标题： jmstate，用于多状态联合建模的灵活Python包 显示英文标题

标题： jmstate, a Flexible Python Package for Multi-State Joint Modeling

摘要： 经典联合建模方法通常依赖于竞争风险或重复事件公式来考虑涉及演变纵向标记和离散事件发生的复杂现实过程。然而，这些框架通常仅能捕捉潜在事件动态的有限方面。多状态联合模型通过通过可能的转换网络表示完整的事件历史，包括重复循环和终端吸收，所有这些都可能受到纵向协变量的影响，提供了一个更灵活的替代方案。在本文中，我们提出了一种统一纵向生物标志物建模与定义在任意有向图上的多状态事件过程的通用框架。我们的方法容纳了马尔可夫和半马尔可夫转换结构，并通过共享潜在结构将非线性混合效应纵向子模型与多状态生存过程耦合，扩展了经典联合模型。我们推导了完整似然性，并基于随机梯度下降开发了可扩展的推理程序。此外，我们引入了一个动态预测框架，能够在复杂的状态转移轨迹上进行个性化风险评估。为了促进可重复性和传播，我们提供了一个开源的Python库\texttt{jmstate}，该方法学的实现可在\href{https://pypi.org/project/jmstate/}{Python包索引}上获得。模拟实验和一个生物医学案例研究展示了该框架在表示复杂纵向和多状态事件动态方面的灵活性和性能。用于重现实验的完整Python笔记本以及本文的源代码可在\href{https://gitlab.com/felixlaplante0/jmstate-paper/}{GitLab}上获得。 显示英文摘要

摘要： Classical joint modeling approaches often rely on competing risks or recurrent event formulations to account for complex real-world processes involving evolving longitudinal markers and discrete event occurrences. However, these frameworks typically capture only limited aspects of the underlying event dynamics. Multi-state joint models offer a more flexible alternative by representing full event histories through a network of possible transitions, including recurrent cycles and terminal absorptions, all potentially influenced by longitudinal covariates. In this paper, we propose a general framework that unifies longitudinal biomarker modeling with multi-state event processes defined on arbitrary directed graphs. Our approach accommodates both Markovian and semi-Markovian transition structures, and extends classical joint models by coupling nonlinear mixed-effects longitudinal submodels with multi-state survival processes via shared latent structures. We derive the full likelihood and develop scalable inference procedures based on stochastic gradient descent. Furthermore, we introduce a dynamic prediction framework, enabling individualized risk assessments along complex state-transition trajectories. To facilitate reproducibility and dissemination, we provide an open-source Python library \texttt{jmstate} implementing the proposed methodology, available on \href{https://pypi.org/project/jmstate/}{PyPI}. Simulation experiments and a biomedical case study demonstrate the flexibility and performance of the framework in representing complex longitudinal and multi-state event dynamics. The full Python notebooks used to reproduce the experiments as well as the source code of this paper are available on \href{https://gitlab.com/felixlaplante0/jmstate-paper/}{GitLab}.