NEURAL NETWORKS AS A PROGNOSTIC TOOL IN C RITICALLY ILL PATIENTS

From 1. 11. 93 to 30. 3. 97, 1149 patients were prospectively studied during their ICU stay. Of them, 114 met the criteria of septic shock, with lethality of 47.3 %, A neural network was trained with datasets f rom 91 of these 114 patients. Testing the trained neural network with the remaining 23 patients, the following result was obtained: all 10 p atients dying from septic shock were correctly predicted; of 13 surviv ing patients, 12 were correctly identified (sensitivity 100 %; specifi ty 92.3 %).