Machine Learning Approach to Predict the Survival Time of Childhood Acute Lymphoblastic Leukemia Patients

Abstract

Survival time prediction (prognosis) is the task of predicting the length of time that a patient will survive. Survival time prediction is a difficult task due to the complex relationships that exist among biological, genetic and environmental factors. Medical practitioners make predictions about the survival time using their previous experiences and observations. The prognosis for different practitioners is often inconsistent. An accurate survival time prediction model can help in treatment scheduling, care of cancer patients and increase the quality of healthcare. In this paper, we present the results of an exploratory study of the survival time prediction of Childhood Acute Lymphoblastic Leukemia patients using a dataset of 512 patients provided by Maharagama National Cancer Institute, Sri Lanka. We investigated three machine learning techniques including multiple linear regression, regression trees and support vector regression. The performances of the models were evaluated using the Relative Absolute Error and Concordance Index in combination with 5-fold cross-validation. Our experiments show that the multiple linear regression and the support vector regression are effective: each predictor achieved an average cross-validated RAE less than 0.3, which is significantly lower than values reported in the previous studies. We also use our prediction models to classify each patient into short survivor versus long survivor where the classification boundary is the average survival time of the entire population. All the prediction modes achieved more than 70% classification accuracy.