评价逆概率删失加权(inverse probability of censoring weighting，IPCW)校正win ratio方法用于两组生存比较的统计性能，为分析带有删失的生存数据提供新的方法选择。采用Monte-Carlo模拟方法，设置等比例风险和非等比例风险下的不同场景，对IPCW校正win ratio法、win ratio法、Cox模型法、log-rank检验法用于生存比较假设检验的统计性能，以模拟样本中P＜0。05的样本比例作为评价指标。对前两种方法参数估计的统计性能，以两组的赢率和赢率比的中位估计值作为评价指标。在等比例风险假设下，4种方法的一类错误率均逼近事先设定的检验水准(0。05);当样本量低于100或删失率大于50％时，IPCW校正win ratio法的检验效能略低于Cox模型法、log-rank检验法，始终高于win ratio法;IPCW校正win ratio法的中位估计值与win ratio法无明显差异。在非等比例风险假设下，从检验效能上看，存在延迟效应时，IPCW校正win ratio法高于win ratio法，低于Cox模型法和log-rank检验法，存在交叉效应时，随着交叉位置不同，几种方法表现不一。从参数估计看，无论在何种情形下，只要删失率不超过50％，该方法均能校正由删失带来的赢率偏倚，并获得无偏的估计，即使删失率超过50％，IPCW校正win ratio法对赢率的校正效果减弱，但仍能维持在较小的偏倚范围内。IPCW校正win ratio法作为一种新的非参数统计方法，可以用于生存数据的比较分析，其统计量能反映组间全局对比赢率情况，且具有较好的无偏性质，可考虑用作非等比例风险时处理效应估计。
Statistical Performance of IPCW Adjusted-win Ratio Method in Survival Comparison
The aim of this study is to evaluate the inverse probability of censoring weighting(IPCW)adjusted win ratio method's statistical performance for the survival comparison between two groups,and provide a novel methodological alternative for analyzing survival data with censoring.Scenarios under proportional and non-proportional hazards were set up to assess the statistical perfor-mance of the IPCW adjusted-win ratio method,win ratio method,Cox model and log-rank test in survival comparisons by Monte-Carlo simulation,the proportion of significant results(P＜0.05)、the median estimates of WR and win proportion under IPCW adjusted-win ratio method and win ratio method were as evaluation metrics.Under the hypothesis of proportional hazards,the type I error rates of four methods were very close to the specified significance level(0.05),when sample size was less than 100 or the censoring rate exceeded 50％,the power of IPCW adjusted-win ratio method was slightly lower than Cox model and log-rank test,yet consistently higher than win ratio method.There was no difference in the median estimates of WR between IPCW adjusted-win ratio method and win ratio method.Under the hypothesis of non-proportional hazards,the power of IPCW adjusted-win ratio method was higher than win ratio method but was lower than Cox model and log-rank test with delayed effects.There was no consistent pattern among the methods with crossover effects.As long as censoring rate was lower than 50％,the IPCW adjusted-win ratio method was able to correct the win proportion bias caused by censoring and obtain an unbiased estimate of WR in any situation.Even when the censoring rate was higher than 50％,the correction effect of IPCW adjusted-win ratio method would be weakened,but WR could still be maintained within a small biased range.As a novel non-parametric statistical approach,IPCW adjusted-win ratio method can be used for the survival comparison.Its statistics can reflect the global comparison of win proportion between groups,and it has a good unbiased estimate,so it can be considered for the estimation of treatment effect under non-proportional hazards.
NETL
NSTL
万方数据
