Ge Lei, Li Yang, Sun Jianguo
Department of Biostatistics and Health Data Science, Indiana University School of Medicine and Richard M. Fairbanks School of Public Health, Indianapolis, IN, USA.
School of Mathematics and Statistics, KLAS of MOE and Key Laboratory of Big Data Analysis of Jilin Province, Northeast Normal University, Changchun, People's Republic of China.
J Appl Stat. 2024 Nov 19;52(7):1423-1445. doi: 10.1080/02664763.2024.2428266. eCollection 2025.
In health and clinical research, panel binary data from recurrent events arise when subjects are surveyed to report occurrence statuses of recurrent events over fixed observation windows. In practice, such data can be cut short by a dependent failure event such as death. For the analysis of panel binary data, tools from generalized linear models overlook the recurrence nature of panel binary data, and other relevant literature does not accommodate the failure time. Motivated by the hospitalization data surveyed from the Health and Retirement Study, we propose a semiparametric joint-modeling-based procedure for analyzing panel binary data with a dependent failure time. For model fitting, we develop a computationally efficient EM algorithm and show the resulting estimates are consistent and asymptotically normal. Theoretical results are provided to enable valid inferences. Simulation studies have confirmed the performance of the proposed method in practical settings. The method is applied to assess important risk factors associated with incidences of hospitalization among the working elderly.
在健康和临床研究中，当对受试者进行调查以报告在固定观察期内复发事件的发生状态时，就会产生来自复发事件的面板二元数据。在实际中，此类数据可能会因诸如死亡等相依失效事件而提前终止。对于面板二元数据的分析，广义线性模型的工具忽略了面板二元数据的复发性质，而其他相关文献并未考虑失效时间。受健康与退休研究调查的住院数据启发，我们提出了一种基于半参数联合建模的程序，用于分析具有相依失效时间的面板二元数据。对于模型拟合，我们开发了一种计算效率高的期望最大化（EM）算法，并表明所得估计是一致的且渐近正态。提供了理论结果以进行有效的推断。模拟研究证实了所提方法在实际环境中的性能。该方法被应用于评估与在职老年人住院发生率相关的重要风险因素。
