Figure 4.

Hemispheric relevance depends on key covariates. Inter-relation between the relevance of lesion load in a given hemisphere (y-axis, left hemisphere: moccasin, right hemisphere: yellow) and marginal posterior parameters of key covariates (x-axis) in predicting cognitive performance. Lesion volume, as well as all four outcome variables were normalized to the same z-score scale. Despite assumed important interactions,³⁷ neither one of typically applied uni- or multivariate models represents an integrative approach to jointly study lesion location and further sociodemographic and clinical covariates. We here present such a joint analysis that is possible based on the combination of our large sample size and generative modelling framework. Top row: The influence of lesion volume on the cognitive outcome prediction varied across the three outcome scores: A one standard unit increase in lesion volume was sufficient to cause a 21% standard unit decrease in MMSE performance. Conversely, one standard unit increases in lesion volume led to only 11–14% standard unit decreases in BN and SVL performances. RC is not shown, as lesion volume was not included as covariate in RC models. Middle row: Years of education had positive effects of comparable magnitude on MMSE, BN and SVL outcome scores: An additional year of education predicted a 5–7% standard unit increase. This effect was even more pronounced for RC, where an additional year of education predicted an increase of 8.5%. Bottom row: One point more on the IQCODE scale, i.e. a higher pre-stroke cognitive decline, predicted a ∼65% drop in BN, SVL and RC standard unit performance. In case of MMSE, on more IQCODE point resulted in an even higher decrease of 77% standard unit performance.