Identifying TBI Physiological States by Clustering Multivariate Clinical Time-Series Data

Area of Science:

• Biomedical Informatics
• Data Science
• Clinical Research

Background:

• Accurate physiological state determination from multivariate time-series data with missing values is critical for acute conditions like Traumatic Brain Injury (TBI).
• Traditional methods using imputation or aggregation can lead to information loss and biased clinical interpretations.
• Novel approaches are needed to preserve data integrity and provide meaningful representations of patient states.

Purpose of the Study:

• To introduce and evaluate the SLAC-Time algorithm for analyzing multivariate time-series data in critical care.
• To identify distinct physiological states in patients with Traumatic Brain Injury (TBI) using a self-supervision-based clustering approach.
• To assess the impact of clinical events and interventions on patient state transitions.

Main Methods:

• Application of the SLAC-Time algorithm, a self-supervision-based method that avoids data imputation or aggregation.
• Clustering of a large research dataset to identify distinct physiological states.
• Validation of identified states using clustering evaluation metrics and clinical domain expert input.

Main Results:

• Identification of three distinct physiological states associated with Traumatic Brain Injury (TBI).
• Characterization of specific feature profiles for each identified TBI state.
• Discovery of relationships between clinical events, interventions, and patient state transitions.

Conclusions:

• SLAC-Time offers a robust method for analyzing complex physiological time-series data, maintaining data integrity.
• The identified TBI states provide a more nuanced understanding of patient conditions and their evolution.
• This approach facilitates better clinical decision-making and personalized treatment strategies for acute conditions.