Meta-analysis of generalized additive models in neuroimaging studies

Area of Science:

• Neuroimaging
• Statistical Genetics
• Computational Neuroscience

Background:

• Analyzing multiple neuroimaging studies increases statistical power but sharing data is often impractical due to privacy and proprietary concerns.
• Existing meta-analytic tools primarily support parametric models, which are often inadequate for capturing complex, non-linear relationships in neuroimaging data, such as age-brain associations.
• There is a need for advanced meta-analytic methods capable of handling semi-parametric models to better analyze neuroimaging datasets.

Purpose of the Study:

• To introduce meta-GAM, a novel method for the meta-analysis of generalized additive models (GAMs) that does not require individual participant data.
• To extend existing meta-analytic capabilities by enabling the analysis of multiple model terms and multivariate smooth functions.
• To provide a robust framework for meta-analysis in neuroimaging that respects data privacy and regulatory constraints.

Main Methods:

• Developed meta-GAM, a method for meta-analyzing generalized additive models, suitable for distributed data.
• Extended capabilities to include analysis of multiple model terms and multivariate smooth functions.
• Enabled computation of meta-analytic p-values for smooth terms.

Main Results:

• Simulation experiments demonstrated the strong performance of the proposed meta-GAM methods.
• The method was successfully applied to real-world data, analyzing hippocampal volume and sleep quality from the Lifebrain consortium.
• The accompanying R package 'metagam' facilitates the application of these novel meta-analytic techniques.

Conclusions:

• Meta-GAM offers a powerful solution for increasing statistical power in neuroimaging meta-analyses while preserving data privacy.
• The method is particularly beneficial for lifespan neuroscience and imaging genetics research, where complex relationships are common.
• The availability of the 'metagam' R package promotes the adoption and application of advanced semi-parametric meta-analytic techniques in the field.