Presentation

Search Abstracts | Symposia | Slide Sessions | Poster Sessions | Poster Slams

Longitudinal analysis of speech and language measures of patients with mild cognitive impairment and amyloid positivity

Poster C10 in Poster Session C, Friday, October 7, 10:15 am - 12:00 pm EDT, Millennium Hall

Sunghye Cho¹, Naomi Nevler², Katheryn Cousins², Sharon Ash², Sanjana Shellikeri², Galit Agmon², Carmen Gonzalez-Recober², Mark Liberman¹, Murray Grossman²; ¹Linguistic Data Consortium, University of Pennsylvania, ²Penn Frontotemporal Degeneration Center, University of Pennsylvania

Introduction: Previous studies have shown that patients with mild cognitive impairment (MCI) show language impairments, including deficits in lexical semantic retrieval, word comprehension, and verbal fluency. However, previous studies on language impairment in MCI rarely use a biological definition of MCI due to Alzheimer’s disease (AD), indicated by amyloid-beta positive (Aβ+) biomarkers, limiting the validity of language measures for longitudinal monitoring of disease progression in AD. In this study, we investigated longitudinal speech samples provided by patients with MCI Aβ+ to examine how their language performance changes over time compared to that of cognitively normal (CN) volunteers. Methods: We analyzed longitudinal speech samples of digitized picture descriptions produced by 117 CN participants and 11 patients with MCI who were Aβ+ based on CSF Aβ42 levels (<192). Groups were matched on age and sex ratio, but not on education level (p=0.033). Recordings were orthographically transcribed and analyzed with our automated lexical and acoustic pipelines. The lexical pipeline tagged part-of-speech (POS) of all words and rated words for several lexical measures (e.g., concreteness, frequency). The acoustic pipeline segmented recordings into speech and pause segments and calculated several durational measures. For each speech or language measure, we built a linear mixed-effects regression model, where the speech measure was included as a dependent variable, and time, group, and the interaction of time and group were included as fixed effects. Individual participants were considered as a random effect, and picture type and education level were included as fixed effects. The CN group was the reference group in all models. Results: The MCI Aβ+ patients produced fewer grammatical particles over time (group x time: β=-0.07, p=0.016), whereas the CN participants produced more particles over time (β=0.06, p=0.031). Also, patients with MCI Aβ+ produced more partial words over time (group x time: β=0.15, p<0.001), whereas partial word count of the CN participants did not change over time (p=0.52). Similarly, patients with MCI Aβ+ repeated words more frequently over time (group x time: β=0.23, p<0.001), whereas the repetition frequency of CN participants did not change (p=0.78). The groups did not differ on other lexical measures. As for the durational measures, patients with MCI Aβ+ paused longer on average over time (group x time: β=0.02, p<0.001), whereas the mean pause duration of the CN participants decreased (β=0.01, p=0.007). The MCI Aβ+ group exhibited less speaking time out of total duration over time (group x time: β=-0.35, p=0.018), whereas the proportion of speech for the CN group did not change over time (p=0.488). Lastly, the MCI Aβ+ patients produced fewer words per minute over time (group x time: β=2.26, p<0.001), whereas the CN group produced more words per minute over time (p=0.002). Conclusion: Speech and language features extracted from one-minute picture descriptions produced by patients with MCI Aβ+ showed different longitudinal trajectories compared to those of the CN participants, suggesting that our automated measures may provide objective, non-invasive, and sensitive speech biomarkers for longitudinal monitoring of patients with MCI and for detection of early-stage AD.

Topic Areas: Disorders: Acquired, Language Production