Variable Selection via Nonconcave Penalized Likelihood in High Dimensional Medical Problems

Abstract

Variable selection is fundamental to high-dimensional statistical modelling in diverse fields of sciences. Specially in health studies, many potential factors are introduced to determine an outcome variable. In our study, different statistical methods are applied to analyze trauma annual data, collected by 30 General Hospitals in Greece. The dataset consists of 6334 observations and at most 131 factors that include demographic, transport and intrahospital data. The statistical methods employed in this work were the nonconcave penalized likelihood methods, SCAD, LASSO, and Hard, the generalized linear logistic regression, and the best subset variable selection, used to detect possible risk factors of death. A variety of different statistical models are considered, with respect to the combinations of factors and the number of observations. A comparative survey reveals differences between results and execution times of each method. The performed analysis reveals several distinct advantages of the nonconcave penalized likelihood methods over the traditional model selection techniques.