The Biostatistics Center, Department of Biostatistics and Bioinformatics, The George Washington University, Washington, MD, USA.
Stat Methods Med Res. 2023 May;32(5):1010-1020. doi: 10.1177/09622802231163333. Epub 2023 Mar 28.
The standard modeling approach for time-to-event outcomes subject to censoring is based on the hazard function, with hazard ratios capturing the effect of exposures on the risk of outcome. The restricted mean survival time, defined as the expected time to event up to a pre-specified time horizon, provides an alternative useful summary of time-to-event outcomes. Restricted mean survival time can be estimated nonparametrically and can be used to compare groups or interventions when the proportional hazards (PHs) assumption does not hold. Moreover, even when the proportional hazards assumption holds, the restricted mean survival time, an additive measure of risk, provides additional information to the hazard ratio, which is a measure of relative risk that can be difficult to interpret in absence of an estimate of the reference risk. Herein, a generalized fiducial approach is proposed for restricted mean survival time, and its asymptotic properties are investigated. Numerical simulations show the proposed approach provides one- and two-sided confidence intervals with coverage probabilities close to nominal values and controls the type-I error for two-group comparisons even for small sample sizes with a low number of events. Data from a type 1 diabetes study is used for illustration.
对于存在删失的生存时间数据，标准的建模方法基于风险函数，使用风险比来捕捉暴露对结局风险的影响。受限平均生存时间定义为在特定时间之前发生事件的预期时间，是一种替代的有用的生存时间结局的总结。受限平均生存时间可以进行非参数估计，并且可以在比例风险（PH）假设不成立时用于比较组或干预措施。此外，即使比例风险假设成立，受限平均生存时间作为风险的加性度量，提供了与风险比相比更多的信息，而风险比是相对风险的度量，在没有参考风险估计的情况下，可能难以解释。在此，提出了一种受限平均生存时间的广义似然比方法，并研究了其渐近性质。数值模拟表明，该方法提供了单侧和双侧置信区间，其覆盖概率接近标称值，并且即使在小样本量和事件数量较少的情况下，也可以控制两样本组比较的Ⅰ类错误。使用 1 型糖尿病研究的数据进行说明。
