Prediction of Cognitive Recovery After Stroke: The Value of Diffusion-Weighted Imaging-Based Measures of Brain Connectivity.

Background And Purpose: Prediction of long-term recovery of a poststroke cognitive disorder (PSCD) is currently inaccurate. We assessed whether diffusion-weighted imaging (DWI)-based measures of brain connectivity predict cognitive recovery 1 year after stroke in patients with PSCD in addition to conventional clinical, neuropsychological, and imaging variables.

Methods: This prospective monocenter cohort study included 217 consecutive patients with a clinical diagnosis of ischemic stroke, aged ≥50 years, and Montreal Cognitive Assessment score below 26 during hospitalization. Five weeks after stroke, patients underwent DWI magnetic resonance imaging. Neuropsychological assessment was performed 5 weeks and 1 year after stroke and was used to classify PSCD as absent, modest, or marked. Cognitive recovery was operationalized as a shift to a better PSCD category over time. We evaluated 4 DWI-based measures of brain connectivity: global network efficiency and mean connectivity strength, both weighted for mean diffusivity and fractional anisotropy. Conventional predictors were age, sex, level of education, clinical stroke characteristics, neuropsychological variables, and magnetic resonance imaging findings (eg, infarct size). DWI-based measures of brain connectivity were added to a multivariable model to assess additive predictive value.

Results: Of 135 patients (mean age, 71 years; 95 men [70%]) with PSCD 5 weeks after ischemic stroke, 41 (30%) showed cognitive recovery. Three of 4 brain connectivity measures met the predefined threshold of <0.1 in univariable regression analysis. There was no added value of these measures to a multivariable model that included level of education and infarct size as significant predictors of cognitive recovery.

Conclusions: Current DWI-based measures of brain connectivity appear to predict recovery of PSCD but at present have no added value over conventional predictors.