Neural Network Medical Decision Algorithms for Pre-Hospital Trauma Care. Phase 1
Abstract
This SBIR Phase I research project is concerned with the problem of civilian and military trauma management, whose paramount issues include: (1) obtaining knowledge about the physiological condition of the injured patient (e.g., injury severity assessment and survival likelihood prediction); and (2) making intelligent use of that information for pragmatic decisional purposes (e.g., triage). The emphasis of our research effort is on assessing the ability of polynomial neural network (PNN) methods to improve on conventional trauma scoring systems and other modeling approaches, such as logistic regression. Using several real-world civilian trauma registry databases, we demonstrated: (1) that PNN models can provide significant improvement over existing pre-hospital and ex post scoring systems, such as T-RTS, TRISS, and ASCOT, in terms of the specificity-sensitivity characteristics of mortality prediction; (2) the ability to discriminate accurately among three or more classes of patients (e.g., RSD, AMBER, and GREEN triage categories); (3) the ability to compensate for missing input variables while achieving results not significantly different from those obtained using models that did not rely on such inputs; and (4) the ability to obtain superior performance through time-series modeling of available patient data.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 01, 1996
- Accession Number
- ADB216627
Entities
People
- B. E. Eugene Jr.