Logistic regression analysis for Predicting Methicillin-resistant Staphylococcus Aureus (MRSA) in-hospital mortality

Abstract

Statistical models have been widely used in public health and made a difference in a wide range of applications. For example, they provide new ideas for efficient feature selection. This paper attempts to demonstrate how to apply regression-based methods to accurately predict in-hospital mortality of Methicillin-resistant Staphylococcus Aureus (MRSA) patients. Logistic regression is used to predict the in-hospital death. It is found that admission age, residency, solid tumor, hemic malignancy, COAD, Dementia, PLT, Lymphocyte, Urea, and ALP are the significant prognostic factors (P<0.1) for in-hospital survival. Using cross validation and random splitting and the prediction accuracy is around 85%. The future research direction is to strengthen the robustness of the predictive model. Possible direction is to make use of other data mining “blackbox” methods, such as k-NN and SVM. These models also need further validation on their performance and feature selection.