Automatic analysis and validation of digitized speech markers in Lewy body spectrum diseases with Alzheimer's disease co‐pathology. (December 2021)

• Record Type: Journal Article
• Authors: Shellikeri, Sanjana; Cho, Sunghye; Liberman, Mark Y.; Howard, Erica; Balganorth, Yvonne; Weintraub, Daniel; Lee, Eddie B.; Trojanowski, John Q.; Irwin, David J.; Grossman, Murray; Nevler, Naomi
• Is Part Of: Alzheimer's & dementia. Volume 17(2021)Supplement 4
• Page Start: n/a
• Page End: n/a
• Publication Date: 2021-12
• Subjects: Alzheimer's disease -- Periodicals; Alzheimer Disease -- Periodicals; Dementia -- Periodicals; Démence; Maladie d'Alzheimer; Périodique électronique (Descripteur de forme); Ressource Internet (Descripteur de forme); 616.83
• DOI: 10.1002/alz.053264
• Languages: English
• ISSNs: 1552-5260

Abstract

Abstract: Background: Lewy Body Spectrum Disorders (LBSD) are hallmarked by α‐synucleinopathy. Up to 50% of LBSD have co‐occurring Alzheimer's Disease (AD) at autopsy and this is associated with earlier dementia onset, greater cognitive impairments, and reduced survival. In vivo identification of AD co‐pathology is currently achievable with invasive cerebrospinal fluid (CSF) AD biomarkers or costly molecular PET imaging. There remains a need for non‐invasive, inexpensive, and validated markers that can serve as screening tools for this purpose. We examined automated speech markers of a short semi‐structured natural speech sample in LBSD with and without neurobiologically‐validated AD co‐pathology. Method: We analyzed digitized speech of Cookie Theft picture descriptions from 62 LBSD patients and 44 healthy controls (HC). AD co‐pathology was determined by neuropathological (25%) or CSF based diagnosis (75%): n=18 LBSD+AD (age= 73.7y (7.4), 67% Parkinson Disease (PD) /PD with Dementia (PDD), 33% Dementia with Lewy Bodies (DLB)); n=44 LBSD‐AD (age= 67.8y (6.5), 61% PD/PDD, 39% DLB). Speech samples were automatically parsed into voiced and silent segments and duration measures were extracted. Transcripts were processed for parts‐of‐speech usage and lexical‐semantic features. We conducted group comparisons of speech measures, examined their effects on determining pathologic group, and assessed relations with MRI cortical thickness. Result: LBSD+AD and LBSD‐AD groups did not differ in disease duration, motor disease severity (UPDRS Part III), nor neuropsychological performance (overall and across individual cognitive domains). Both LBSD+AD and LBSD‐AD spoke in shorter speech segments and longer and more frequent pauses; and speaking rate was slower compared to HC. Compared to LBSD‐AD, speakers with LBSD+AD used shorter words (by phoneme count) acquired at a younger age, and produced more conjunctions per 100 words. The features showed excellent model fit for determining pathologic group (McFadden R 2 = 0.23), but speech features did not distinguish between PDD and DLB. Reduced word age of acquisition related to atrophy in left and right temporal regions in LBSD+AD only. Conclusion: Automated speech measures distinguished biologically‐confirmed LBSD+AD from LBSD‐AD regardless of clinical phenotype. Findings highlight the potential use of automated digitized speech analytics as a valid screening tool for AD co‐pathology in LBSD.