Using Artificial Intelligence to Develop an Adult Trauma Triage Decision Rule

Tuesday, 14 July 2009: 2:05 PM

Linda J. Scheetz, EdD, RN, FAEN
College of Nursing, New York University, New York, NY

Learning Objective 1: describe how artificial intelligence can be used to develop a clinical decision rule.

Learning Objective 2: recognize the development of a trauma triage decision rule as one step in the translation of evidence to the clinical care of injured persons.

Purpose: Globally, millions of traffic-related deaths and injuries occur each year, with a predicted 60% increase through 2020 (World Health Organization, 2004). Regardless of advances in trauma care, decision making at the crash scene to determine whether patients should be transported to a trauma center (TC) or a non-trauma center (NTC) hospital remains problematic. Many patients with unrecognized life-threatening injuries are transported to NTCs, which lack the resources to provide timely definitive care, resulting in potentially preventable deaths and disability for those who might have benefited from TC care.  Many clinical questions, including which injured patients should be transported to TCs and NTCs, can be addressed by the use of decision rules developed with artificial intelligence methods. Classification tree analysis is one such method for identifying predictors of life-threatening injury and creating a decision rule for triaging patients to TCs and NTCs. The purpose of this secondary analysis was to develop a trauma triage rule for adults.

Methods: This study used classification tree analysis to examine 74,626 NASS CDS database records of adults who were involved in vehicular crashes. Thirteen predictors were examined for their ability to classify (predict) patients having severe injury, an indicator of the need for TC care. Using the classification tree data, a triage decision rule was developed to identify patients with severe injuries who should be transported to a TC.

Results: Police-estimated injury severity, manner of collision, number of persons injured, and age were the best predictors of severe injury and became the decision points for a triage decision rule. Sensitivity and specificity of the rule in statistical modeling were 99.18% and 73.96%, respectively.

Conclusion: Artificial intelligence methods classified injury data and identified patients for TC transport. These classification data were used to develop a triage decision rule.  Prospective validation of the rule is needed.