【摘要】 目的多分类结局指标中两类占比的比较目前尚无相应统计方法,本研究旨在建立多类别中某两类占比差的统计推断方法。方法根据多项分布理论,用正态近似法建立两类别发生占比差的假设检验方法,分别基于Wald法和Newcombe法构建其置信区间,包括连续性校正和非连续性校正两种情形。通过模拟验证假设检验方法的一类错误及检验效能和置信区间方法的覆盖率,最后以实例进行说明。结果基于占比差的假设检验在大样本下可以较好的控制一类错误。两种方法置信区间的覆盖率均在95%左右,Newcombe法优于Wald法,但在发生率较低时两种方法均不理想。结论本文提出多分类结局指标中两类占比差的假设检验及置信区间方法均能满足应用需求,其中置信区间方法推荐Newcombe法,但当样本量太小(如20例左右),所有方法均失效,建议使用描述方法。更多还原

【Abstract】 Objective No statistical inference method has been developed to define the significant difference between tw o categories of the multinomial outcome. This study aims to develop hypothesis test methods and interval estimation methods base on the percent difference( PD). Methods Hypothesis test method w as developed based on the theory of large sample and multinomial distribution. The confidence interval w as estimated based on the Wald method and the New combe’ s method separately,including the corresponding continuity-corrected methods. Type I error and pow er of the hypothesis test and the coverage rate of confidence interval w ere tested by M onte Carlo simulation methods. Results The type I error of the developed hypothesis test method w as w ell controlled under large sample. Confidence interval methods based on New combe’s method w ith or w ithout continuity-correction w ere better than Wald methods in the coverage rate. How ever,all of the interval estimation methods suffered from poor coverage rates w hen the percent of the tw o categories w as too low. Conclusion Both the hypothesis test and confidence interval methods brought up in the paper can meet application requirements and the methods based on New combe’ s method are recommended for confidence interval estimation. All methods w ork badly under small sample( such as 20),so descriptive methods are recommended for that case.更多还原