ELICITING PROBABILITIES FOR A BAYESIAN TRIAGE

Author: Afsaneh Barzi, MD, MS; Sarmad Sadegi, MD, MS

Primary Advisor: Craig W. Johnson, PhD

Committee Members:

Masters thesis, The University of Texas School of Health Information Sciences at Houston.

 
Objective: Development of a triage application using Bayesian networks as reasoning system. Design: Triage has become part of emergency medicine over time because the volume of patients using the emergency room has been increasing and some of these conditions have non-critical conditions. Ideally a ?triage expert? would be the best-qualified person to perform medical triage. However, when the demand for both economic and human resources exceeds availability, innovative solutions are required. It has been verified that the triage problem lends itself to an AI solution. Bayesian networks are an artificial intelligence modality that is superior to rule-based and decision tree based systems in that it handles uncertainty properly. However, implementing a successful medical Bayesian network requires a lot of numerical values that are not readily available from the formal medical resources. We have used specific techniques and methods to extract the numbers required for such Bayesian networks from domain experts (doctors) and the literature (textbooks, papers, etc). The methods for probability elicitation are reviewed in this paper. Conclusion: The application could help decrease the load work in emergency departments by different mechanisms, including self triage for out of the ED patients