Physiologic phenotypes in blunt thoracic aortic injury: implications for risk stratification and surgical...

Area of Science:

• Trauma Surgery
• Cardiovascular Surgery
• Machine Learning in Medicine

Background:

• Blunt thoracic aortic injury (BTAI) is a rare but often fatal condition.
• Current management strategies face challenges due to diverse patient presentations.
• Injury severity alone may not fully predict patient risk and outcomes.

Purpose of the Study:

• To apply machine learning to identify distinct physiologic phenotypes in BTAI patients.
• To analyze the outcomes associated with each identified phenotype.
• To utilize explainable AI (XAI) to understand the physiologic drivers of these phenotypes.

Main Methods:

• Analysis of 1,375 patients from the Aortic Trauma Foundation registry.
• K-means clustering (k=3) to define three distinct physiologic phenotypes based on eleven variables.
• XGBoost classification with SHapley Additive exPlanations (SHAP) for feature importance and validation.

Main Results:

• Three phenotypes identified: 'Stable' (n=431), 'Shock' (n=398), and 'Neurologically Compromised' (n=254).
• Mortality rates were significantly higher in 'Shock' (18.9%) and 'Neurologically Compromised' (18.1%) compared to 'Stable' (7.2%).
• Physiologic markers like blood pressure, lactate, and heart rate were key discriminators, with similar treatment patterns across phenotypes.

Conclusions:

• Machine learning successfully identified distinct BTAI phenotypes with varying mortality risks.
• Physiologic status, particularly perfusion and metabolic markers, is crucial for BTAI phenotyping and prognostication.
• These findings support the use of physiologic phenotyping to improve BTAI risk stratification and clinical decision-making.