在臨床實驗中,基於成本與管理的考量,常使用群序檢定以有機會提早結束試驗。當資料為橫斷面(cross-sectional data)型態,且欲檢定兩處理平均數之差異時,常使用的群序檢定方法有Pocock(1977)、O’Brien與Fleming(1979)以及Lan與DeMets(1983)等方法。Lee與DeMet(1991)提出在長期追蹤資料(longitudinal data)中,利用線性混合模式檢定兩處理之線性趨勢。本文延伸Lee與DeMets之方法,推廣其線性混合模式至多項式趨勢型態,提出在多種處理下之群序檢定方法。各階段檢定統計量之聯合分配為多變量卡方分配,文中使用Pocock及O’Brien與Fleming概念,提出二階段與三階段檢定之臨界值,並使用實例說明此檢定程序。 For ethical, economical and administrative considerations, group sequential methods are frequently applied for possibly early determination clinical trials. For cross-sectional data, three common group sequential methods for comparing means between two treatments are proposed by Pocock (1977), O''Brien and Fleming (1979) and Lan and DeMets (1983). For longitudinal data, Lee and DeMets (1991) proposed a sequential comparison for testing the rates of change with linear trend between two treatments. In this article, a group sequential method for multi-armed trials is proposed, which is a generalization of Lee and DeMets'' method, for testing responses changes with time in polynomial trend. The asymptotic joint distribution of proposed test statistics is a multivariate chi-squared distribution. Boundaries of the proposed methods based on Pocock-type and O''Brien and Fleming-type are provided for practical use. The proposed testing procedure is illustrated by a clinical example.