Proposta e avaliação de um modelo de prognóstico para pacientes com septicemia

Abstract

The technology used in medicine evolves all the time, and with that new solutions emerge. With artificial intelligence and machine learning applied to medical problems, the ability to anticipate diagnoses and prognoses can be considered, due to the data processing capacity that a computer can perform. No machine learning problem was applied, a prognostic end of a disease known in medicine, SEPSIS, which causes a generalized infection. Sepsis is an effective prevention of measures not correctly identified, significantly compromising the destruction of Organs following organs, and can cause the destruction of multiple organs, so it is necessary to adopt measures to be able to contain the advance of pre-health. Therefore, the methodology of this work is to present a model of supervision for the prognosis of patients who present symptoms of sepsis. Finally, a proposal integrated the elaboration of learning models in order to obtain a prediction from an ICU prediction data program in the monitoring of inpatients from PhysioNet S in the monitoring of inpatients from ICU research data - eICU Collaborative Database through data used in the research database. With that, with the models vector models according to the solution, K-nearest neighbors (KNN) and Support machines. The best result of receiver operation (ROC) was according to the method of each model (ROC), respectively. And for the prediction of patient time, according to the best scenario, the models obtained AUC 0.85, 0.64 and 0.32, respectively.