Identifying key physiological and clinical factors for traumatic brain injury patient management using network analysis and machine learning

Abstract

In the intensive care unit (ICU), managing traumatic brain injury (TBI) patients presents significant challenges due to the dynamic interaction between physiological and clinical markers. This study aims to uncover these subtle interconnections and identify the key ICU markers for the timely care of TBI patients using advanced machine-learning techniques. We combined correlation-based network analysis and graph neural network (GNN) techniques to explore relationships among electrocardiography (ECG) features, vital signs, pathology test results, Glasgow Coma Scale (GCS) scores, and demographics from 29 TBI patients admitted to the Gold Coast University Hospital (GCUH). Our findings highlighted that the final GCS index strongly correlated with arterial and diastolic blood pressure variations, patient demographics such as gender and age, and certain heart rate variability (HRV) features. Variability in diastolic blood pressure, GCS, and pNN50 (an HRV measure) demonstrated strong associations with several other physiological and clinical markers during the first 12 hours post-ICU admission. HRV features and variability in physiological signals during the first 12 hours in the ICU are important factors in assessing the severity of TBI patients.