Artificial neural network in prediction of the outcome of critically ill patients with perforative peritonitis
Aim. The aim of the present paper is to compare the use ofArtificial Neural Network (ANN) to APACHE II, MOF, TISS-28 and MPI scoring system in prediction of peritonitis-relateddeath in patients with perforative peritonitis. Patients andmethods. A prospective study was performed of 145 patientswith perforative peritonitis, treated in the Surgical IntensiveCare Unit. The main outcome of this study was peritonitisrelateddeath. The Levenberg-Marquardt method was usedfor training, and 16 variables for entrance into the ArtificialNeural Network. Sensitivity and specificity of scoring systemsare graphically shown for the different values of cut-off pointswith the receiver-operating characteristic curve (ROC) curve.Results. We tested 92 cases in a network and found that theAPACHE II system predicted the lowest number of wrongassessments with a score of 12, with all the other scoringsystems predicting 19 wrong assessments. The area underthe curve for the first postoperative day was 0.87 for TISS-28score, 0.86 for APACHE II score, 0.83 for MOF and 0.72 forMPI score. The highest rate of correlation between the observedand the expected mortality rate was in the APACHEII system. This demonstrated that TISS-28 and APACHE IIare significantly better than other systems (P<0.01). In addition,this discriminatory ability was also retained on the thirdand seventh postoperative days. Conclusion. APACHE II issuperior in the prediction of patient outcome to the ArtificialNeural Network and other tested scoring systems.