群集資料之復發間隔時間存活函數無母數估計

Abstract

在追蹤研究中，針對某一感興趣事件記錄並蒐集個體重複此事件的相關資料稱為復發事件資料，而隨著各種研究的需求或資料的特性，將研究個體劃分於各群集中為常見處理個體間關聯性的分析方式，例如家族資料與多中心的研究需要分別以家庭和各個中心為群集，而結合復發事件資料與群集資料稱為群集復發資料，在此資料結構下兩相鄰復發事件的間隔時間通常是感興趣的分析對象，且同時存在相同群集內不同個體的相關性與同一個體的不同復發間隔時間彼此的相關性，本文嘗試引入兩個脆弱變數分別解釋資料中兩種不同的關聯性，並以Wang 和 Chang(1999)所提出的加權風險集合法以及Hoffman et al.(2001)提出的群集內重抽法分別處理這些關聯性，針對群集復發資料結合兩法成為四種無母數的統計方法，估計其邊際間隔時間的存活函數，最後藉由模擬呈現此四種方法與過去僅考慮部分相關性或不考慮任何相關性的方法比較其估計結果。

Clustered recurrent event data are a common data structure in longitudinal follow-up studies, in which more than one event of the same type within an individual may be observed during follow-up. The main outcome of interest is the gap time between two successive recurrent events. For clustered recurrent event data, there are two potential sources of correlation, within-individual correlation between recurrent gap times and within-cluster correlation between individual. The aim of this study is to estimate the marginal survival function of recurrent gap time by assuming that two latent frailty variables result in the within-individual and within-cluster correlations and recurrent gap times for an individual in the same cluster are identically independently distributed, We propose four non-parametric methods to estimate the marginal survival function of gap time by combining the weighted risk set method for recurrent gap time data (Wang and Chang, 1999) and within cluster resampling method for clustered survival data (Hoffman et al., 2001), to deal with the within subject correlation and within cluster correlation. A series of simulation studies are conducted to investigate the statistical properties and performance of the proposed methods and to compare with naive methods which ignore within-subject correlation, within-cluster correlation, or both.