Heimann G, Neuhaus G
Institute for Mathematical Stochastics, University of Hamburg, Germany.
Biometrics. 1998 Mar;54(1):168-84.
In the random censorship model, the log-rank test is often used for comparing a control group with different dose groups. If the number of tumors is small, so-called exact methods are often applied for computing critical values from a permutational distribution. Two of these exact methods are discussed and shown to be incorrect. The correct permutational distribution is derived and studied with respect to its behavior under unequal censoring in the light of recent results proving that the permutational version and the unconditional version of the log-rank test are asymptotically equivalent even under unequal censoring. The log-rank test is studied by simulations of a realistic scenario from a bioassay with small numbers of tumors.
在随机删失模型中，对数秩检验常用于比较对照组与不同剂量组。如果肿瘤数量较少，通常会应用所谓的精确方法从排列分布中计算临界值。本文讨论了其中两种精确方法，并证明它们是不正确的。根据最近的结果，即对数秩检验的排列版本和无条件版本即使在不等删失情况下渐近等价，推导出了正确的排列分布，并研究了其在不等删失下的行为。通过对生物测定中肿瘤数量较少的实际场景进行模拟，研究了对数秩检验。
