| 摘要: | 在臨床實驗中，有研究者提出了新的比較藥效的方法。其做法是將患者分
為使用新藥物或新療程治療的實驗組，及使用安慰劑或不治療的控制組。再將實驗組與控制組的病患根據他們的風險狀況做倆倆配對，藉由排序疾病導致的症狀嚴重性的設計，以心血管疾病為例：優先考慮是否死亡，再來考慮是否中風。若是使用新藥物的病患沒有死亡而使用安慰劑的病患死亡則記為獲勝，反之記為失敗。或配對後的病患皆沒有死亡，但使用新藥物的病患沒有中風而使用安慰劑的病患中風也記為獲勝，反之也記為失敗。若這兩位配對後的病患皆沒有死亡也沒有中風則記為平手。在多組倆倆比較後可以得到獲勝組數、失敗組數與平手組數。因此可得到 net benefit、win ratio、win odds 的估計量，再進一步去評估新藥物或新療程的治療效果。
本文提出使用概似方法 (likelihood method) 對配對設計下上列三個參數做推論，我們透過模擬研究與實例分析來探討概似方法得到的推論表現。並將信賴區間與 Matsouaka (2022) 所使用的在 Zou and Donner (2008) 與 Donner and Zou (2012) 所提出的 MOVER (method of variance estimates recovery) 與 Pocock et al. (2011) 提出的信賴區間做比較。
;In clinical trials, some researchers have proposed new methods for comparing drug efficacy. It is done by dividing patients into an experimental group that gets a new drug or course of treatment, and a control group that gets a placebo or no treatment. Then, the patients in the experimental group and the control group are paired according to their risk status, and the design is designed by sorting the severity of symptoms caused by the disease, taking cardiovascular disease as an example: priority is given to death, and then to stroke. If the patients who use the new drug do not die and the patients who use the placebo die, it is recorded as a win. On the contrary, it is recorded as a lose. Or if the two paired patients did not die, but the patient who received the new drug did not have a stroke and the patient who received the placebo had a stroke, it was also recorded as a win, and vice versa was also recorded as a lose. If neither of the two matched patients dies or suffers a stroke, it is a tie. After comparing multiple groups, the number of winning groups, the number of losing groups and the number of tie groups can be obtained. Therefore, the estimates of net benefit, win ratio, and win odds can be obtained, and then the therapeutic effects of new drugs or new courses of treatment can be further evaluated.
In this paper, we propose to use likelihood method to make inferences on the above three parameters under the matching design. We discuss the inference performance obtained by the likelihood method through simulation research and case analysis. And compare the confidence interval with the MOVER (method of variance estimates recovery) proposed by Zou and Donner (2008) and Donner and Zou (2012) used by Matsouaka (2022), and the confidence interval proposed by Pocock et al. (2011). |
