臨床試驗之多重檢定

Multiplicity in Clinical Trials

藥物的療效須藉由臨床試驗之假說檢定來驗證。多重檢定常見於臨床試驗設計，包括多個主要與次要評估指標、多個試驗組別、多個主要評估時間點、多個分析群體以及子群體分析等。為避免因多重檢定導致試驗將無效藥誤判為有效的機率擴增，使得試驗對療效結果的宣稱受到質疑，必須選用適當的統計方法控制整體型一誤差。本文目的在討論何種設計下會有多重檢定問題的產生，以及介紹幾種可適當處理多重檢定問題的統計方法，並佐以範例說明供臨床研究者參考。

A clinical trial is often designed to test multiple hypotheses on the efficacy of a new drug. Multiplicity in clinical trials may be induced by multiple endpoints, multiple treatment arm comparisons, multiple time points, multiple analysis populations, and subgroup analyses. It is well recognized that multiplicity can have a substantial influence on the rate of false positive findings if no appropriate statistical method is used to safeguard against the inflation of false positive findings from multiple tests. The paper accordingly focuses on examining situations with the potential of triggering multiplicity problems and introducing some common multiple testing methods for controlling the false positive rate. An example comparing different statistical methods is provided for the reference of clinical researchers.