In survival analysis, the accelerated failure time model is a useful alternative to the popular Cox proportional hazards model due to its easy interpretation. Current estimation methods for the accelerated failure time model mostly assume independent and identically distributed random errors, but in many applications the conditional variance of log survival times depend on covariates exhibiting some form of heteroscedasticity. In this paper, we develop a local Buckley-James estimator for the accelerated failure time model with heteroscedastic errors. We establish the consistency and asymptotic normality of the proposed estimator and propose a resampling approach for inference. Simulations demonstrate that the proposed method is flexible and leads to more efficient estimation when heteroscedasticity is present. The value of the proposed method is further assessed by the analysis of a breast cancer data set.

中文翻译：

---

异方差加速失效时间模型的局部 Buckley-James 估计

在生存分析中，加速故障时间模型是流行的 Cox 比例风险模型的有用替代方案，因为它易于解释。当前加速故障时间模型的估计方法大多假设独立且同分布的随机误差，但在许多应用中，对数生存时间的条件方差取决于表现出某种形式的异方差的协变量。在本文中，我们为具有异方差误差的加速失效时间模型开发了一个局部巴克利-詹姆斯估计器。我们建立了所提出估计量的一致性和渐近正态性，并提出了一种用于推理的重采样方法。模拟表明，当存在异方差时，所提出的方法是灵活的，并且可以更有效地进行估计。