Characterizing progressive speech changes in prodromal‐to‐mild Alzheimer’s disease using natural language processing

• DOI: 10.1002/alz.063244
• Publication Date: 2022-12-20T16:20:31Z
• Authors: Jessica Robin; Mengdan Xu; Aparna Balagopalan; Jekaterina Novikova; Laura Kahn; Abdi Oday; Mohsen Hejrati; Somaye Hashemifar; Mohammadreza Nega...; Bill Simpson; Edmond Teng
• Keywords: Clinical Dementia Rating

Abstract Background Novel automated tools for analyzing speech and language may provide new insights into Alzheimer’s disease (AD). Although changes occur in AD other neurodegenerative diseases, current clinical assessments to monitor these symptoms can be burdensome have limited sensitivity. Through analyses of open‐ended naturalistic collected from a standardized interview, we developed novel measure characterize progressive AD. Methods We analyzed Clinical Dementia Rating (CDR) recordings subset 101 participants (58F, 43M, mean age=69 years, SD=7) the Tauriel trial semorinemab prodromal‐to‐mild CDR were at baseline, 6‐month, 12‐month 18‐month timepoints. Recordings processed using Winterlight analysis platform which generates >500 acoustic linguistic features. After controlling age, sex level education, identified multiple features that had significant linear effects time (indicating longitudinal change). These combined an unweighted composite score, was compared with endpoints. Results The score included six (related word duration, frequency, syntactic depth, use nouns, pronouns particles) three power spectrum vocal recordings). When endpoints, similar effect size detecting change (β=0.29) CDR‐Sum Boxes (CDR‐SB; β=0.30) Disease Assessment Scale‐Cognitive Subscale (ADAS‐Cog; β=0.22). Notably, it significantly greater Repeatable Battery Neuropsychological Status (RBANS; β=‐0.15, p<0.01) subscales ADAS‐Cog measuring finding difficulty (β=0.12, spoken ability (β=0.09, p<0.01). Conclusions Progressive are detectable early measurable via processing tools. Speech scores potential more sensitive measures progression and/or treatment response speech‐related do not contribute additional patient burden. Further validation is needed replicate findings confirm neuropathological relevance this measure.