多重路徑模式的邏輯斯迴歸分析

Abstract

許多生物醫學方面之研究，經常需要分析數個事件 (multiple endpoints) 的資料。而處理這種屬於多重路徑 (multi path) 的問題，常遭遇的難度之一是如何處理那些因設限而無法辨識其路徑的觀測值。 本論文首先從半競爭風險和長期倖存者兩種觀念來研究此問題並建構多重路徑的資料模式。進一步將以logistic regression探討解釋變數對於路徑選擇的影響，並分別在上述的兩種資料模式下提出合理的參數估計方法。

Many interesting biomedical applications involve analysis of multiple endpoints. Statistical inference based on multi-path is especially challenging due to the problem of non-identifiability. In this article, we concern about two concepts of semi-competing risks and long term survivors, then build the multi-path models. We use logistic regression to discuss the inference of path choosing. Our approach provide reasonable parameter estimator under two kinds of data structure.