Use of artificial intelligence to identify cardiovascular compromise in a model of hemorrhagic shock

• Published in: Critical care medicine Vol. 32; no. 2; pp. 450 - 456
• Main Authors: Glass, Todd F, Knapp, Jason, Amburn, Philip, Clay, Bruce A, Kabrisky, Matt, Rogers, Steven K, Garcia, Victor F
• Format: Journal Article
• Language: English
• Published: Hagerstown, MD by the Society of Critical Care Medicine and Lippincott Williams & Wilkins 01-02-2004; Lippincott
• Subjects: Anesthesia. Intensive care medicine. Transfusions. Cell therapy and gene therapy; Animals; Artificial Intelligence; Biological and medical sciences; Cardiovascular Diseases - diagnosis; Cardiovascular Diseases - etiology; Disease Models, Animal; Female; Intensive care medicine; Male; Medical sciences; Prospective Studies; Shock, Hemorrhagic - complications; Swine; Shock; Intensive care; Cardiovascular disease; cardiovascular; Hemorrhage; Artificial intelligence; Trauma; Monitoring; Resuscitation
• ISSN: 0090-3493; 1530-0293
• DOI: 10.1097/01.CCM.0000109444.02324.AD

OBJECTIVETo determine whether a prototype artificial intelligence system can identify volume of hemorrhage in a porcine model of controlled hemorrhagic shock.
DESIGNProspective in vivo animal model of hemorrhagic shock
SETTINGResearch foundation animal surgical suite; computer laboratories of collaborating industry partner.
SUBJECTSNineteen, juvenile, 25- to 35-kg, male and female swine.
INTERVENTIONSAnesthetized animals were instrumented for arterial and systemic venous pressure monitoring and blood sampling, and a splenectomy was performed. Following a 1-hr stabilization period, animals were hemorrhaged in aliquots to 10, 20, 30, 35, 40, 45, and 50% of total blood volume with a 10-min recovery between each aliquot. Data were downloaded directly from a commercial monitoring system into a proprietary PC-based software package for analysis.
MEASUREMENTS AND MAIN RESULTSArterial and venous blood gas values, glucose, and cardiac output were collected at specified intervals. Electrocardiogram, electroencephalogram, mixed venous oxygen saturation, temperature (core and blood), mean arterial pressure, pulmonary artery pressure, central venous pressure, pulse oximetry, and end-tidal CO2 were continuously monitored and downloaded. Seventeen of 19 animals (89%) died as a direct result of hemorrhage. Stored data streams were analyzed by the prototype artificial intelligence system. For this project, the artificial intelligence system identified and compared three electrocardiographic features (R-R interval, QRS amplitude, and R-S interval) from each of nine unknown samples of the QRS complex. We found that the artificial intelligence system, trained on only three electrocardiographic features, identified hemorrhage volume with an average accuracy of 91% (95% confidence interval, 84–96%).
CONCLUSIONSThese experiments demonstrate that an artificial intelligence system, based solely on the analysis of QRS amplitude, R-R interval, and R-S interval of an electrocardiogram, is able to accurately identify hemorrhage volume in a porcine model of lethal hemorrhagic shock. We suggest that this technology may represent a noninvasive means of assessing the physiologic state during and immediately following hemorrhage. Point of care application of this technology may improve outcomes with earlier diagnosis and better titration of therapy of shock.