Goldman A I
Division of Biostatistics, School of Public Health, University of Minnesota, Minneapolis 55455.
J Clin Epidemiol. 1991;44(12):1327-40. doi: 10.1016/0895-4356(91)90094-p.
The use of the Kaplan-Meier estimator for the analysis and testing of cure model data is discussed and results are compared with the more commonly used logrank test. The Kaplan-Meier estimator is particularly appropriate for describing the shape of the underlying survival distribution of data from bone marrow transplantation but is also relevant for different types of timed events data from other chronic diseases. The estimate of the event-free fraction by the product limit, may have a bias depending on the extent of follow-up, the parameters of the model, and confounding by competing risk events. The test based on the product limit has appropriate size and is more powerful as long as follow-up is sufficient to minimize the bias. On the other hand, the logrank test, which is optimal for testing differences in time-to-event curves under proportional hazard assumptions, may be inappropriate and misleading for evaluating the difference in event-free fractions under the cure model.
本文讨论了使用Kaplan-Meier估计量分析和检验治愈模型数据，并将结果与更常用的对数秩检验进行比较。Kaplan-Meier估计量特别适合描述骨髓移植数据潜在生存分布的形状，但也适用于来自其他慢性疾病的不同类型的定时事件数据。通过乘积限估计无事件分数可能会有偏差，这取决于随访时间、模型参数以及竞争风险事件的混杂情况。只要随访时间足够长以尽量减少偏差，基于乘积限的检验就具有合适的规模且更具效力。另一方面，对数秩检验在比例风险假设下对于检验事件发生时间曲线的差异是最优的，但对于评估治愈模型下无事件分数的差异可能不合适且具有误导性。
