本文考慮合併使用兩種藥物的複合藥之早期臨床試驗設計，其中包含至少一種標靶藥物。標靶藥物的藥效可能在低劑量時增加，然後在較高的劑量時趨於平穩，因此試驗目的是找尋複合藥的最佳劑量組合，使得服用該劑量組合的病患獲得最高的藥效但是維持目標毒性機率。本文提出兩種早期臨床試驗設計，根據二元毒性及藥效反應的觀察值，進行配置下一世代病人劑量組合。第一種設計為推展單一藥物的STEIN設計至複合藥，藉分別比較毒性機率與藥效機率的最大概似估計值與對應的毒性機率與藥效機率臨界值建立劑量升降方法。第二種設計的劑量升降為比較劑量組合毒性機率的 100(1-α)% 之最短長度可信區間及TTP，以及比較藥效機率之貝氏估計值與期望的藥效機率。因此第一種設計記為 mSTEIN，第二種設計則記作 CLUB12 。兩種設計皆有提早結束試驗的機制，而且當人數用盡時，利用效用函數估計最佳劑量組合。因為本文提出的早期臨床試驗設計無需假設任何特定的劑量組合毒性機率或劑量組合藥效機率模式，故稱為模型輔助設計。本文最後在各種不同的毒性機率與藥效機率的組合情境下進行模擬，研究所提設計的效率。模擬結果顯示當藥效機率呈現組合劑量之高原型態時，在正確估計最佳劑量組合與適當配置病人劑量組合方面mSTEIN設計表現優於CLUB12設計，但是當藥效機率為組合劑量之遞增型態時，CLUB12設計則比 mSTEIN設計具有更佳的設計效率。;This thesis considers early clinical trial designs for the combination of two drugs, which includes at least one Molecularly Targeted Agent (MTA). The efficacy of the MTA may increase at low doses and then plateau at higher doses. Therefore, the purpose of the trial designs is to find the optimal dose combination (ODC) so that patients taking this dose combination achieve the highest efficacy but maintain the target toxicity probability (TTP). Two early clinical trial designs are proposed where the dose combination escalation/de-escalations procedures are based on observations of binary toxicity and efficacy responses. The first design is to extend the single-drug STEIN design to a design for combination drugs, where the dose escalation/de-escalation procedure is constructed by comparing the maximum likelihood estimators (MLEs) of the toxicity and efficacy probabilities with the corresponding thresholds, respectively. The dose escalation/de-escalation procedure of the second design is established by comparing the 100(1-α)% highest posterior probability (HPD) credible interval of the toxicity probability with the TTP and comparing the Bayesian estimate of the efficacy probability with the expected efficacy probability. Therefore, the first design is denoted as mSTEIN and the second one is denoted as CLUB12. Both the designs have rules for early termination and the utility function is used to estimate the ODC when the maximum number of patients is reached. Also, the mSTEIN and CLUB12 are model-assisted designs since there are not any assumption about the specific dose combination-toxicity or dose combination-efficacy model. Finally, a simulation study is carried out to investigate the efficiency of the proposed designs under various combinations of toxicity and efficacy probabilities. The results show that when the efficacy is plateau over the combination doses, the mSTEIN design performs better than the CLUB12 in correctly selecting the ODC and dose-combination assignment to patients. However, when the efficacy probability is ordered, the design efficiency of the CLUB12 design is superior to that of the mSTEIN design.
