Comparative method in PCA and PLS COX regression to solve multicollinearity

Cox regression is one of the method in survival analysis which is used to explain the correlation and influence between the individual failure at a certain time with one or more predictor variables. There is an assumption in Cox regression that is non multicollinierity. If non multicollinierity assumption not fulfill it means that using Cox regression is no longer appropriate. Therefore some alternatif method developed to solve this multicollinerity problem such Partial Least Square-Cox regression (PLS-Cox) [3], and Principal Component Analysis-Cox regression (PCA-Cox) [4]. The objective of this research is to observe different influence of predictor variables to individual failure probability and compare which one of the best method between PLS-Cox and PCA-Cox in survival data analysis that contains multicollinierity. This research using survival data of the probability of myocardial recovery in patients with CVA hospitalize at Rumah Sakit Haji Surabaya [14]. Subjects studied as many as 69 patients. Response variable (Y) is the length of patient hospitalization (days), with predictor variables were age (X1), gender (X2, 0 = female, 1 = male), diagnose (X3, 4 categories, complications of diabetes, complications of hypertension, more than one disease complications, and without complications), LDL cholesterol (X4), HDL cholesterol (X5), and Blood Pressure (X6).Based on the adjusted value of cross validation (Q²_adjusted), it can be concluded that modeling using PLS-Cox regression on data containing multicollinearity can provide a good prediction capability and relatively better.