临床试验中剂量-反应率关系研究中的样本含量估计方法

Sample size calculation in dose-response research

目的:Ⅱ期临床试验中,剂量-反应率关系研究所需要的样本含量估计方法介绍及评价。方法:目前常用于二分类变量资料剂量-反应关系研究中样本量估计的方法包括Jun-Mo Nam提出的基于正态近似Cochran-Armitage趋势检验的估算方法和Chang提出的unified contrast估算方法,本文在假设反应率P在logit尺度下与剂量呈线性关系的前提下,分别用上述两种方法估算在不同的斜率和0剂量组反应率下所需的样本含量,并采用计算机模拟抽样技术评价不同样本含量所对应的检验效能。结果:当反应率P在logit尺度下与剂量呈线性关系时,在不同参数组合下,两种样本含量估算方法得到的结果均比较接近;Cochran-Armitage趋势检验正态近似法的模拟检验效能接近期望效能,而Chang's unified contrast方法的检验效能受对比系数的影响较大,若预设的对比系数的形状与实际反应率比较接近,则此时模拟得到的检验效能将高于设计时的检验效能。结论:当反应率P在logit尺度下与剂量呈线性关系时,若Chang's unified contrast方法中的对比系数的设定与反应率P形状相同时,两种样本含量估计方法基本一致。

AIM. To introduce and evaluate two methods for sample size calculation in doseresponse research. METHODS. The methods used for sample size calculation in dose-response research include Cochran-Armitage trend （CAT） test proposed by Jun-mo Nam and the Unified contrast （CUC） method proposed by Chang. This paper takes the example of binary data, uses the two methods to calculate the required sample size and evaluate the power by simulation under the assumption that the response rate had a linear relationship with dosage under the logit scale. RESULTS： When the probability of response followed a linear trend on the logit scale, the sample sizes estimated from the two methods are approximately equal in various parameter＇s settings. The simulated power of CAT test was close to the expected one, however, the simulated power of CUC method was obviously affected by the values of contrast coefficient, if a contrast whose coefficients parallel the expected response, the CUC method was more powerful. CONCLUSION： When the probability of response follows a linear trend over dosage on the logit scale, the two methods reach almost the same result, in such circumstance, the Chang＇s unified contrast method will obtain an increasing power.