一般的群序檢定方法中,每位受測者僅有單一觀察值,所以各階段檢定統計量之間具有IIS(independent increments structure)性質。 常見的群序檢定方法有Pocock(1977)、O'Brien與Fleming(1979)以及 Lan與DeMets(1983)等三種方法。然而在長期追蹤資料(longitudinal data)下,每位受測者有重覆測量值,而且這些測量值彼此間具有相關性,因此各階段檢定統計量之間不再具有IIS性質。針對分析重覆測量值或者多重反應變數之資料型態,Armitage等人(1985),Geary(1988),Tang等人(1989)以及Lee與DeMets(1991)分別提出不同的方法。本文將以Lee-DeMets方法為基礎,推廣其線性混合模式概念至多項式趨勢型態,應用二次式檢定統計量進行群序檢定。此外,藉由模擬研究討論各階段檢定統計量之邊際抽樣分配和所有檢定統計量之聯合分配,並使用實例說明其檢定程序。 關鍵字:IIS性質,線性混合模式,長期追蹤資料。
Classical group sequential methods are based on the assumption of independent increments structure (IIS) between the interim test statistics. Three common classical group sequential methods are proposed by Pocock(1977), O'Brien and Fleming(1979), and Lan and DeMets(1983). However, for longitudinal data the IIS assumption between the interim test statistics does not hold because of the correlation between the measurements from the same subject. Several parametric methods of group sequential test for analyzing the data with repeated measurements or multiple observations have been developed by Armitage et al.(1985), Geary(1988), Tang et al.(1989) and Lee and DeMets(1991). The proposed quadratic form test statistic is a generalization of Lee and DeMets' statistic to polynomial setting. The sampling distributions of the proposed test statistic at each stage as well as the joint distribution are discussed by simulation studies. The proposed testing procedure is illustrated by a clinical example.