Can Artificial Intelligence Improve the Appropriate Use and Decrease the Misuse of REBOA?

Area of Science:

• Trauma surgery
• Medical artificial intelligence
• Clinical decision support

Background:

• Resuscitative endovascular balloon occlusion of the aorta (REBOA) use for noncompressible torso hemorrhage is debated.
• Hemodynamically unstable blunt trauma patients require optimized interventions.

Purpose of the Study:

• To apply a novel interpretable AI method, Optimal Policy Trees (OPTs), to enhance REBOA appropriateness.
• To reduce REBOA misuse in blunt trauma patients with hemorrhagic shock.

Main Methods:

• Trained and validated OPTs on a large trauma patient database (2010-2019 ACS-TQIP).
• Defined hemorrhagic shock by SBP ≤90 or ≥4 units transfused within 4 hours.
• Compared 24-hour mortality rates between OPT-prescribed REBOA and actual REBOA use.

Main Results:

• Analyzed 100,615 patients, with 803 receiving REBOA; REBOA patients had more severe injuries.
• Observed higher 24-hour mortality in REBOA (47%) vs. non-REBOA (21%) groups.
• OPTs demonstrated an 18% mortality reduction for REBOA and 0.8% for non-REBOA, by guiding appropriate use and avoiding misuse.

Conclusions:

• Interpretable AI models like OPTs can significantly improve mortality in unstable blunt trauma.
• This study highlights AI's potential beyond prediction, enabling intervention prescription to alter patient outcomes.
• Optimizing REBOA use through AI can decrease mortality in critical trauma care.