不同相依設限情境下序列二元間隔時間的 時間區塊交叉比估計

Abstract

在長期追蹤的醫學研究中，個體可能經歷反覆發生有序的二元事件，像是慢性病患者反覆發生入院與出院兩種情況。本研究感興趣的是反覆發生兩種事件的二元間隔時間，例如，病人在醫院治療的時間以及離開醫院後再次進去醫院治療的時間，此資料型態為序列二元間隔時間資料。二元復發事件會因設限而不再發生事件，例如，研究終止時。一般常見假設設限時間與序列二元間隔時間獨立，然而，實際上設限時間可能會與過去序列復發事件的歷史有關。交叉比為常用於二元間隔時間相關性的一種測度，本文目的估計在不同相依右設限情境下的全域以及時間區塊之交叉比。當分析第二間隔時間與後續的間隔時間時，會有誘導性相依設限的問題，可使用設限機率倒數權重(inverse probability of censoring weigh，簡稱IPCW)解決此問題，本文考慮三種設限情境下，利用三種設限機率倒數權重來估計交叉比。最後，針對兩種不同序列二元間隔時間的聯立分布並考慮在不同相依設限情境下，以蒙地卡羅模擬分析比較不同交叉比估計方法的表現。

In a follow-up medical study, subjects may experience ordered bivariate events alternately over time, for example, repeated discharges and readmissions for chronic disease patients. In the study, the time variables of interest are the bivariate gap times between bivariate events, such as the length of stay in a hospital and the time interval between the discharge from the hospital and readmission. This data set is called the serial bivariate gap time data. In general, the censoring, for example, end of study, may stop the bivariate recurrent event process. Typically, the censoring time is assumed to be independent of serial bivariate gap times. However, in reality, the censoring time may depend on the history of serial recurrent gap times. The purpose of our study is to estimate the global and time-segment cross ratios, which are used to measure the global and local associations between bivariate gap times, under certain dependent censoring situations. The problem of the influence of the induced dependent censoring for analyzing the second or later gap times can be solved by using the inverse probability of censoring weight (IPCW). We consider three IPCWs to estimate cross ratio subject to three censoring situations. Our Monte Carlo simulation studies are conducted under various dependent censoring scenarios with two different joint distribution of serial bivariate gap times to evaluate the performance of the proposed estimates of the cross ratios.