Discovery of generalizable TBI phenotypes using multivariate time-series clustering.

Area of Science:

• Neuroscience
• Data Science
• Medical Informatics

Background:

• Traumatic Brain Injury (TBI) is highly heterogeneous, complicating patient stratification and treatment.
• Existing TBI phenotyping methods often lack generalizability across different populations and settings.
• Identifying consistent TBI phenotypes is crucial for advancing personalized medicine.

Purpose of the Study:

• To develop and validate a generalizable TBI phenotyping approach using multivariate time-series clustering.
• To identify distinct, reproducible TBI phenotypes across diverse clinical datasets.
• To explore the relationship between identified phenotypes, clinical presentation, and demographic factors like age.

Main Methods:

• Employed a self-supervised learning-based approach for clustering multivariate time-series data with missing values (SLAC-Time).
• Applied SLAC-Time to two distinct datasets: the research-focused TRACK-TBI and the real-world MIMIC-IV.
• Validated the stability and generalizability of the clustering method and identified phenotypes across datasets.

Main Results:

• The SLAC-Time method demonstrated consistent optimal hyperparameters and cluster numbers across both TRACK-TBI and MIMIC-IV datasets.
• Identified three generalizable TBI phenotypes (α, β, and γ) with distinct clinical and temporal features.
• Phenotype α characterized mild TBI, β severe TBI with diverse presentations, and γ moderate TBI; age was a significant factor in mortality but not core phenotype characteristics.

Conclusions:

• The SLAC-Time approach provides a stable and generalizable method for TBI phenotyping.
• The three identified phenotypes offer a framework for understanding TBI heterogeneity.
• These findings support the potential for improved patient stratification and targeted therapeutic strategies in TBI care.