Utilizando método de aprendizado de máquina para análises de sobrevida de pacientes oncológicos

Abstract

The cancer is a complex and multifactorial disease, described as disordered growth of cells from different types of tissues, caused by genetic alteration. Factors that will contribute to the development of the disease vary in each individual and in each type of cancer, and survival analysis is one of the ways to assess the probability of survival according to the characteristic of a given factor. Thus, by using machine learning techniques, it is possible to predict the probability based on all factors associated to each patient, so that the risk pattern of an entire patient set can be identified optimally. Considering the local scenario, this study aimed to analyze the risk pattern of the oncological patients attended by the Hospital Mossoroense de Estudos e Combate ao Câncer during the period of 2006 to 2016, using machine learning e data mining methods for survival analysis. In present study, was used randomForestSRC library, developed in R language. The algorithm was used to identify the most relevant variables for survival prediction, and also the error rate and performance of the method were evaluated. Considering the 11 groups analyzed, 7 presented error rate below 20% and 4 presented Brier Score below 0.1.