以核平滑概似函數估計加乘法風險迴歸模型;Using Kernel Smoothed Likelihood Functions to Estimate Additive-Multiplicative Hazards Regression Models

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/86298

Title:	以核平滑概似函數估計加乘法風險迴歸模型;Using Kernel Smoothed Likelihood Functions to Estimate Additive-Multiplicative Hazards Regression Models
Authors:	林浩田;Lin, Hao-Tian
Contributors:	統計研究所
Keywords:	比例風險模型;加乘法風險模型;最大概似估計;proportional hazards model;additive-multiplicative hazards model;MLE
Date:	2021-07-27
Issue Date:	2021-12-07 12:29:01 (UTC+8)
Publisher:	國立中央大學
Abstract:	存活分析常使用Cox 比例風險模型，來解釋共變量與存活時間的關係；不過在某些資料中，並非所有的共變量皆符合比例風險。在比例風險假設未滿足的情況下，Cox 模型的估計結果會存在偏誤。為了處理此情形，以加乘法模型來配適，將不符比例風險的共變量放置於加法效應風險。然而，現有的計數方法Cox-Aalen 模型，僅能處理符合比例風險的乘法效應，不符合者的設定為加法效應，且需假設時間相關。本研究以完整概似函數方法的Cox-Aalen 模型，進行時間獨立的共變量效應分析。另外，對於不符合比例風險假設的情況，使用AFT-Aalen 模型，亦即乘法效應的準線風險基於加速失效風險。使用核平滑方法建構AFT-Aalen 模型的概似函數，並藉此以Nelder-Mead 方法進行估計。以模擬資料對模型做效力的評估，並將其用於皮膚黑色素瘤資料的效應估計。;The Cox proportional hazards model is popular for survival analysis to explain the relations between survival time and covariates. However, if the proportional hazard (PH) assumption is not satisfied by some covariates of data, the estimation of PH model would be biased. To solve these situation, we fit the additive-multiplicative hazards model, which puts those covariates not satisfying proportional hazards into additive terms. The existing methods for additive-multiplicative hazards (AMH) models like Cox-Aalen model by counting process can only describe the covariates satisfying proportional hazards in multiplicative submodels, and the remains in additive terms must be assumed to have time-varying regression coefficient. The study use the full likelihood Cox-Aalen model to analyse the time-independent effects of covariates. On the other hand, we use accelerated failure time to construct AFT-Aalen model when PH assumption fails. We build likelihood functions for AFT-Aalen model by kernel smoothing method and estimate the parameters by Nelder-Mead method. We evaluate the performance of our AMH models through simulation study and apply them to Melanoma data.
Appears in Collections:	[Graduate Institute of Statistics] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	121	View/Open

社群 sharing

Loading...