Zou Haotian, Lutz Michael W, Welsh-Bohmer Kathleen, Luo Sheng
Department of Biostatistics & Bioinformatics, Duke University, Durham, North Carolina, USA.
Department of Neurology, Duke University School of Medicine, Durham, North Carolina, USA.
Alzheimers Dement. 2025 Sep;21(9):e70094. doi: 10.1002/alz.70094.
Accurate prediction of Alzheimer's disease (AD) dementia onset and progression to mild cognitive impairment (MCI) is crucial for early intervention and clinical trial design. This study presents a predictive framework leveraging Bayesian model averaging (BMA) with a multivariate functional mixed model (MFMM) to integrate multivariate longitudinal outcomes and survival data.
The training cohort included 1012 participants from the Alzheimer's Disease Neuroimaging Initiative (ADNI). The validation cohort comprised 2087 participants from the National Alzheimer's Coordinating Center (NACC). BMA methods, including stacking and pseudo-BMA+, aggregated predictions across candidate models to enhance accuracy and robustness. Predictive performance was evaluated using the C-index, a measure of discrimination.
Compared to the composite model, BMA improved prediction accuracy. The C-index was 0.777 (stacking) and 0.771 (pseudo-BMA+) in ADNI and 0.743 and 0.738 in NACC.
This framework offers a robust tool for personalized medicine, enabling accurate predictions and enhanced generalizability across diverse populations.
We introduced a novel joint modeling framework integrating multivariate longitudinal outcomes (Mini-Mental State Examination and Clinical Dementia Rating Sum of Boxes) with survival data to predict Alzheimer's disease dementia onset and progression. We validated the framework across complementary datasets: Alzheimer's Disease Neuroimaging Initiative (training) and National Alzheimer's Coordinating Center (NACC; validation), with NACC providing a demographically diverse population to assess generalizability. The model enhanced predictive accuracy using Bayesian model averaging, which synthesizes insights across multiple models to reduce uncertainty and improve robustness. The model demonstrated consistent and clinically relevant performance, supporting its applicability for early intervention, precision medicine, and clinical trial design.
准确预测阿尔茨海默病（AD）痴呆症的发病及进展为轻度认知障碍（MCI）对于早期干预和临床试验设计至关重要。本研究提出了一个预测框架，该框架利用贝叶斯模型平均法（BMA）与多变量函数混合模型（MFMM）来整合多变量纵向结果和生存数据。
训练队列包括来自阿尔茨海默病神经影像倡议（ADNI）的1012名参与者。验证队列由来自国家阿尔茨海默病协调中心（NACC）的2087名参与者组成。BMA方法，包括堆叠法和伪BMA +，汇总了候选模型的预测结果，以提高准确性和稳健性。使用C指数（一种区分度指标）评估预测性能。
与复合模型相比，BMA提高了预测准确性。在ADNI中，C指数堆叠法为0.777，伪BMA +法为0.771；在NACC中，C指数分别为0.743和0.738。
该框架为个性化医疗提供了一个强大的工具，能够进行准确预测并增强在不同人群中的通用性。
我们引入了一种新颖的联合建模框架，该框架整合多变量纵向结果（简易精神状态检查表和临床痴呆评定量表总和）与生存数据，以预测阿尔茨海默病痴呆症的发病及进展。我们在互补数据集上验证了该框架：阿尔茨海默病神经影像倡议（训练）和国家阿尔茨海默病协调中心（NACC；验证），NACC提供了人口统计学上多样化的人群以评估通用性。该模型使用贝叶斯模型平均法提高了预测准确性，该方法综合了多个模型的见解以减少不确定性并提高稳健性。该模型展示了一致且与临床相关的性能，支持其在早期干预、精准医疗和临床试验设计中的适用性。
