固定点处k组生存率比较的非参数统计方法研究

Methods of Comparing Multiple Survival Rates at a Fixed Time Point

目的当仅需比较某固定时间点上组间生存率的差异,或者不满足组间风险率成比例假设时,如生存曲线存在交叉,log-rank检验不再适用,且现有固定点检验法仅限于两组间,故本文发展固定点处多组间生存率的比较方法。方法首先提出多组间五种固定点检验法(经典法、对数转换法、双对数转换法、反正弦平方根转换法及逻辑转换法),并通过Monte Carlo模拟评价五种方法在不同情形下的一类错误和检验效能。最后对满足和不满足风险率成比例假设两个实例用上述方法进行分析。结果综合Monte Carlo模拟得到的一类错误及检验效能结果,以经典法和反正弦平方根转换法最为激进,对数转换法略保守,逻辑转换法最为保守,而双对数转换法最为稳健。结论在进行多组间生存率比较时,当仅想比较多组间某固定点处生存率差异或者组间不满足成比例假设,可使用上述五种固定点检验法,其中优先建议使用双对数转换法。

Objective In comparing multiple survival curves at a fixed pointin time, log-rank test is inapplicable. Be- sides, its power would be worse in crossing survival curves because of not meet the proportional hazard assumption. Hence, we use the method of comparing survival rates at fixed point to deal with them. However, this method can only use for two groups. In view of the above,we considered the comparison of multiple survival curves at fixed point. Methods We first proposed 5 methods to compare multiple survival curves at fixed point（ naive, log, cloglog, arcsin, logit）. Monte Carlo simulations were car- fled out to evaluate the type I error and power of these methods. Finally, we used two examples for analysis by using above meth- ods. Results Comprehensive results of type I error and power, naive and arcsin were the most radical ways ;log and logit were more conservative; and cloglog was the most robust. Conclusion In comparison of multiple survival rates, someone can choose our methods of comparing multiple survival curves at fixed-point when these survival curves do not meet the proportional hazard assumption or only interested at fixed-point in time. And we suggested cloglog method.