Predicting mortality of cancer patients using artificial intelligence, patient data and blood tests

Several authors have shown that hematological parameters can be used to detect poor prognosis in patients with cancer. Thus, such features could be used in artificial intelligence (AI-based) models to predict mortality among these patients. This work aimed to develop and compare several AI-based models to predict the prognosis (death vs. survival) of cancer in patients using blood tests and patient data as inputs. At total, 908 cancer patients were assisted in a prospective study. Four artificial intelligence models were compared: artificial neural networks (ANN), supporting vector machines (SVM), decision trees and neuro-fuzzy networks. Also, four different input strategies were tested, considering the use of 49, 45, 22 and 14 inputs. The results of this study showed that the ANN and the SVM presented the best results, using 45 inputs. The ANN was the best model since it presented better statistical values for the positive (death) and negative (survival) classes. The use of blood parameters as inputs for AI-based models could be used to predict death in patients with cancer, and this methodology can be expanded to other diseases.