Clinical Manifestations.

Background: Automated analysis of natural speech is emerging as a promising digital biomarker of Alzheimer's disease (AD). As speech is a complex process, relying on multiple interacting cognitive functions, fine-grained analysis of speech may have the potential to capture subtle cognitive deficits in the very early stages of AD. Here, we examined the association between amyloid-beta (Aβ) pathology and acoustic speech characteristics in a group of cognitively normal Dutch adults.

Method: We included 50 cognitively normal (n = 23 Aβ-positive, 46%) older adults from three clinical cohorts at the Alzheimer Center Amsterdam (Table 1). Aβ-status was based on local cut-offs for Aβ-concentrations in cerebrospinal-fluid or visual inspection of amyloid positron emission tomography (PET)-imaging. In a five day remote burst assessment, seventeen tablet-based tasks (picture description, journal-prompt storytelling, verbal-fluency) were administered, with a daily administration time of approximately 10 minutes. Various acoustic features, including silent pauses, pause-to-word-ratio, and perceived loudness, were extracted from the voice recordings. Mean burst scores were calculated over the five-day period. The association between Aβ-pathology and acoustic features was investigated using linear regression models separately for each subtask.

Results: Pause-to-word ratio was higher in Aβ-positive individuals in the repetitive picture description subtask (β = -0.05, p = 0.04) and the journaling subtask (β = -0.07, p = 0.03) compared to the Aβ-negative individuals. Although statistical significance between groups was not reached for other acoustic features, a pattern was observed that the Aβ-positive group consistently displayed more medium pauses, longer audio duration, more local jitter, higher intensity variance, a higher fundamental frequency, and fewer long pauses in all multi-day subtasks (Figure 1).

Conclusion: Our results demonstrated that Aβ-pathology was associated with more pauses during free speech in cognitively healthy adults. This supports the notion that remote multi-day speech assessments have the potential to monitor cognitive changes, for example in observational studies and decentralized clinical trials in preclinical AD. Next steps include examining the association between Aβ-pathology and linguistic features using the same sample.