Presenter of 1 Presentation
CLUSTERING OF LONG-TERM FUNCTIONAL RECOVERY PATTERN OF STROKE PATIENTS: THE KOSCO STUDY
Abstract
Background and Aims
This study aimed to cluster multifaceted functional recovery patterns of first-ever stroke patients using the unsupervised learning algorithm of artificial intelligence (AI).
Methods
This study was an interim analysis of the Korean Stroke Cohort for Functioning and Rehabilitation (KOSCO) dataset, which is a prospective multicenter cohort. Multifaceted functional assessments were performed five or six times repeatedly from seven days to 24 months of stroke onset. Out of 7,858 patients enrolled, 4,389 ischemic stroke (IS) and 1,146 hemorrhagic stroke (HS) patients who completed the functional assessment at 24 months after onset were included in this analysis. In k-means clustering analysis, multifaceted functional assessment scores, demographic features, and clinical information were used as input variables. The optimal cluster k number was determined by finding the highest k with at least a silhouette score of 0.2. All statistical analysis was implemented using R (version 4.0.3).
Results
IS patients were clustered into 11 groups while HS patients were clustered into 13 groups. In both IS and HS, each cluster showed distinct clinical characteristics and functional recovery pattern which included the overall high function group, the overall low function group, the gradual improvement group, the late declining group, etc. Each cluster were clearly distinguished by age, initial stroke severity, and lesion location, etc.
Conclusions
Early identification and accurate prediction of long-term functional outcome will be useful for developing customized treatment for these patients.
Acknowledgment
Supported by a grant from the Korea Centers for Disease Control and Prevention (2019E-320202) and the NRF grant, provided by the Korean government (NRF-2020R1A2C3010304).