Automated Decision-Support Technologies for Prehospital Care of Trauma Casualties

The military has long been interested in advanced decision-support capabilities for combat casualty care in which an automated computer algorithm processes available data and, through artificial intelligence, offers caregivers accurate information about the state of the casualty. However, two major obstacles have impeded these capabilities. First, routine vital signs have been speculated to be insensitive to prehospital major traumatic pathology. Second, there are numerous potential sources of decision-support failure, and it is not possible to investigate and address such potential limitations and demonstrate utility within the confines of a research laboratory. To address these obstacles, we retrospectively mined our trauma database consisting of vital signs and attribute data from 898 patients, and employed various signal-processing, artificial intelligence, and knowledge engineering technologies to develop an automated decision-support system. Our system for major hemorrhage diagnosis yielded an area under the receiver operating characteristic curve of 0.85 (95% confidence interval 0.80-0.90), with an 85% sensitivity and a 73% specificity, when retrospectively applied to the testing set of basic vital-sign data. In parallel, we developed a novel plug-and-play software/hardware system (termed APPRAISE) for automated, real-time data collection and prospective testing of decision-support algorithms in prehospital, clinical settings. Through simulations, we verified APPRAISE’s real-time capability. Here, we summarize our technologies and findings in the development of an advanced medical system to reach the long-awaited goal of field deployment of automated decisionsupport tools for the triage and diagnosis of trauma casualties.