在強化設計及考克斯比例風險模型下設限資料評估標靶藥物統計分析方法之研究

Abstract

在傳統的臨床實驗中，納入以及排除是基於臨床指標所考量的，但往往未考慮到受試者的基因或是基因體的變異。在人類基因體計畫完成後，許多疾病的分子標的可以被鑑別，因此可以發展出分子標的治療方法，但鑑定分子標的之診斷試劑通常並非完全準確，所以有些診斷標的臨床實驗的陽性病人實際上可能並沒有分子標的，因此對於真正擁有分子標的之病人族群而言，標的臨床實驗下之標的療法的療效估計值會有偏差。因此，我們提出對於真正擁有分子標的之病人配合標的治療之不偏推論統計方法。在強化設計的臨床試驗及半參數考克斯比例風險模型下，我們針對設限資料來探討處理效應並同時可考慮許多共變數來鑑定分子標的之診斷試劑的準確度，我們採用Eng K.H.及Hanlon B.M. (2014) 所提出之混合考克斯比例風險模型，並加以應用，且藉由EM演算法推導出考克斯比例風險模型下風險比例的估計式並且利用拔靴法來計算估計值之變異數。運用模擬研究來驗證所得之估計值與檢定程序而加以比較與現有方法之間的差異，及以實例數據以說明方法的應用。

In traditional clinical trials, inclusion and exclusion criteria are considered based on some clinical endpoints, the genetic or genomic variability of the trial participants are not totally utilized in the criteria. After the Human Genome Project is completed, many molecules underlying disease can be identified, it is possible to develop a targeted molecular therapy. However, the accuracy of diagnostic devices for identification of such molecular targets is usually not perfect. Some patients with positive diagnosis result is actually might not have the specific molecular targets. As a result, the treatment effect may be underestimated in the patient population truly with the molecular target. In order to resolve this issue, we propose a method based on the mixture Cox’s proportional model for the k latent classes (Eng K.H. and Hanlon B.M., 2014) and under the enrichment design. We develop inferential procedures for the treatment effects of the targeted drug based on the censored endpoints in the patients truly with the molecular targets which also incorporates the inaccuracy of the diagnostic device for detection of the molecular targets on the inference of the treatment effects. We propose using the EM algorithm in conjunction with the bootstrap technique for estimation of hazard ratio and its variance. Though the simulation study, we empirically investigate the performance of the proposed methods and to compare with the current method. The numerical examples illustrate the proposed procedures.