摘要
目的针对两条生存曲线出现交叉的情况,探讨其组间生存率比较的检验方法,并分析Log-rank、Breslow、Tarone-Ware、Renyi、Lin-Wang、Two-stage六种检验方法的适用情形。方法采用Monte Carlo方法模拟不同参数分布情形下,无删失和平均删失比例为10%、20%、30%时,六种方法的检验效能和I类错误。结果 Log-rank检验在两组危险率成比例时具有最优的检验效能,Breslow与Tarone-Ware检验对后期的差异不敏感,Two-stage、Lin-Wang、Renyi检验在生存曲线交叉时具有较好的检验效能,其中Renyi检验较为保守。当两组危险率满足比例风险假定或两条生存曲线后期差异大时,六种方法的检验效能随删失比例的增加而减小。结论不论生存曲线是否出现交叉,综合检验效能和I类错误两方面,Two-stage方法在两组生存率比较的问题上较其他五种方法具有更广的适用性,而当两条生存曲线交叉时,用Log-rank等常规方法可能会得到不合理的结论。
Objective To choose a proper statistical inference method for two crossing survival curves and to evaluate the suitable situa- tions of six methods. Methods Monte Carlo simulations are carded out to compare the power and type I error under a variety of situations with dif- ferent censoring rates and various parameter distributions. Results Log- rank test has optimal power when two hazard rates hold proportional hazard assumption. Breslow and Tarone-Ware test are insensitive to the late differ- ences. Two-stage, Lin-Wang and Renyi test are appropriate to detect the differences between two crossing survival curves. Among them, Renyi test is more conservative. When two survival curves hold proportional hazard assumption or they are close in the beginning but then separate, the power of six methods goes down with the increase of censoring rates. Conclusion Considering the power and type I error, no matter two survival curves cross or not,Two-stage test is more widely applicable to compare two sur- vival rates than the others, while the conventional tests may draw unreason- able conclusions especially for two crossing survival curves.
出处
《中国卫生统计》
CSCD
北大核心
2013年第5期668-672,675,共6页
Chinese Journal of Health Statistics
基金
国家自然科学基金(81202288)
广东省科技计划(20108031600100)
广州市科技计划(2012J5100023)
高等学校博士学科点专项科研基金资助课题(20114433120010)
