群随机试验与传统随机试验统计方法对比分析

Comparison of Statistical Approaches between Cluster Randomization Trials and Classic Randomization Trials

目的 介绍群随机试验的统计方法并与传统方法比较。方法 以青少年戒烟干预试验为例 ,介绍群随机试验的假设检验及样本含量估计方法。结果 用传统统计方法分析群随机试验会产生虚假的统计显著性结论 ,增大了Ⅰ型错误 ;用传统样本含量计算公式将低估群随机试验所需的样本含量 ,增大了Ⅱ型错误。

Objective Introduction of cluster randomization trials and comparison of statistical methods between cluster randomization trials and classic randomization trials.Method Demonstration of hypothesis test and estimation of sample size of cluster randomization trials by a school-based intervention to adolescent tobacco use prevention.Results Application of classic statistical methods to the analysis of cluster randomization trials will tend to bias observed P-values downward,thus lead to a type I error substantially above that which is planned for.On the other hand,application of calssic sample size formulas will lead to underpowered studies and a substantially elevated type Ⅱ error.Conclusion Serious analytic bias will occur if the effect of clustering is ignored.