Jin Wei, Gao Qiuxin, Xu Yanxun
Department of Applied Mathematics and Statistics, Johns Hopkins University, Baltimore, Maryland, USA.
Division of Biostatistics and Bioinformatics, Johns Hopkins School of Medicine, Baltimore, Maryland, USA.
Stat Med. 2025 Mar 30;44(7):e70071. doi: 10.1002/sim.70071.
Testing differences in longitudinal trajectories among distinct groups of population is an important task in many biomedical applications. Motivated by an application in Alzheimer's disease, we develop BayTetra, an innovative Bayesian semiparametric approach for estimating and testing group differences in multivariate longitudinal trajectories. BayTetra jointly models multivariate longitudinal data by directly accounting for correlations among different responses, and uses a semiparametric framework based on B-splines to capture the non-linear trajectories with great flexibility. To avoid overfitting, BayTetra encourages parsimonious trajectory estimation by imposing penalties on the smoothness of the spline functions. The proposed method converts the challenging task of hypothesis testing for longitudinal trajectories into a more manageable equivalent form based on hypothesis testing for spline coefficients. More importantly, by leveraging posterior inference with natural uncertainty quantification, our Bayesian method offers a more robust and straightforward hypothesis testing procedure than frequentist methods. Extensive simulations demonstrate BayTetra's superior performance over alternatives. Applications to the Biomarkers of Cognitive Decline Among Normal Individuals (BIOCARD) study yield interpretable and valuable clinical insights. A major contribution of this paper is that we have developed an R package BayTetra, which implements the proposed Bayesian semiparametric approach and is the first publicly available software for hypothesis testing on trajectory differences based on a flexible modeling framework.
在许多生物医学应用中，检验不同人群组之间纵向轨迹的差异是一项重要任务。受阿尔茨海默病应用的启发，我们开发了BayTetra，这是一种创新的贝叶斯半参数方法，用于估计和检验多变量纵向轨迹中的组间差异。BayTetra通过直接考虑不同响应之间的相关性，对多变量纵向数据进行联合建模，并使用基于B样条的半参数框架来灵活地捕捉非线性轨迹。为避免过拟合，BayTetra通过对样条函数的平滑度施加惩罚来鼓励简约的轨迹估计。所提出的方法将纵向轨迹的假设检验这一具有挑战性的任务转化为基于样条系数假设检验的更易于管理的等效形式。更重要的是，通过利用带有自然不确定性量化的后验推断，我们的贝叶斯方法提供了比频率主义方法更稳健、更直接的假设检验程序。广泛的模拟证明了BayTetra相对于其他方法的优越性能。应用于正常个体认知衰退生物标志物（BIOCARD）研究产生了可解释且有价值的临床见解。本文的一个主要贡献是我们开发了一个R包BayTetra，它实现了所提出的贝叶斯半参数方法，并且是第一个基于灵活建模框架进行轨迹差异假设检验的公开可用软件。
