Predicting Outcome in Critically Ill Patients using Artificial Intelligence Models

M. Santos, Á. Silva, A. Abelha, and L. Gomes (Portugal)


Critically Ill Patients, Mortality Predicting Models based on Intermediate Outcomes,Knowledge Discovery from Databases, Data Mining, Artificial Neural Networks.


Outcomes research in the Intensive Care Units (ICUs) is gaining momentum supported and motivated on recent social, medical and technological developments. A lot of models for mortality prediction have been proposed and adopted in the ICUs (e.g., APACHE, SAPS); however none of these models takes into account the intermediate outcomes (incidence and duration of Out of-Range Measurements of the monitored parameters). The existence of several databases containing clinical data collected from ICUs enabled the application of Data Mining techniques like the Artificial Neural Networks (ANNs) in a Knowledge Discovery from Databases (KDD) process to induce predictive models in a more flexible and efficient fashion than the classical approaches as the Logistic Regression. This paper argues in this direction presenting an experimental and comparative study on the use of ANNs in “outcome prediction” analysing the impact of intermediate outcomes (physiological impairment). The overall KDD process is dissected and some preliminary results are presented and discussed.

Important Links:

Go Back