Tang An-Min, Tang Nian-Sheng, Zhu Hongtu
Key Laboratory of Statistical Modeling & Data Analysis of Yunnan Province, Yunnan University, 650091, Kunming, China.
Department of Biostatistics, University of North Carolina at Chapel Hill, Chapel Hill, U.S.A.
Stat Med. 2017 Apr 30;36(9):1476-1490. doi: 10.1002/sim.7211. Epub 2017 Jan 9.
The normality assumption of measurement error is a widely used distribution in joint models of longitudinal and survival data, but it may lead to unreasonable or even misleading results when longitudinal data reveal skewness feature. This paper proposes a new joint model for multivariate longitudinal and multivariate survival data by incorporating a nonparametric function into the trajectory function and hazard function and assuming that measurement errors in longitudinal measurement models follow a skew-normal distribution. A Monte Carlo Expectation-Maximization (EM) algorithm together with the penalized-splines technique and the Metropolis-Hastings algorithm within the Gibbs sampler is developed to estimate parameters and nonparametric functions in the considered joint models. Case deletion diagnostic measures are proposed to identify the potential influential observations, and an extended local influence method is presented to assess local influence of minor perturbations. Simulation studies and a real example from a clinical trial are presented to illustrate the proposed methodologies. Copyright © 2017 John Wiley & Sons, Ltd.
测量误差的正态性假设是纵向数据和生存数据联合模型中广泛使用的分布，但当纵向数据呈现偏态特征时，可能会导致不合理甚至误导性的结果。本文通过将非参数函数纳入轨迹函数和风险函数，并假设纵向测量模型中的测量误差服从偏态正态分布，提出了一种用于多变量纵向数据和多变量生存数据的新联合模型。开发了一种蒙特卡罗期望最大化（EM）算法，结合惩罚样条技术和吉布斯采样器中的Metropolis-Hastings算法，以估计所考虑联合模型中的参数和非参数函数。提出了案例删除诊断措施来识别潜在的有影响观测值，并提出了一种扩展的局部影响方法来评估微小扰动的局部影响。给出了模拟研究和一个来自临床试验的实际例子，以说明所提出的方法。版权所有© 2017约翰威立父子有限公司。
