比較不同特徵點比例風險模式於半競爭風險資料之平均餘命的估計表現

Abstract

關於病患在接受治療或發病後的平均餘命是臨床實務上重要的課題。在慢性疾病中，隨時間變化的生物標記可能會是重要的預後因子。本文目標是藉由病患們生物標誌的動態過程，提供病患們平均餘命的估計。針對死亡事件或設限事件會中止生物標誌的動態過程之半競爭風險資料，本研究考慮了不同特徵點比例風險模式，包含以在特徵時間點之生物標誌指標作為模式的共變數，以獲得平均餘命之估計。在研究中模擬的部分，是以均方差作為平均餘命估計表現的評估準則，並將所提估計方法應用於二筆實際資料。

The life expectancy of patients after receiving treatment or onset is an important issue in clinical practices. The information of time-varying biomarkers may be important prognostic factors of chronic diseases. The goal of this paper is to provide the estimated life expectancy of the patient based on the patient's own dynamic biomarker process information. The study considers several landmark proportional hazards models including such biomarkers at a specific landmark time as covariates. Then, estimates of patient life expectancy can be obtained based on the semi-competing risks data, in which the biomarker process can be stopped by death or censoring event, but not vice versa. In simulation study, the performance of estimated life expectancy is assessed via mean square errors. Our methods are applied to two real data sets for illustration.