固定点处组间生存率比较的统计检验法

Statistical methods for comparing survival rates at a fixed time point

组间生存曲线的比较是研究疾病预后的重要内容，目前常用Log-rank检验，但当生存曲线交叉或者两曲线后期开口较大，即比例假设不成立时，采用该法则失效。为此本文讨论适用这种情形的比较法。介绍两条生存曲线某固定点处生存率比较的5种（经典法、对数转换法、cloglog转换法、反正弦转换法和logit转换法）统计检验方法。通过实例分析发现，若用组间生存曲线整体检验的Log-rank检验法或Two-stage检验法均很难对两组生存率比较给出合理的结果，而选取几个固定点进行检验时，可得到固定时间点上组间生存率差异是否有统计学意义。在5种固定点时刻的检验方法中，以cloglog转换法最佳。

Comparison of survival curves between two groups is an important part of disease prognosis study. Log-rank test is commonly used, but when the two curves＇ later intersecting opening is too large, the proportion of assumptions is not established, thus the Log-rank test is ineffective. We introduces five statistical tests to compare two survival curves at a fixed time points： classic method, logarithmic transformation, cloglog transformation, arcsine transformation and logit transformation. Through the study we found that if the overall survival curves are difficult to compare between groups tested with Log-rank test method or Two-stage test method, the fixed time point test can effectively determine whether there was significant difference in survival rate at a fixed time point. Among the five fixed time ooint tests, cloglog transformation could give more precise result.