A simulation study for cut points analysis in Logist regression and Cox proportional hazards model

• Main Authors: Chia-Chiung Liu, 劉佳峻
• Other Authors: Chung Chang
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/yn3698

碩士 === 國立中山大學 === 應用數學系研究所 === 107 === 　　In clinical practice, it is often necessary to segment the continuous variables for risk assessment, that is, to convert the continuous prognostic factors into categories to facilitate clinical judgment and interpretation. There are three problems to be solved in the study of estimating cut points. The first problem is to determine the optimal number of cut points. In the traditional methods, many of them have been developed to find one optimal cut point to categorize variables into two subgroup. However, in a lot of situations, finding more than one cut points is of interest. The second one is to find the location of optimal cut points. The last one is the statistical inferences after finding the optimal number and locations of cut points, including correcting the p-value, relative risks, powers, etc. 　　In previous theses(Tsai, Y.H.(2018), and Chiu, Y.C.(2018)), they proposed a new approach in both the logistic and Cox regression models, combining the cross-validation and Monte Carlo methods(CVM), to find the optimal number and locations of cut points. However, in their theses, the proposed method was not compared with other methods. In this thesis, we conducted simulation studies to compare the proposed CVM with three other methods, including naive approach(without any correction,NA), split-sample approach(SS), and cross-validation approach(CV). We compares the performance between these four methods in estimating the number and location s of cut points, relative risks, and powers in both univariate and multivariate analysis and for different sample sizes.