Abstract:
|
This presentation will discuss a project to provide situational awareness to U.S. Dept. of Defense epidemiologists monitoring the health of military personnel and dependents. Data sources of increasing specificity and timeliness are becoming available. The challenge is to combine these sources for a coherent, current view of population health. Available data sources included administrative clinical encounter records including free-text chief complaints, prescription orders, laboratory test orders, and others. The fusion capability combines data streams and algorithm outputs using Bayesian Networks to derive the level of concern for possible outbreaks or other events, and to clarify the basis for these concerns, including drill-down to selected patient details. The presentation will explain the fusion approach and demonstrate the validation of this capability based on 3.75 years of data, 30 documented outbreaks, and 10-fold data-driven cross-validation. This approach avoids separate examination of evidence sources and encourages investigation by reducing nuisance alerting and directing attention to events likely connected with serious illness, with readily accessible detail.
|