Classifying Admission Characteristics of TBI Patients Using K-means Clustering

Session Number

Project ID: MEDH 32

Advisor(s)

Ali Mansour, University of Chicago

Discipline

Medical and Health Sciences

Start Date

17-4-2024 9:40 AM

End Date

17-4-2024 9:55 AM

Abstract

Traumatic Brain Injury (TBI) remains a significant health concern that often results in long-term cognitive impairments, coma, or even mortality. Current classification methods are primarily reliant on the Glasgow Coma Scale (GCS) and face limitations in representing the complexity and variability of TBIs. This study utilizes unsupervised learning through a k-means algorithm to cluster TBI patients at Beth Israel Deaconess Medical Center between 2008 and 2019 based on admission characteristics, in order to enhance prognosis and treatment approaches. Analysis of clusters uncovers diverse patient profiles which reveal correlations between age, GCS scores, and post-hospital outcomes. Clusters characterized by extreme age or GCS scores demonstrate varied mortality rates suggesting the ineffectiveness of GCS as a sole classifier. Younger age emerged as a highly expected yet crucial predictor of favorable outcomes. The study establishes the potential of clustering algorithms in patient stratification, offering insights for prognosis and post-hospital outcome prediction. However, there are limitations stemming from a lack of generalizability due to a single-hospital dataset. More validation across diverse datasets is required for broader clinical applicability in critical care settings for TBI patients.

Share

COinS
 
Apr 17th, 9:40 AM Apr 17th, 9:55 AM

Classifying Admission Characteristics of TBI Patients Using K-means Clustering

Traumatic Brain Injury (TBI) remains a significant health concern that often results in long-term cognitive impairments, coma, or even mortality. Current classification methods are primarily reliant on the Glasgow Coma Scale (GCS) and face limitations in representing the complexity and variability of TBIs. This study utilizes unsupervised learning through a k-means algorithm to cluster TBI patients at Beth Israel Deaconess Medical Center between 2008 and 2019 based on admission characteristics, in order to enhance prognosis and treatment approaches. Analysis of clusters uncovers diverse patient profiles which reveal correlations between age, GCS scores, and post-hospital outcomes. Clusters characterized by extreme age or GCS scores demonstrate varied mortality rates suggesting the ineffectiveness of GCS as a sole classifier. Younger age emerged as a highly expected yet crucial predictor of favorable outcomes. The study establishes the potential of clustering algorithms in patient stratification, offering insights for prognosis and post-hospital outcome prediction. However, there are limitations stemming from a lack of generalizability due to a single-hospital dataset. More validation across diverse datasets is required for broader clinical applicability in critical care settings for TBI patients.