序列監測下現時狀態資料之半參數迴歸模型分析

Abstract

許多臨床及流行病學研究中，研究個體經常受到多次監測並記錄其主要事件狀態，該類型研究的資料蒐集方式即形成序列監測資料。一般來說，該種監測過程下主要事件的發生時間會受到區間設限，而區間設限資料在計算上較為複雜且費時，並且沒有辦法完整運用到監測訊息。因此，本研究目的即為在序列監測資料下發展一半參數的回歸分析。本研究推廣單次監測資料下的現時狀態資料估計方法(Lin, Oakes和Ying, 1998)以運用到更完整的監測訊息，並彙總監測過程的收集的一系列現時狀態資料提出二階段估計法對半參數加法風險模型的迴歸係數進行估計。本研究也對估計值的近似分布性質進行探討，並同時提出近似變異數的模型假設下及穩健估計。進一步對多種設限率及監測次數條件下進行模擬分析，以探討該方法在不同情況下的表現。

Many clinical and epidemiology studies often monitor the study subjects more than one time during the follow-up period and record their statuses of the event of interest at each monitoring time. The data collected from such studies are called the sequential monitoring data. Typically, the time to event of interest is subject to interval censoring under the monitoring process. The computation of interval censored data is complicated and time-consuming. However, the analysis of interval censored data does not fully utilize all the monitoring information. Therefore, this study is to develop a semiparametric estimation method for regression analysis based on the sequential monitoring data In this study, we extend the estimation method for current status data under a single monitoring time (Lin, Oakes, and Ying, 1998) to utilize the comprehensive monitoring information. We propose a two-stage estimation procedure by pooling the sequence of the current status data at monitoring times to estimate the regression coefficients in the semiparametric additive hazard model. The asymptotic properties of the proposed estimators are established and the model-based and robust estimates of the asymptotic variance are also provided. We conduct extensive simulation studies with various censoring rates and monitoring frequencies to investigate the performance of the proposed methods.