Federated Statistical Harmonization and Multivariate Modeling of Brain Aging Trajectories in Severe Mental Diseases

About This Grant

This project aims to develop federated multivariate harmonization and neuroimaging data analysis frameworks to investigate altered brain aging patterns in individuals with severe mental illness (SMI) and assess the modifying effects of metabolic syndrome (MetS). SMI, including schizophrenia spectrum disorders, bipolar disorder, and major depressive disorder, affects nearly 1 in 25 U.S. adults, is associated with reduced quality of life, and has been linked to accelerated brain aging and increased risk of early-onset dementia. Characterizing brain aging in SMI is critical for understanding its underlying neurobiological mechanisms and reducing excess morbidity and mortality in this population. Individuals with SMI face a 50% higher risk of developing MetS, a cluster of cardiometabolic risk factors associated with reduced brain health. However, how MetS modifies brain aging trajectories in SMI remains poorly understood. This project will address this knowledge gap by leveraging the ENIGMA (Enhancing NeuroImaging Genetics through Meta-Analysis) Consortium, which aggregates neuroimaging data from tens of thousands of participants across 47 countries. Despite ENIGMA’s unparalleled scope and population heterogeneity, key methodological challenges remain: (1) its distributed structure restricts data sharing to only summary-level information; (2) residual site-level heterogeneity from scanner and acquisition differences can bias results; and (3) high-dimensional, interdependent imaging features complicate statistical inference and reduce replicability of research findings. The overarching goal of this project is to develop novel, federated statistical methods that enable reliable and replicable neuroimaging analyses in large-scale, distributed collaborations, allowing participating sites to harmonize neuroimaging data and collaboratively build analytical models without sharing individual-level information. This project focuses on the following aims: Aim 1 will develop a network-aware multivariate harmonization strategy that identifies and leverages the covariance network structure among neuroimaging features to guide site-effect removal and design a summary-statistics-based federated learning method for distributed data. Aim 2 will develop a joint modeling framework that incorporates latent factor modeling to capture shared variations across neuroimaging outcomes and decorrelates the test statistic for robust multiple testing correction, and a federated learning method. In Aim 3, we will apply these methods to ENIGMA data from distributed sites to identify brain areas with abnormal aging patterns in SMI and further modified by MetS and releases user-friendly software to support scalable, distributed neuroimaging workflows. The proposed methods are broadly applicable beyond this study, enabling summary-statistics-based neuroimaging data integration, facilitating large-scale neuroscience collaborations and expanding opportunities for broader scientific partnerships.