相依截切資料的分析方法探討

Abstract

在長期追蹤的研究當中，研究者若要收取起始事件發生在研究開始前的個體資料，經常會利用橫截面的方式來收集資料，以這種收集方式所收得的資料稱為截切資料 (truncated data)。針對隨機截切模型，現有與存活函數相關的統計推論以及迴歸分析都假設存活時間與截切時間在可觀測範圍內獨立，但是這樣的部分獨立假設在許多的情況下是不合理的。在本篇文章中，我們提出了給定相依截切時存活函數的條件累積涉險函數之無母數估計量。最後以 Hyde (1977) 中所提到的老人社區醫療照護資料檔為實例來說明本文所提出的方法。

In longitudinal studies, researchers often collect data from a cross-sectional sampling scheme, in which subjects had an initial event prior to recruitment. This kind of data collected under a cross-sectional sampling criterion is called truncated data. In the random truncation model, existing statistical inferences about the survival function and regression analysis assume that the survival time and truncation time are independent in the observable region. However, the assumption of quasi-independence is unrealistic in many situations. In this study, we provide a nonparametric estimate of the conditional cumulative hazard function of survival time given truncation time under informative truncation. Finally, we use the Channing House data (Hyde, 1977) to illustrate our proposed method.