隨著醫療技術的不斷進步，治療腫瘤的選擇越來越多元化。為了評估新的治療方法與傳統方法之間的差異，我們需要進行臨床試驗以驗證藥品的治療效果。常見的評估方法包括確定對治療有反應的患者比例以及評估患者反應持續時間。在進行研究時，我們考慮了不同醫院之間存在的變異性，將分層變量作為隨機效應解釋未被觀察或測量的特徵，並使用了具有隨機效應的加速失效時間模型來建立患者反應持續時間的模型。通過估計模型中的參數，我們可以推算出對治療有反應患者的期望反應持續時間。同時，將有反應患者的期望反應持續時間與反應率結合，得到了所有隨機患者的期望反應持續時間，利用它建立一個公正的假說檢定。為了比較我們的檢定方法與其他方法之間的差異，我們進行了模擬研究。根據模擬結果，我們提出的檢定方法相對保守，即使在存在有限制的情況下，這方法也表現出良好的測試結果。最後，我們以急性骨髓性白血病患者的資料作為實例進行了分析，來說明我們提出的檢定方法。

With the continuous advancement of medical technology, the options for treating tumors have become increasingly diversified. In order to evaluate the differences between new treatment methods and conventional methods, we need to conduct clinical trials to validate the therapeutic effects of drugs. Common evaluation methods include determining the proportion of patients who respond to the treatment and assessing the duration of patient response. In this study, we took into account the variability among different hospitals, treating stratified variables as random effects to explain unobserved or unmeasured characteristics. We used an accelerated failure time model with random effects to establish a model for the duration of patient response. By estimating the parameters in the model, we can infer the expected response duration for patients who respond to the treatment. Additionally, by combining the expected response duration of responsive patients with the response rate, we obtained the expected response duration for all randomized patients, which was used to establish a fair hypothesis test. In order to compare our testing method with other methods, we conducted a simulation study. It was observed that our proposed testing method is relatively conservative, and this method demonstrates good testing performance even in the presence of data limitations. Finally, we applied our method to analyze the data of patients with acute myeloid leukemia as an example.

[1] Abdul-Fatawu, M. Accelerated failure time models:: An application in insurance attrition. The Journal of Risk Management and Insurance 24, 2 (2020), 12–35.
[2] Cary, N. Sas/stat® 14.1 user's guide. Cary, NC: SAS Institute Inc (2015).
[3] Do Ha, I., Jeong, J.-H., and Lee, Y. Statistical modelling of survival data with random effects. Springer, 2017.
[4] Doob, J. L. The limiting distributions of certain statistics. The Annals of Mathematical Statistics 6, 3 (1935), 160–169.
[5] Ellis, S., Carroll, K. J., and Pemberton, K. Analysis of duration of response in oncology trials. Contemporary clinical trials 29, 4 (2008), 456–465.
[6] Lambert, P., Collett, D., Kimber, A., and Johnson, R. Parametric accelerated failure time models with random effects and an application to kidney transplant survival. Statistics in medicine 23, 20 (2004), 3177–3192.
[7] Le-Rademacher, J. G., Peterson, R. A., Therneau, T. M., Sanford, B. L., Stone, R. M., and Mandrekar, S. J. Application of multi-state models in cancer clinical trials. Clinical Trials 15, 5 (2018), 489–498.
[8] Lee, Y., Nelder, J. A., and Pawitan, Y. Generalized linear models with random effects: unified analysis via H-likelihood, vol. 153. CRC Press, 2018.
[9] Wang, Y. Estimation of accelerated failure time models with random effects. Ph.D. Thesis, Department of Statistics, Iowa State University, USA, 2006.