多重競爭風險復發事件資料之統計分析

Abstract

長期追蹤資料的研究中，經常搜集到多重競爭風險且具復發可能性的資料型態。此外研究對象亦可能會因為研究時間結束、發生死亡或其他研究本身不感興趣的事件發生而退出研究。疾病的複雜機制與資料的不完整性，使此類資料的統計推論問題充滿挑戰。 在論文中，我們以統整的架構探討上述資料結構，可發現文獻常見的資料型態為此架構的特例。我們回顧了如何利用“脆弱模式”做為處理關聯性的模式建構原則，並提出數個資料生成演算法。“累積發生函數”常被選用在描述競爭風險型態資料，於是我們在模擬中將所提出的演算法用於估計此函數，並檢驗估計量在不同假設下的表現。

In longitudinal follow-up studies, recurrent events data in presence of competing risks are commonly seen. Besides the end-of-study effect, subjects may leave the study due to different reasons such as loss to follow-up, withdrawal or the occurrence of terminal events. The complicated mechanism s as well as the censoring issue becomes the major challenges for statistical inference. We study the complicated phenomenon under a unified framework which includes some familiar data structures as special cases. We also introduce the frailty model which is a popular and useful approach to constructing correlated random variables. Then we propose several data generation algorithms and apply them in our simulation study. Specifically we consider estimation of the cumulative incidence function (CIF), which is useful descriptive measure for describing competing risks data, when the data are generated according to the proposed algorithms.