학술논문
Predicting Cognitive Impairment in Cerebrovascular Disease Using Spoken Discourse Production
Document Type
Journal Articles
Reports - Research
Reports - Research
Author
Roberts, Angela; Aveni, Katharine; Basque, Shalane; Orange, Joseph B.; McLaughlin, Paula; Ramirez, Joel; Troyer, Angela K.; Gutierrez, Stephanie; Chen, Angie; Bartha, Robert; Binns, Malcolm A.; Black, Sandra E.; Casaubon, Leanne K.; Dowlatshahi, Dar; Hassan, Ayman; Kwan, Donna; Levine, Brian; Mandzia, Jennifer; Sahlas, Demetrios J.; Scott, Christopher J. M.; Strother, Stephen; Sunderland, Kelly M.; Symons, Sean; Swartz, Richard
Source
Subject
Language
English
ISSN
0271-8294
Abstract
Purpose: Dementia due to cerebrovascular disease (CVD) is common. Detecting early cognitive decline in CVD is critical because addressing risk factors may slow or prevent dementia. This study used a multidomain discourse analysis approach to determine the spoken language signature of CVD-related cognitive impairment. Method: Spoken language and neuropsychological assessment data were collected prospectively from 157 participants with CVD as part of the Ontario Neurodegenerative Disease Research Initiative, a longitudinal, observational study of neurodegenerative disease. Participants were categorized as impaired (n = 92) or cognitively normal for age (n = 65) based on neuropsychology criteria. Spoken language samples were transcribed orthographically and annotated for 13 discourse features, across five domains. Discriminant function analyses were used to determine a minimum set of discourse variables, and their estimated weights, for maximizing diagnostic group separation. Results: The optimal discriminant function that included 10 of 13 discourse measures correctly classified 78.3% of original cases (69.4% cross-validated cases) with a sensitivity of 77.2% and specificity of 80.0%. Conclusion: Spoken discourse appears to be a sensitive measure for detecting cognitive impairment in CVD with measures of productivity, information content, and information efficiency heavily weighted in the final algorithm. [This article was written with ONDRI Investigators.]