摘要
目的探讨贝叶斯方法和Simon二阶段方法应用于篮子试验设计的优劣及实例分析判断试验疗效。方法模拟研究针对二分类结局变量,考虑不同试验层数、期中分析纳入人数等参数设置,比较两种方法的篮子试验设计在检验效能、控制I错误率及样本量的区别,并通过实例来介绍其应用。结果模拟试验结果表明,除异质性较高的2种情景外,贝叶斯方法的篮子试验设计效能均高于应用Simon二阶段方法,且对假阳性率控制更为严格,代价便是无效篮子增多时检验效能降低。贝叶斯方法的篮子试验的样本量总体较少,尤其在试验层数较多和同质情况下。结论贝叶斯方法的篮子试验更加灵活,可借用层间信息,相比Simon二阶段方法,更严格地控制整体一类错误率,且在试验层数和期中分析次数多时更节省平均总样本量。
Objective To evaluate the advantages and disadvantages of the basket trial design using Simon’s two-stage design and Bayesian method.Application of discovering histologic types in which the treatment is active is discussed.Methods In this paper,we focus on binary outcome and conduct series of simulation studies under various considerations,such as disease types,accrual of interim analyses and so on.We compare the difference of power,type I error rate and sample size between two design methods.At last,we implement the Bayesian basket trial through the actual applications.Results Our simulation studies show that basket trial using Bayesian method often has higher power to detect the treatment effect except for two scenarios with heterogeneity.On the other hand,Bayesian basket trial design has better controlled false positive rate in exchange for power decrease when there are more inefficient subgroups.The sample size needed in Bayesian basket trial is generally smaller,especially when the number of strata is large and treatment effect is homogeneous across tumor strata.Conclusion The Bayesian basket trial design provides more flexibility,which allows information borrowing across cancer types.This design compared to the basket trial design using Simon’s two-stage design controls better family wise error rate and saves more average total sample sizes when the number of strata and interim analysis is large.
作者
谭明敏
杨旻
仲子航
于全骥
于朝雷
周佳薇
倪森淼
蔡丽馨
于浩
柏建岭
陈峰
Tan Mingmin;Yang Min;Zhong Zihang(Department of Biostatistics,School of Public Health,Nanjing Medical University(211166),Jiangsu)
出处
《中国卫生统计》
CSCD
北大核心
2020年第3期349-353,共5页
Chinese Journal of Health Statistics
基金
国家自然科学基金(81773554)
国家自然科学基金青年基金(81302512)。
