A Comparison between the Use of Cox Regression and the Use of Partial Least Squares-Cox Regression to Predict the Survival of Kidney-Transplant Patients

摘要

The aim of this research is to model survival data from kidney-transplant patients using the partial least squares (PLS)-Cox regression, which can both meet and not meet the no-multicollinearity assumption. The secondary data were obtained from research entitled "Factors affecting the survival of kidney-transplant patients". The research subjects comprised 250 patients. The predictor variables consisted of: age (Xi), sex (X2); two categories, prior hemodialysis duration (X3), diabetes (X4); two categories, prior transplantation number (X_5), number of blood transfusions (X_6), discrepancy score (X7), use of antilymphocyte globulin(ALG) (X_8); two categories, while the response variable was patient survival time (in months). Partial least squares regression is a model that connects the predictor variables X and the response variable y and it initially aims to determine the relationship between them. Results of the above analyses suggest that the survival of kidney transplant recipients ranged from 0 to 55 months, with 62% of the patients surviving until they received treatment that lasted for 55 months. The PLS-Cox regression analysis results revealed that patients' age and the use of ALG significantly affected the survival time of patients. The factor of patients' age (Xi) in the PLS-Cox regression model merely affected the failure probability by 1.201. This indicates that the probability of dying for elderly patients with a kidney transplant is 1.152 times higher than that for younger patients.