Abstract

In otolaryngologic studies, subjects may contribute either unilateral observations from only one ear or bilateral observations from a pair of ears. For bilateral subjects, each person contributes information from two ears, the values of which are generally highly correlated. To avoid the confounding effect in stratified otolaryngologic studies, stratified data analysis is an important research topic. Based on the dependence model in stratified designs, this article presents five simultaneous confidence intervals (CIs) and two bootstrap simultaneous CIs for proportion differences with combined unilateral and bilateral data. Six approximate CIs and two bootstrap CIs for the common proportion difference are also developed. The proposed CIs are evaluated by empirical coverage probability, empirical coverage width and the ratio of mesial non-coverage probability to the non-coverage probability. Simulation results show that the simultaneous CIs based on Wilson method, the inverse hyperbolic tangent transformation, score statistic and the bootstrap percentile simultaneous CIs perform well and hence be recommended for applications, score CI and the CI based on Cochran statistic for the common proportion difference behave well even under small sample sizes, other CIs produce good performance when sample size is not small. A real otolaryngology study data is used to illustrate the proposed methodologies.

摘要

在耳鼻喉科研究中，受试者可以仅从一只耳朵提供单侧观察结果，也可以从一对耳朵提供双边观察结果。对于双边受试者，每个人从两只耳朵提供信息，其值通常高度相关。为了避免分层耳鼻喉科研究中的混杂效应，分层数据分析是一个重要的研究课题。基于分层设计中的依赖模型，本文提出了五个同时置信区间 (CI) 和两个自举同步 CI，用于结合单边和双边数据的比例差异。还开发了常见比例差的六个近似 CI 和两个引导 CI。所提出的 CI 通过经验覆盖概率进行评估，经验覆盖宽度和中间非覆盖概率与非覆盖概率的比率。仿真结果表明，基于Wilson方法、反双曲正切变换、评分统计和Bootstrap百分位数联动CI表现良好，值得推荐应用，评分CI和基于Cochran统计的CI对于共同比例差异表现良好即使在小样本量下，其他 CI 在样本量不小的情况下也能产生良好的性能。使用真实的耳鼻喉科研究数据来说明所提出的方法。得分统计和引导百分位数同步 CI 表现良好，因此推荐用于应用程序，评分 CI 和基于共同比例差异的 Cochran 统计的 CI 即使在小样本量下也表现良好，其他 CI 在样本量不小时也能产生良好的性能。使用真实的耳鼻喉科研究数据来说明所提出的方法。得分统计和引导百分位数同步 CI 表现良好，因此推荐用于应用程序，评分 CI 和基于共同比例差异的 Cochran 统计的 CI 即使在小样本量下也表现良好，其他 CI 在样本量不小时也能产生良好的性能。使用真实的耳鼻喉科研究数据来说明所提出的方法。