Meta-analytic Gaussian Network Aggregation.

Area of Science:

• Network Science
• Statistical Modeling
• Psychiatric Epidemiology

Background:

• Estimating Gaussian graphical models (GGMs) is crucial but challenging due to limited sample sizes and debated generalizability.
• Existing methods for comparing networks across studies often neglect cross-study heterogeneity.
• Insufficient sample sizes hinder accurate estimation of complex network structures.

Purpose of the Study:

• To develop and present methods for estimating GGMs by aggregating data from multiple datasets.
• To introduce meta-analytic Gaussian network aggregation (MAGNA) as a novel approach for network estimation.
• To address limitations in current network comparison methods by accounting for cross-study heterogeneity.

Main Methods:

• Developed a general maximum likelihood estimation framework to embed various GGM estimation models.
• Introduced two variants of MAGNA: fixed-effects (ignoring heterogeneity) and random-effects (modeling heterogeneity).
• Assessed MAGNA's performance through large-scale simulation studies.

Main Results:

• MAGNA demonstrates effective GGM estimation by aggregating information across multiple datasets.
• The random-effects MAGNA variant successfully accounts for and models heterogeneity between studies.
• Simulations confirm the robustness and performance of the proposed MAGNA methods.

Conclusions:

• MAGNA offers a powerful framework for robust GGM estimation, particularly when dealing with multiple, potentially heterogeneous datasets.
• The method enhances the generalizability and replicability of network structures derived from aggregated data.
• Application to PTSD symptom data highlights MAGNA's utility in real-world psychiatric research.