Using a Semi-Automated Modeling Environment to Construct a Bayesian, Sepsis Diagnostic System
Peter J. Haug, Jeffrey P. Ferraro · 2016
We have developed an analytics environment, which uses clinical data from an Enterprise Data Warehouse to construct diagnostic models for use in clinical settings. We have focused on models based on Bayesian networks. The resulting system allows flexible development and testing of different Bayesian models based on 1) varying the data made available for use in the model and 2) manually and programmatically altering the models to improve their behavior. Here we illustrate the use of this system in exploring a group of models designed to identify sepsis patients in an emergency department setting. Varying the data elements used in the models and the structure of the models provides a range of diagnostic models whose operating behavior can be compared and contrasted.