Development of voice digital biomarkers in AD, their associations with CSF amyloid beta (Aß1‐42) and their underlying neural representations. (31st December 2021)

• Record Type: Journal Article
• Authors: Hajjar, Ihab; Choi, Jinho; Moore, Elliot; Calhoun, Vince D.; Abrol, Anees; Goldstein, Felicia C.
• Is Part Of: Alzheimer's & dementia. Volume 17(2021)Supplement 5
• Page Start: n/a
• Page End: n/a
• Publication Date: 2021-12-31
• 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.056549
• Languages: English
• ISSNs: 1552-5260

Abstract

Abstract: Background: Subtle changes in connected speech may be detectable years before symptomatic phases of Alzheimer's disease (AD). In this study we combine machine learning with innovative natural language processing (NLP) and automatic speech recognition (ASR) to develop digital voice biomarkers. We investigate their associations with levels of Ab (1‐42) and map the significant features to underlying brain networks. Method: Data for the current case‐control analysis were drawn from participants in the Brain Stress Hypertension and Aging Research Program (B‐SHARP) at Emory University. B‐SHARP participants undergo cognitive assessments, neuroimaging, and lumbar punctures as well as annual audio recordings to capture connected speech. A picture description task was included for NLP analysis In addition, participants were asked to describe the room they were in and to list events they went through from the moment they arrived at the study site. Each audio recording went through 2 NLP pipelines to obtain lexical‐semantics and acoustic features. Result: The sample included 206 B‐SHARP participants (age=65.1 years, 51% African Americans, 42% with mild cognitive impairment (MCI), 61% women, follow‐up period =2 years). NLP analysis and to a lesser extent acoustic analysis outperformed traditional cognitive screening in identifying MCI status. NLP based analysis outperformed other screening tools of Ab positive status (defined as <250 pg/ml) status especially in the MCI group. Area under the curve (AUC) for identifying MCI, Ab positive, and MCI/Ab positive and NC/Ab positive vs negative. Results are in the Table. We then assessed the association between baseline digital voice biomarkers and disease progression reflected by CDR‐SOB change/2 years. Both NLP‐based scores (Beta= ‐0.18;p=0.0097) and acoustic scores (beta=0.26, p=0.009) were associated with 2‐year change in CDR‐SOB after adjusting for baseline demographics. Finally functional brain connectivity, based on resting state MRI, was associated with digital voice NLP‐based scores in multiple brain networks (Figure). Conclusion: A brief voice recording analysis detected cognitive status, increased likelihood of identifying Abeta positivity, and predicted disease progression over 2 years. Our protocol for digital biomarkers had an underlying neural representation in language‐related neural networks. Our study provides multi‐faceted evidence for validity of using voice recording as a tool for biomarker and cognitive screening.