Sun Y, Sherman M
Department of Mathematics, University of North Carolina at Charlotte 28223, USA.
Biometrics. 1996 Mar;52(1):87-97.
We introduce two new classes of tests for censored data. The first tests for association between survival and a covariate and the second tests for equality of survival distributions between K groups. Both tests are permutation tests based on nonparametric test statistics and, unlike the Wald test in the proportional hazards model or the log rank test, can detect alternatives of the "crossed hazards" type. The tests are simple to implement and require no theoretical analysis by the user nor an appeal to asymptotic distribution theory. Simulation results show that the tests are competitive with their classical counterparts under proportional hazards and superior under certain nonproportional hazards alternatives in both the continuous and K-sample cases. The tests are applied to data from the British Medical Research Council's fourth myelomatosis trial and from a Danish study of malignant melanoma.
我们引入了两类针对删失数据的新检验。第一类检验用于生存与协变量之间的关联性，第二类检验用于K个组之间生存分布的相等性。这两类检验都是基于非参数检验统计量的置换检验，并且与比例风险模型中的 Wald 检验或对数秩检验不同，它们能够检测出“交叉风险”类型的备择假设。这些检验易于实施，用户无需进行理论分析，也无需诉诸渐近分布理论。模拟结果表明，在比例风险情况下，这些检验与经典检验具有竞争力，而在连续和K样本情况下的某些非比例风险备择假设下则更具优势。这些检验被应用于英国医学研究委员会的第四次骨髓瘤试验数据以及丹麦一项恶性黑色素瘤研究的数据。
