医疗器械临床试验单组目标值法样本量计算不同方法的比较

Comparison of Different Methods for Sample Size Calculation in Single Arm Performance Goal Studies of Medical Device Clinical Trials

目的:对单组目标值法样本量计算的不同方法进行比较,为实际应用中选择合适的方法提供依据。方法:构建目标值π_(0)和预计值π_(1),以正态近似法(2种)、通用法、反正弦法、确切标准法及确切保守法共6种方法分别计算各自所需的样本量,编程计算相应的最低成功率,并进行计算机模拟获得检验效能。结果:6种方法在π_(0)及π_(1)不接近0或1时表现较为相似,但π_(0)逐渐接近0时,正态近似法B和通用法得到的样本量相对较小,并逐渐损失了检验效能;π_(0)逐渐接近1时,正态近似法B和通用法得到的样本量相对较大,检验效能也比预设值逐渐增高。从检验效能来看,反正弦法的结果与确切标准法接近而显得更为离散,而确切保守法几乎能保证预设的检验效能,但在π_(0)>0.5时,确切保守法所需样本量逐渐多于确切标准法。不同方法对最低成功率的要求总体相似,但存在微小差别。结论:单个率比较的样本量计算方法的选择较为复杂,对检验效能要求比较高时,宜优选确切标准法和确切保守法,其次可考虑反正弦法,而通用法和正态近似法在率偏向两侧时,样本量会过大或过小,应根据实际应用情况具体权衡。

Objective:This study aims to compare different methods of sample size calculation in sing arm performance goal studies,providing a basis for selecting appropriate methods in practical applications.Methods:Performance goalπ0 and expected valueπ1 are constructed.Six methods,including two approaches of normal approximation,general method,arcsine method,exact standard method,and exact conservative method,are used to calculate the required sample size.The corresponding minimum success rates are calculated through programming,and the powers are obtained through computer simulation.Results:The six methods perform similarly whenπ_(0)andπ_(1)are not close to 0 or 1.However,whenπ_(0)gradually approaches 0,the sample size obtained by the normal approximation method B and the general method is relatively small,resulting in a gradual loss of power.Conversely,asπ_(0)gradually approaches 1,the sample size obtained by the normal approximation method B and the general method is relatively large,leading to an increase in power compared to the preset value.From the perspective of power,the results of the arcsine method are close to the exact standard method but appears more discrete.The exact conservative method almost always guarantees the preset power,but whenπ_(0)>0.5,the sample size required for exact conservative method gradually increases compared to the exact standard method.The requirements for the minimum success rate with different methods are generally similar with slight differences.Conclusion:The selection of sample size calculation methods for single arm performance goad studies is relatively complex.When there is a high requirement for power,it is advisable to choose the exact standard method and the exact conservative method,followed by the arcsine method.Sample size with the general method or the normal approximation method tends to be either too large or too small when proportions are biased towards 0 or 1,respectively.Hence,caution should be used based on the specific application context.