两个有序分类变量构建一个分类复合终点指标方法的模拟评价

Simulation Evaluation of Constructing a Categorical Composite Endpoint from Two Ordered Categorical Variables

目的对于临床试验有效性评价中两个或可以转变为两个均为有序分类变量的主要终点指标,提出一种最乐观或最悲观的构建分类复合终点的方法,分析这种方法的合理性及应用性。方法采用MonteCarlo模拟的方法,考虑调整样本量和相关系数,分析分类复合终点指标进行疗效评价的Ⅰ型错误和检验效能,并与多重检验和连续复合终点指标的结果进行比较。结果Ⅰ型错误方面,随着样本量和相关系数的增大,两个主要终点指标均有统计学意义的多重检验的Ⅰ型错误远低于检验水准0.05,至少一个主要终点指标有统计学意义的多重检验的Ⅰ型错误在0.04至0.05之间,分类复合终点指标和连续复合终点指标的Ⅰ型错误均保持在0.05左右。检验效能方面,整体上,分类复合终点指标的检验效能、连续复合终点的检验效能和至少一个主要终点指标有统计学意义的多重检验的检验效能接近,三者均大于两个主要终点指标均要有统计学意义的多重检验的检验效能,后者最保守。各方法的检验效能与两个主要终点指标间相关系数的关系因赋值不同而有不一样的变化趋势。结论对于临床试验两个或可以转变为两个均为有序分类变量的主要终点指标的资料,可根据临床实际意义构建最乐观或最悲观分类复合终点指标,其能得出可解释的综合水平,能控制Ⅰ型错误且具有较高的检验效能。而且无论相关系数大小,都可以构建分类的复合终点指标,因为乐观与悲观之间没有固定的优劣关系,使得研究者在实际研究过程中根据实际情况来构建评价指标,而不是倾向于选择乐观的方法来构建,避免这一倾向带来的偏倚。

Objective For two or can be converted to two ordered categorical primary endpoints of clinical trials , prop- osethe most optimistic or pessimistic method to construct categorical composite endpoint and evaluate reasonableness and applica- bility of this method. Methods Through Monte Carlo simulation, consider adjusting the sample size and correlation coefficient, compare type Ⅰ error and power of efficacy evaluation among three methods （ categorical composite endpoint index, multiple tes- ting and continuous composite endpoint index）. Results In terms of type Ⅰ error,with the increase of sample size and correla- tion coefficient,type Ⅰ error of multiple testing that two primary endpoints are statistically significant is far below 0. 05, and multiple testing that at least one primary endpoint is statistically significant is between 0. 04 and 0. 05, while type Ⅰ error of cat- egorical composite endpoint and continuous composite endpoint indexes are maintained around 0. 05. In terms of power, power of categorical composite endpoint, power of continuous composite endpoint and power of multiple testing that atleast one primary endpoint is statistically significant are close. The former thre epowers are much larger than power of multiple testing that two pri- mary endpoints are statistically significant, which is the most conservative. But there is different trend of power change for differ- ent correlation coefficients between the two primary endpoints. Conclusion For two or can be converted to two ordered categor- ical primary endpoints of clinical trials, we can constructthe most optimistic or pessimistic categorical composite endpoint accord- ing to actual clinical meaning, which can provide useful interpretable comprehensive level and increase power under the control of type I error. And whatever the size of the correlation coefficient, we can build categorical composite endpoint, because there is no fixed relationship about the pros and cons between optimistic and pessimistic methods. So in real clinical trials, researchers will construct categorical composite endpoint index according to the actual situation, rather than tending to choose optimistic ap- proach and avoiding the tendency to bring bias.