在臨床試驗中，雙臂試驗是常見的試驗方法之一，參與者被隨機分成兩組，其中一組接受實驗用藥(新藥)，而另一組則接受參考用藥或安慰劑。然而由於缺乏內部對照組，雙臂試驗的結果可能受到其他變數的影響，難以準確區分實驗用藥的真實效果，特別是存在族群或地域差異的情況下，試驗結論可能不具有普適性。 為解決雙臂試驗的不足，本篇論文探討三臂非劣性試驗，假設主要終點為右設限且包行政設限和失去追蹤設限，利用三種半母數模型，Cox比例風險模型、PO模型和AFT模型比較實驗用藥組、參考用藥組和安慰劑組在不同分配比例下，淨效應和分數效應所需要的樣本數差異，以期找出不同條件下最少樣本數的最適效應。此外，在型一誤差的部分，利用限制概似比檢定( CLRT )，透過不同的權重計算方法改善在聯合檢定時被低估的情況，使其符合預先設定的數值。最後將上述方法應用於膀胱癌資料，以驗證模擬試驗的可行性。;Two-arm trials are commonly used in clinical research, assigning participants to either an experimental group or a control group receiving a reference treatment or placebo. However, the absence of an internal control arm may introduce confounding, particularly in the presence of population or regional heterogeneity, thereby limiting the generalizability of the findings.
This thesis investigates three-arm non-inferiority trials with right-censored endpoints, including both administrative and loss-to-follow-up censoring. Three semiparametric models—the Cox proportional hazards model, the proportional odds (PO) model, and the accelerated failure time (AFT) model—are utilized to compare sample size requirements for net and fraction effects under various allocation ratios among the three groups, with the goal of identifying the most sample-efficient effect type under varying conditions.
To ensure valid control of the Type I error rate, a Constrained Likelihood Ratio Test (CLRT) is employed, incorporating alternative weighting strategies to address the underestimation observed in joint hypothesis testing. The proposed methods are further validated using a real-world bladder cancer dataset to demonstrate their practical feasibility.
