五种最小不平衡指数法的平衡能力比较

目的比较临床试验分组方法之一最小不平衡指数法中五种不同方法的平衡能力。方法采用随机样本和固定样本两种样本,以完全随机分组法为对照,比较最小不平衡指数法中的极差法(RM)和方差法(VM)、信号法(SM)、动态上限法(UM)在不同样本量、最优分组概率、试验分组数组合情形下,每种参数组合计算机模拟运算1000次后,分别观察对组间人数、各个变量、人数相对较少变量水平的平衡能力的相对大小,并计算出在考虑所有需要平衡变量平衡程度时的总体平衡能力胜出率。结果优度法在总体平衡能力胜出率、变量平衡和人数相对较少变量水平平衡三个方面强于其他四种最小不平衡指数法,方差法、极差法、信号法结果近似,动态上限法最差但强于完全随机分组法。当最优分组概率增大时,优度法的优势更加明显。极差法、方差法和优度法组间人数平衡能力相近,当P=1.0时,信号法的组间人数平衡能力与前者接近。结论优度法(GM)的平衡能力最强,建议取代目前广为应用的极差法和方差法。

Simulation Studies on the Balance Ability for Five Minimization Schemes in Clinical Trials

Objective To compare the balance ability of five minimization schemes in clinical trials.Methods 36 simulation studies were conducted with samples randomly sampled from a hypothetical population to observe the balance ability with respect to balance across groups,win probability considering four factors desired to keep balance across groups,two simulation studies with a fixed sample from the same population to observe the balance ability from the angles of each factor and factor level with small number of participants.Results goodness method outperformed others four minimization methods from group,factor and level angles.Conclusion Goodness method(GM)is encouraged to put into practice to replace popular range method and variance method.