治癒模式之文獻回顧

Abstract

傳統存活分析假設感興趣的事件一定會發生，然而當事件由“死亡”推廣到如“發病”、“復發”… 的情形時，這個假設便不盡合理。文獻修正的方向是承認存在所謂的“免疫者”永遠也不會發生此事件，這類方法統稱為“治癒模式”。本論文回顧治癒模式的部份文獻，將其分成三個方向討論。最常見的治癒模式對免疫者無明確的定義；有一類模式以存活是否超過某固定時間做為免疫的指標；另一方向則以競爭風險發生順序為指標，後兩者對免疫均有明確的定義。所考慮的推論方法，則有迴歸分析、非迴歸分析以及無母數分析三種，論文的重點放在迴歸分析方法的回顧。

In classical survival analysis, it is implicitly assumed that the event of interest will occur eventually. However, this assumption may not implausible when there exist a proportion of subjects who will never experience the event despite of long-term follow-up. In the thesis, we will review the literature on cure models. Three types of cure models are considered according to the definition of “immune” or “cure”. In the first type, there is no explicit definition for the immune. In other words, cured individuals are always mixed with susceptible but censored ones. The second class defines cure as being able to survive beyond a pre-specified time period. For the third type, whether a subject is immune is determined by the order of competing events. Inference methods for cure models include parametric, semi-parametric and nonparametric analysis. The focus here is on semi-parametric regression analysis.