二元無序間隔時間資料之相關性分析

Abstract

在許多長期的生物醫學研究當中，受試者可能會隨著時間的推移經歷多次不同類型的事件，且這些事件之間並不存在特定的發生順序。例如：在癌症研究當中，局部復發與遠端轉移復發就是兩種無序二元復發事件。此外，同一類型的兩個連續事件之間的間隔時間通常是實務上感興趣的主題，並且每個類型的事件在追蹤時間內可能發生不止一次的復發，因此這類型的數據也稱之為無序二元復發間隔時間資料。本文研究的目的是估計無序二元復發間隔時間資料下，不同類型事件的間隔時間之間的相關性。交叉比是目前進行二元存活時間相關性的常見相關性測度，因為它可以表示為風險比。其估計方法是可透過結合合成概似函數與倒數機率設限權重來消弭間隔時間資料下的誘導相依設限問題。蒙地卡羅模擬則是用來評估本文提出方法的表現。

In many long-term biomedical studies, subjects may experience several events of different types over time, where the different types of events occur in no chronological order. For example, local and distant recurrences are two unordered bivariate events in cancer studies. The gap time between the two successive events of the same type is often of interest in practices and each type of event may occur more than once during the follow-up period. This type of data is called the unordered bivariate recurrent gap times data set. The purpose of this study is to estimate the association between gap times of bivariate events under the unordered bivariate recurrent gap times data. The cross ratio is a popular association measurement for bivariate survival times since it can be expressed into a hazard ratio. The estimation method of the cross or hazard ratio, which is the association measure between gap times of bivariate events is developed by using the composite likelihood approach together with the inverse probability of censoring weighting to eliminate the problem of induced dependent censoring. The Monte Carlo simulation study is conducted to show the performance of the proposed estimation.