Identifying diagnostic biomarkers for Delirium and predicting cognitive Outcomes in hospitalized older adults using automated Speech Analysis (IDOSA) - Abstract
Delirium affects up to 50% of hospitalized older patients (75+) and is associated with a 12-fold increased risk
for subsequent development of Alzheimer’s Disease (AD) and AD-related dementias (AD/ADRD) as well as
accelerated cognitive decline in those with pre-existing cognitive impairment. Yet, over 75% of delirium cases
in the hospital setting remain undiagnosed. It is also unknown who will develop persistent cognitive decline
post-hospitalization and who will return to baseline cognition. This project seeks to develop digital speech
biomarkers for delirium diagnosis and prediction (exploratory) of post-hospital cognitive decline, with the
ultimate goal of detecting vulnerability toward persistent cognitive impairment and AD/ADRD. “Disturbance in
language” is among the diagnostic criteria for delirium, but speech is largely unexplored as a diagnostic or
prognostic tool due to the difficulty of achieving standardized and reliable assessments. Automated speech
analysis is an accurate, non-invasive, and efficient method for quantifying acoustic and textual speech
features. The computational speech features generated from such analyses are objective, non-invasive
markers that have been used to accurately diagnose and predict dementia but have not been examined for
diagnosing delirium or predicting post-hospital cognitive decline. In our pilot study, we found that computational
speech features could be feasibly collected in the hospital setting and allowed for more accurate classification
of delirium status among older adults than demographics and illness severity alone. Further validation is
needed. The objectives of our proposal are to use automated speech analysis and machine learning (ML) to
develop digital biomarkers for: (Aim 1) diagnosis of delirium in older adults with and without Mild Cognitive
Impairment (MCI) and AD/ADRD; and (Aim 2) explore the prediction of post-hospital cognitive decline. Study
Design: We will recruit 210 hospitalized older adults (75+), including at least 40% with pre-existing cognitive
impairment (MCI and AD/ADRD). Participants will be assessed for the presence and severity of pre-existing
cognitive impairment as well as delirium, and provide audio-recorded speech samples at 2 timepoints during
hospitalization while completing a series of language tasks. Speech samples will undergo verbatim
transcription and automated computerized processing to quantify speech features related to tempo, prosody,
organization, lexical characteristics, and dysfluencies. We will assess 3-month post-hospital cognitive
outcomes. Expected Outcomes: If successful, we will significantly improve the timeliness and accuracy of
delirium diagnosis during hospitalization, and identify patients with vulnerability for AD/ADRD and lasting
cognitive decline. Future longitudinal studies will evaluate the relationship between delirium and vulnerability
detected through automated speech analysis and subsequent development of AD/ADRD.
