Functional Connectivity Heterogeneity and Consequences for Clinical and Cognitive Prediction

Functional connectivity is frequently used to assess dynamic brain functioning and predict individual differences in behavioral outcomes, such as psychopathology. Inferences from functional connectivity analyses typically rely on group-averaged model statistics. However, heterogeneity between individuals may lead to group-level models that poorly reflect each individual. Poor individual-level precision may limit the ability to make individual-level predictions, which is necessary for key goals such as clinical translation. This registered report examined between-person heterogeneity in resting-state functional connectivity strength patterns by assessing similarity between group- and individual-level connectivity models in the Adolescent Brain Cognitive Development study. Using intraclass correlation coefficients, we found that a group-averaged region-of-interest connectivity model was a poor reflection of every individual. In contrast, a group-averaged model of between- and within-network connectivity was a good representation of most individuals. We then examined how individual-level distinctness from the group moderated predictive performance of several clinical and neurocognitive scales. We found that higher similarity to the group improved prediction of performance on the pattern comparison test but did not significantly moderate prediction of other outcomes. Results suggest that region-of-interest-based functional connectivity networks are highly heterogeneous and group-based models are inappropriate for individual-level inferences, but that network-based connectivity is largely similar across individuals. Additionally, we provide preliminary but modest evidence of the impacts of heterogeneity on prediction that future studies should build on.