Year of Graduation
The models and methods of an assessment of various types of survival in practice of the children's oncologist
School of Applied Mathematics and Information Science
Abstract One of the main conditions for subsequent chemotherapy and adjustments of further clinical research is a comprehensive analysis of the results of treatment. In the area of health statistics this analysis is called as survival analysis.Nowadays there are few software tools that would allow researcher to quickly load the patient database and make a comprehensive analysis of efficiency of the conducted research. According to this there was developed a program with a graphical interface which allows researcher to read information and conduct analysis of survival by various methods.Applied scope in this paper is lymphoblastic leukemia, a type of cancer that occurs most frequently in children.The objective of this work are the patients subjected to chemotherapy in the treatment lymphoblastic leukemia.The subject of research is to evaluate the effectiveness of methods used for analysis of survival.Methodological basis of this work are methods of medical statistics and machine learning.The general goal of this paper is the realization of mathematical model in the form of a program that will allows to use methods of survival analysis on input databases. In addition, the paper presents an experiment - consideration of competing risks as a problem of classification.In the course of the experiment there was attempted application of machine learning techniques to predict events. There were considered three methods of classification in this work - support vector machine (SVM) method, naive Bayes method and the method of decision trees. High performance showed support vector method and trees method which gave a good accuracy of the prediction. Due to classifier of trees which has the property to build clear rules for the occurrence of the event, we got a good class of events with high precision. Using this class in the Cox regression model we bring a new, statistically significant risk factor "Age".