AI-Based Speech Analysis May Flag Cognitive Impairment

Machine-learning models trained on routine patient-physician conversations achieved similar performance in development and external validation cohorts in predicting cognitive impairment.

By
Andrea Surnit
June 19, 2026
4 min

Conexiant

Overview

A machine-learning analysis of conversations between primary care physicians and older patients shows potential for identifying cognitive impairment with moderate accuracy. The highest-performing model achieved an area under the receiver operating characteristic curve of over 0.73.

Background

Cognitive impairment is a growing concern in aging populations. This study explores the use of machine learning in analyzing speech patterns as a potential screening method for cognitive decline.

Data Highlights

Metric	Development Cohort	Validation Cohort
AUROC	0.73	0.73
Positive Predictive Value	N/A	30%
Sensitivity	N/A	68%
Specificity	N/A	64%
Cognitive Impairment Prevalence	N/A	21%

Key Findings

The highest-performing machine-learning model achieved an AUROC of over 0.73 in both development and validation cohorts.
In the validation cohort, the best-performing algorithm had a positive predictive value of 30%, sensitivity of 68%, and specificity of 64%.
Greater variability in pause duration and increased energy in unvoiced speech were associated with cognitive impairment classification.
Deep neural network-derived acoustic features outperformed expert-defined acoustic measures.
Preserving conversational dynamics improved model performance.