Identifying trigger cues for hospital blood transfusions based on ensemble of machine learning methods

Area of Science:

• Emergency Medicine
• Trauma Surgery
• Critical Care

Background:

• Traumatic shock is a leading cause of preventable death, with most fatalities occurring within six hours of hospital arrival.
• Prehospital interventions, including early blood transfusion, are crucial for improving survival rates in trauma patients.
• Optimizing transfusion triggers in the prehospital setting is essential for managing hemorrhagic shock and occult shock.

Purpose of the Study:

• To identify factors available to prehospital clinicians that predict early in-hospital blood transfusion requirements.
• To develop a simple algorithm for prehospital transfusion decisions, particularly for patients with occult shock.

Main Methods:

• Analysis of trauma patient data from a single critical care transport service to a Level I trauma center (2012-2019).
• Utilized logistic regression, Fast and Frugal Trees (FFTs), and Bayesian analysis to identify predictors of early transfusion.
• Evaluated 13 clinically relevant factors for their association with transfusion needs.

Main Results:

• Out of 2,157 patients, 9.60% required blood transfusion within four hours of admission.
• A Fast and Frugal Tree model incorporated Systolic Blood Pressure (SBP), prehospital lactate, Shock Index, and Abbreviated Injury Scale (AIS) of the chest.
• Prehospital lactate concentration was a significant predictor (OR=2.31) in Bayesian analysis.

Conclusions:

• A simple, clinically relevant prehospital algorithm was developed using a combination of statistical and machine learning methods.
• This algorithm aids in identifying patients likely to require transfusion within four hours of hospital arrival.
• The developed tool supports timely and appropriate prehospital transfusion decisions for trauma patients.