Comparison of Quantile Regression and Parametric Regression Models in Heterogeneous Survival Data

Abstract

It was aimed to compare the performances of quantile regression and parametric regression models in predicting survival times in heterogeneous survival data. In this study, heterogeneous survival data sets were generated according to various sample sizes. Performances of methods were compared according to root mean square error and mean absolute error criteria after 1000 repetitive simulation. In addition, performances of quantile regression and parametric regression methods (exponential, Weibull, log-normal and log-logistic) on survival data were compared using German Breast Cancer Study data which is heterogeneous and related to 1984-1992 years. Models were built by performing parameter estimation of quantile regression and parametric regression methods on training set. In simulation and German Breast Cancer Study data set applications, the minimum error values were observed in quantile regression model. It was also determined that only the number of nodules had an effect on survival time. It is concluded that quantile regression method enables researchers to choose the model with minimum error among all models built according to each of the quantile values and is useful in heterogeneous survival data set. For this reason, it was concluded that quantile regression method is a method that estimates survival time with less error than parametric regression methods.