科学研究

Scientific Research

当前位置: 首页 > 科学研究 > 学术报告 > 专题报告 > 正文

Beyond Overall Diagnosis: A Longitudinal Multi-Domain Latent Structure Model for Domain-Specific Parkinson's Disease Progression

来源:
|
发布时间:2026-11-16
报告时间:
2026年11月21日(周六)上午10:30-11:30
报告地点:
6号楼C405
报告人:
康凯
报告人单位:
中山大学
邀请人:
晏挺
摘要:
Parkinson’s disease (PD) progresses heterogeneously, with motor, cognitive, and other non-motor symptoms often worsening at different times. Conventional models based on overall diagnoses or composite scores can obscure this asynchronous progression and limit individualized prognosis and care. Clinically, a domain-specific progression time marks when a patient meaningfully worsens in one symptom domain, helping clinicians identify when focused monitoring or care may be needed. We propose a longitudinal multi-domain latent structure model that groups PD-related markers into clinically distinct domains and represents progression within each domain through a shared latent disease state, while allowing latent disease states to differ across domains. The unobserved timing of the domain-specific progression is inferred from repeated clinical measurements whose distributions differ before and after progression, and baseline risk factors are related to this timing through a Cox proportional hazards model. We establish theoretical guarantees for recovering the underlying grouping of clinical markers into domains. Applied to longitudinal measurements of 30 retained MDS-UPDRS items from 849 PPMI participants, the method identifies five clinically interpretable symptom domains and individualized domain-specific progression trajectories. Faster progression is associated with male sex in selected domains and with smaller baseline putamen volume across domains, whereas education shows no clear association.

康凯(中山大学)