Design of an XML-based alerting system prototype using mortality prediction models

Abstract

This thesis presents an XML based clinical alerting system prototype that has the ability to accurately detect and warn medical decision makers of critical events or complications prior to their occurrence. The alerting system makes use of existing MIRG ANN infant mortality prediction models to develop 'alert generation rule' algorithms using the four most highly ranked variables in each of the mortality prediction models. Prior to the use of this system, physicians did not have a way of knowing which ANN mortality model was better at predicting infant mortality as these models are similar and there is no currently known method to establish which one is better except by generally computing sensitivity and specificity. This alerting system is a first attempt at finding an alternate way to assess how each model performs in detecting the outcome 'death' and can be used by MIRG researchers in selecting the best model.