多變量整合分析與多指標多區域臨床試驗之多維度隨機效應模型之應用

Abstract

製藥領域中，新的治療療效時常藉由數個相關的指標做為評估。例如阿茲海默症的主要指標是參考認知測試、日常活動、與整體變化來評估。在此篇論文中，我們的目標希望解決在臨床試驗中考量數個指標時的兩個議題:(1) 對於具有異質性變異數試驗的多維度整合分析的傳統方法，其過去文獻的方法指出其經驗值(empirical values)的準確度是不夠好的。我們提出由Wishart類型多維度t分配的改進方法，且將其應用至多維度整合分析時常使用的多維度隨機效應模型之中；(2) 多區域臨床試驗，其收集來至世界各地不同區域且符合相同計畫書的參與實驗者，此試驗亦遇到在區域之間具有異質性變異數的情況。然而，過去的相關文獻多只考慮單一的指標。因此，我們將多維度隨機效應模型應用至多指標多區域臨床試驗。多維度整合分析的模擬結果指出我們的方法較其他方法精確度較高，以及實際例子的結果亦顯示傳統的方法相對較保守。對於多指標多區域臨床試驗，我們提供其所需樣本數計算以及評估特定區域療效是否與整體療效具有一致性之機率機算，此結果亦顯現Wishart類型多維度t分配的方法相較其他方法有較高的機率。

In some therapeutic areas, the clinical efficacy of a new treatment may be evaluated by a set of possibly correlated endpoints. For example, the main endpoints for Alzheimer’s disease are usually referred to cognitive test, activities of daily living, and global assessment of change. In this thesis, our aim is to deal with two issues based on the consideration for multiple endpoints in clinical trials: (1) the conventional approaches for the multivariate meta-analysis (MMA) having heterogeneous covariances across different trials may have a less accuracy of empirical values. We propose a refined method constructed by a Wishart type multivariate t distribution and apply the method to the multivariate random effect model (MREM) for MMA to raise the accuracy; (2) multi-regional clinical trials (MRCTs), incorporating subjects from many regions around the world under the same protocol, also meet the situation of existing heterogeneous covariances across regions. However, all of its related approaches are concerned with only one primary endpoint. We will apply the MREM to the MRCTs with multiple endpoints. The simulation studies for MMA indicate that our method has more accuracy than other approaches, and the real example also shows that the conventional multivariate normal distribution might be conservative. Besides, for the MRCTs, we provide the required sample size determination and the probabilities of observing the consistent trend between specific region and overall region. The performance also shows that the Wishart type multivariate t distribution has the higher probability among all approaches.