Network disruption in post-stroke depressive symptoms: a Bayesian multicohort study

Post-stroke depressive symptoms are common, but studies on the contributions of lesion locations to depression symptoms remain inconsistent. Bayesian statistics offer the possibility to clarify these inconsistencies by providing evidence in favor and against observed associations. We reanalyzed four previously published observational cohorts with 1,864 patients with stroke characterized for depressive symptoms from France, Korea, Germany and the USA. We performed voxel-wise Bayes factor mapping using general linear models. Lesion maps were used to test for associations of post-stroke depressive symptoms with lesion location. Network level associations were tested based on structural disconnection and functional lesion network maps estimated by reference to normative connectome data. Due to a sample size with broad lesion coverage, Bayesian inference was informative in large parts of the brain. Lesion associations were found in clusters in the right hemisphere including cortical (frontal, insular, parietal) and subcortical regions (striatum, amygdala, hippocampus) and adjacent white matter. Evidence of no contribution to post-stroke depressive symptoms was observed in large parts of the left hemisphere. Structural disconnection was linked to depressive symptoms in the bilateral frontal white matter prominently involving the corpus callosum and anterior thalamic radiation. Lesion network mapping revealed evidence of an association of post-stroke depressive symptoms with network connectivity in the left posterior cingulate cortex and precuneus, and less consistently, the bilateral anterior insula, along with widespread absence of associations in other regions. Depressive symptoms were most strongly associated with right frontal lesions and right-dominant structural disconnection in frontal cortico-subcortical circuits.