Developing a Multimodal Screening Algorithm for Mild Cognitive Impairment and Early Dementia in Home Health Care:...

Area of Science:

• Gerontology and Cognitive Neurology
• Health Informatics and Machine Learning
• Speech and Language Pathology

Background:

• Mild cognitive impairment and early dementia (MCI-ED) are often missed in home health care (HHC) due to time constraints and incomplete documentation.
• Current assessments like OASIS may not capture subtle cognitive changes.
• Speech patterns and clinical notes offer potential for earlier MCI-ED identification.

Purpose of the Study:

• To develop and evaluate a multimodal screening approach for MCI-ED in HHC.
• Integrate speech/interaction features, LLM-extracted data from notes, and OASIS variables.
• Enhance early detection of cognitive decline in the HHC setting.

Main Methods:

• Cross-sectional case-control study involving adults aged ≥60 receiving HHC.
• Collected audio-recorded patient-nurse encounters, EHR data (including OASIS and notes).
• Extracted speech/interaction features; used LLMs for text extraction; integrated data using machine learning.

Main Results:

• 114 HHC patients assessed (55 MCI-ED cases, 59 controls).
• Patient-nurse encounters analyzed for speech, interactional, and linguistic features.
• Exploratory analyses indicated multimodal models outperformed single-source models.

Conclusions:

• A reproducible multimodal framework for MCI-ED screening in HHC was developed.
• Data collection and processing are feasible.
• Integrating speech, text, and OASIS variables shows promise for improved MCI-ED detection.