Detecting Mild Cognitive Impairment Using Follow-Up Call Speech and Electronic Health Record Data in Home Health Care...

Area of Science:

• Gerontology
• Computational Linguistics
• Health Informatics

Background:

• Alzheimer's disease and related dementias are frequently undiagnosed in home health care settings.
• Early symptoms of cognitive impairment are often not documented in electronic health records (EHRs).

Purpose of the Study:

• To develop and evaluate a speech-based screening algorithm for detecting mild cognitive impairment (MCI).
• To utilize routine follow-up calls in home health care for cognitive screening.

Main Methods:

• Speech data were collected from semi-structured follow-up calls, Clinical Dementia Rating (CDR) interviews, and EHR data (Outcome and Assessment Information Set [OASIS]).
• Machine learning models were trained on individual and combined data modalities.
• Model performance was assessed using the area under the curve (AUC) of the receiver operating characteristic.

Main Results:

• The multimodal model combining follow-up call speech and OASIS variables achieved the highest performance (AUC = 91.03%).
• This multimodal model outperformed models using CDR interviews (AUC = 80.67%), follow-up calls alone (AUC = 81.09%), and OASIS alone (AUC = 78.53%).
• Patients with MCI exhibited greater comorbidity burden, higher dual eligibility rates, and lower health literacy.

Conclusions:

• Combining speech from routine follow-up calls with structured clinical data improves early MCI detection in home health care.
• This low-burden, multimodal approach integrates with existing care coordination workflows.
• The method shows potential for scalable cognitive screening in diverse, aging populations.