In:
Stroke, Ovid Technologies (Wolters Kluwer Health), Vol. 52, No. Suppl_1 ( 2021-03)
Abstract:
Neurocognitive disorder (NCD) appears in 10% of survivors of a first-ever stroke and in 30% after a recurrent one. We aimed to classify clinical- and imaging factors related to the development of major NCD 3 months post stroke, so as to examine whether early development ( 〈 1 yr) is mainly of a neurodegenerative or vascular origin. Based on previous literature we hypothesized that early post-stroke major NCD would mainly be of neurodegenerative origin. 227 stroke survivors (age = 71.69 (11.25), NIHSS = 3.79 (4.75), females (43,61%)) were included from the ‘Norwegian COgnitive Impairment After STroke’ study. Clinical- and MRI data at baseline, and neuropsychological data at baseline and 3 months was used. Cortical thicknesses were automatically measured using Freesurfer and white matter hyperintensity (WMH) volumes were semi-automatically measured using FSL Bianca. Stroke lesion volumes were semi-automatically measured using ITK-snap. Support Vector Machine (SVM) classification was used in order to investigate the prognostic value of the neuroimaging findings and the clinical factors. Model performance was measured using Area Under Receiving Operating Characteristics (ROC) and -Curve (AUC). The best model correctly predicted major NCD in 88% of the patients and was driven by 19 features, with the top five features being stroke volume, WMH volume, left hemisphere occipital- and temporal thickness, and right hemisphere cingulate thickness. In conclusion, clinical- and MRI findings may be used in prediction of cognitive outcome for stroke patients. Various meta-studies of post-stroke major NCD prediction show levels ranging from .48 to .91, showing that the current model is performing well. Early development of major NCD seems dependent not on neurodegenerative causes alone, but also on vascular factors, as well as aspects of the stroke itself. This highlights the complexity of post-stroke NCD and thus also its prediction. Figure 1: ROC curve of SVM model, with an AUC of .88.
Type of Medium:
Online Resource
ISSN:
0039-2499
,
1524-4628
DOI:
10.1161/str.52.suppl_1.P66
Language:
English
Publisher:
Ovid Technologies (Wolters Kluwer Health)
Publication Date:
2021
detail.hit.zdb_id:
80381-9
detail.hit.zdb_id:
1467823-8
