Rapid and accurate triage is essential when dealing with acutely injured persons. To that end, a number of trauma triage decision aids have been proposed. Unfortunately, mistriage rates for trauma patients remain high, due in part to aids that lack precision or discrimination, are not used, or are misused or misapplied. Improving the trauma triage process is an important problem because doing so improves trauma triage outcomes by optimizing access to and utilization of specialized resources as well as reducing morbidity and mortality. A promising method for improving this process is to integrate machine learning algorithms as a rapid and accurate clinical decision support technique. Machine learning techniques have been used to develop rules to diagnosis myocardial infarction and to provide rapid and accurate decision support in nonmedical settings. They have also been used to predict mortality following injury. Furthermore, using machine learning algorithms for clinical decision support employs highly accurate predictive models to counter the previously mentioned pitfalls of existing trauma decision aids in the following ways: (1) collecting relevant readily available clinical information, (2) choosing the most accurate predictive model for the situation at hand, and (3) utilizing computational speed and power to optimize predictive accuracy. The goal of this AREA proposal is to test the feasibility of machine learning-based predictive modeling techniques to support trauma triage. We plan to use well-established predictive model development and evaluation techniques to: (1) Compare the existing American College of Surgeons trauma triage decision aid to four different machine learning predictive modeling techniques; (2) Compare the performance of four different machine learning predictive modeling techniques with and without a novel measure of change in physiological status; and (3) Compare the performance of full machine learning models to those derived using information gathered as a byproduct of the workflow. Models will predict persons with severe injury or need for specialized trauma resources. Successful integration of machine learning algorithms is a first step toward developing an adaptive computer assisted trauma triage system that not only helps clinicians make better decisions but also facilitates rapid access to specialized resources. ? ? ?
