Title: Efficient Implementation of a Semiparametric Joint Model for Multivariate Longitudinal Biomarkers and Competing Risks Time-to-Event Data Show Chinese title

Title: 半参数联合模型在多变量纵向生物标志物和竞争风险时间至事件数据分析中的高效实现

Abstract: Joint modeling has become increasingly popular for characterizing the association between one or more longitudinal biomarkers and competing risks time-to-event outcomes. However, semiparametric multivariate joint modeling for large-scale data encounter substantial statistical and computational challenges, primarily due to the high dimensionality of random effects and the complexity of estimating nonparametric baseline hazards. These challenges often lead to prolonged computation time and excessive memory usage, limiting the utility of joint modeling for biobank-scale datasets. In this article, we introduce an efficient implementation of a semiparametric multivariate joint model, supported by a normal approximation and customized linear scan algorithms within an expectation-maximization (EM) framework. Our method significantly reduces computation time and memory consumption, enabling the analysis of data from thousands of subjects. The scalability and estimation accuracy of our approach are demonstrated through two simulation studies. We also present an application to the Primary Biliary Cirrhosis (PBC) dataset involving five longitudinal biomarkers as an illustrative example. A user-friendly R package, \texttt{FastJM}, has been developed for the shared random effects joint model with efficient implementation. The package is publicly available on the Comprehensive R Archive Network: https://CRAN.R-project.org/package=FastJM. Show Chinese abstract

Abstract: 联合建模已成为表征一个或多个纵向生物标志物与竞争风险事件时间结局之间关联的流行方法。然而，对于大规模数据，半参数多元联合建模面临显著的统计和计算挑战，主要由于随机效应的高维性和非参数基线风险估计的复杂性。这些挑战通常导致计算时间延长和内存使用过度，限制了联合建模在生物样本库规模数据集中的实用性。本文介绍了一种基于期望最大化（EM）框架的高效半参数多元联合模型实现，该实现通过正态近似和定制化的线性扫描算法支持。我们的方法显著减少了计算时间和内存消耗，使得分析数千个受试者的数据成为可能。通过两个模拟研究验证了我们方法的可扩展性和估计准确性。此外，我们还展示了对包含五个纵向生物标志物的原发性胆汁性肝硬化（PBC）数据集的应用实例。为共享随机效应联合模型提供了一个用户友好的R包\texttt{FastJM}，并实现了高效实施。该包已在综合R归档网络上公开发布：https://CRAN.R-project.org/package=FastJM。