dynamic Time Warping and k-means clustering. We employed network analysis to determine prevalent PSP trajectories and compared patient demographics, PSP manifestations, and subtypes. Causal relationships within these trajectories were inferred using the Greedy Equivalence Search algorithm.

Results: Among 1,205 PSP patients, 247 were included in the final analysis, with 258 unique multi-step PSP disease trajectories identified. Three trajectory clusters emerged: 1) binocular movement disorder (e.g., cerebrovascular diseases → disorders of binocular movement → PSP; N=29), 2) Parkinson’s disease (e.g., joint disorder → Parkinson’s disease → PSP; N=72), and 3) neurodegenerative diseases (e.g., sleep disorders → degenerative nervous system diseases → PSP; N=168). Significant differences in demographics, symptoms, and PSP subtype distributions were observed across clusters, with 36.8% of trajectory edges identified as causal.

Conclusion: Our analysis revealed distinct PSP disease trajectories, integrating temporal and causal insights. These findings can improve PSP risk assessment and early diagnosis, with the potential for future targeted interventions.

PSP disease trajectory clusters

« Back to 2025 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/temporal-disease-patterns-prior-to-progressive-supranuclear-palsy-diagnosis/