- 學位論文
臨床試驗群集設計之期中樣本數估計
Sample Size Estimation for Cluster Design in the Middle of Clinical Trial
摘要
樣本數計算在臨床試驗中扮演很重要的角色，因為樣本數太少會無法證明要藥物的療效，樣本數太多浪費資源，增加試驗時間或浪費經費，由於缺乏臨床先導研究資料，無法得知計算樣本數的參數，此研究在討論臨床試驗群集設計之期中樣本數估計。此研究以群集隨機對照試驗(cluster randomized controlled trials)設計實驗，將群集(cluster)作為隨機化的單位，而不是將每個患者單獨分配到治療組中。以半盲設計為基礎，使用給定的轉移機率將群集從原始治療組隨機轉換到新的治療組，並在不知道群集轉換的情況下重新評估樣本量。採用試驗期中轉換後的數據和轉換機率，計算樣本數，轉換後的變異數分析(ANOVA)模型的樣本數大於原始資料的變異數分析(ANOVA)模型的樣本數。隨著轉移機率越來越接近 0 或 1，轉換後的變異數分析(ANOVA)模型的樣本數會更接近原始的變異數分析(ANOVA)模型的樣本數。當轉移機率等於0或1時，就代表是原始的治療組，所以轉換沒有意義。如果該機率接近 0.5，將導致獲得較低的披露風險和較大的樣本數。藥廠應在樣本數和披露風險之間取得平衡。本文推薦轉移機率10%或90%。為了保持盲性設計，參與試驗的研究參與者不知道轉移概率的真實值，但知道轉移機率為10%或90%。
並列摘要
The sample size calculation plays a very important role in clinical trials. We cannot prove the efficacy of the study drug if the number of sample sizes is too small. It wastes the resources and increases the study time if the number of sample sizes is too large. Due to the lack of previous clinical data, it is impossible to know parameters for calculating the sample size. In this thesis we study the semi-blinded sample size estimation of the clinical trial for the cluster design. It is a cluster randomized controlled trial, using the cluster as the unit of randomization rather than individually assigning each patient to a treatment group. Based on a semi-blind design, clusters are randomly switched from the original treatment group to the new treatment group based on a given probability and then the sample size was re-evaluated in the middle of trial. Using the new data and this probability in the middle of trial to calculate the number of samples, the number of samples in the new data for ANOVA model is greater than that in the original data. As the transition probability gets closer to 0 or 1, the sample size of the transformed ANOVA model will be closer to the sample size of the original ANOVA model. When the transfer probability is equal to 0 or 1, the new data equals to the original treatment group. If the probability is close to 0.5, it will lead to lower disclosure risk and larger sample size. Pharmaceutical companies should strike a balance between sample size and disclosure risk. This paper recommends a transfer probability of 10% or 90%. In order to maintain a blinded design, the study participants who participated in the trial did not know the true value of the transition probability, but knew that the transition probability is either 10% or 90%.