适应性富集设计中的效应点估计方法比较  

A Comparison of Point Estimation Methods in Adaptive Enrichment Designs

作　　者：王爽 仲子航 朱子涵 焦阳 柏建岭 于浩 Wang Shuang;Zhong Zihang;Zhu Zihan(Department of Biostatistics,School of Public Health,Nanjing Medical University,Nanjing 211166)

机构地区：[1]南京医科大学公共卫生学院生物统计学系,211166

出　　处：《中国卫生统计》2025年第1期38-43,49,共7页Chinese Journal of Health Statistics

基　　金：国家自然科学基金(82273738);江苏省高校“青蓝工程”。

摘　　要：目的探讨适应性富集设计中的效应点估计方法,为富集设计下的疗效估计提供参考。方法通过模拟适应性富集设计中的疗效数据,比较极大似然方法、第二阶段估计方法、Lindley估计方法、Bootstrap参数方法和条件无偏估计方法这五种方法在估计药物疗效方面的统计学性能。结果极大似然估计方法往往产生较大偏倚,且容易受参数影响;Lindley估计方法和Bootstrap参数方法偏倚较小,但Lindley估计方法可能低估药物疗效;第二阶段估计方法和条件无偏估计方法的估计几乎无偏。除了第二阶段估计方法外,其他方法的均方误差均较小。结论在适应性富集设计中,条件无偏估计方法的偏倚较小,均方根误差在可接受范围内,因此推荐使用该方法来估计药物疗效。Objective To explore methods for point estimation in adaptive enrichment design,providing a reference for effect estimation inn this context.Methods We simulated efficacy data under adaptive enrichment design and subsequentially estimated the effects using five methods:maximum likelihood method,stage 2 method,Lindley′s method,parametric Bootstrap method,and the uniformly minimum variance conditionally unbiased estimated(UMVCUE)method.The performance of these methods was further evaluated.Results The simulation study showed that the maximum likelihood method has a large bias and is very sensitive to some parameters,while Lindley’s method tends to underestimate and the parametric bootstrap method sometimes overestimates.The stage 2 method and UMVCUE are almost unbiased.In terms of mean squared error(MSE),all methods perform well except stage 2 estimator.Conclusion In adaptive enrichment design,the UMVCUE performs well in terms of bias and MSE.Therefore,we recommend it as the point estimator for the treatment effect in adaptive enrichment design.

关 键 词：富集设计 效应估计 期中分析 亚组 

分 类 号：R195.1[医药卫生—卫生统计学]