可加风险模型下现状数据的降维问题

Dimension Reduction for Additive Risk Model with Current Status Data

文章基于可加风险模型假设,采用偏最小二乘回归和有监督的主成分回归两种投影降维方法,研究了高维协变量情况下现状数据的降维问题。通过深入地模拟试验,对比两种降维方法在高维相关现状数据的生存预测方面的表现,最后将两种降维方法结合实际数据集进行实证分析。模拟和实证结果表明这两种降维方法能很好地处理具有高维、强相关协变量的小样本数据集,比如基因微阵列数据。在后续的研究中,有望将现状数据扩展至其它更一般的区间删失数据。

This paper considered the regression analysis of high-dimensional current status data for the additive hazard model.The partial least square regression and the supervised principal component regression are adopted to reduce the dimension of covariates.The performance of these two methods in survival prediction are compared through the simulation experiments.Finally,two real datasets are analyzed with the proposed methods of this paper.The simulation result shows that two dimensionality reduction methods can deal with small sample covariate data with high dimension and strong correlation,such as gene microarray data.In the follow-up study,these methods may be extended from the current status data to other general interval-censored data.