Acoustic Analysis of Primary Care Patient-Clinician Conversations to Screen for Cognitive Impairment

Cognitive impairment (CI) is often underdetected in primary care due to time and resource constraints. Passive analysis of clinical dialogue may offer an accessible approach for screening.

To assess whether audio recordings of patient-physician dialogue during routine primary care visits can be used to identify CI using acoustic speech features and machine learning (ML).

This diagnostic study, conducted from August 2020 through December 2021 among older primary care patients, involved audio recording primary care visits using a microphone and portable device. An external validation cohort was recruited in a separate city to assess reproducibility of findings. Analysis was performed from January 2025 through June 2025. The study was conducted in primary care practices in New York, New York, with additional participants recruited from primary care practices in Chicago, Illinois, for validation. Eligible patients were recruited from primary care practices during routine visits with no prior diagnosis of mild CI or dementia. The study included English-speaking patients aged 55 years and older without documented history of dementia or mild CI. For validation, patients meeting the same eligibility criteria were recruited from primary care practices in Chicago.

Multiple 30-second speech segments were extracted from recordings. Acoustic features were derived using foundation models (Whisper, HuBERT, wav2vec 2.0) and expert-defined methods (eGeMAPS, prosody).

The primary outcome, CI, was defined as Montreal Cognitive Assessment score 1.0 or more standard deviations below age- and education-adjusted norms. ML classifiers were trained to predict CI status from audio recordings. Area under the receiver operating characteristic curve (AUROC) and maximum F1 score (Fmax) were calculated for identifying participants with CI.

The study included 787 English-speaking patients aged 55 years and older without documented history of dementia or mild CI and an additional 179 patients meeting the same eligibility criteria for validation. In total, 966 participants were recruited, among whom 530 (55%) were female, mean (SD) age was 67.2 (8.1) years, and CI prevalence was 21%. Models using Whisper-derived acoustic features performed best (AUROC, 0.733; 95% CI, 0.714-0.752; Fmax[CI], 0.502; 95% CI, 0.471-0.533). Results generalized to the external site with similar performance (AUROC, 0.727; 95% CI, 0.714-0.740; Fmax[CI], 0.459; 95% CI, 0.441-0.477). Model interpretation identified pitch, timing, and variability features as key predictors. When used for screening, the algorithm achieved positive predictive value of 30.4% (95% CI, 28.7%-32.1%), sensitivity of 68.2% (95% CI, 61.8%-74.6%), and specificity of 63.6% (95% CI, 59.8%-67.4%) on the holdout cohort.

In this diagnostic study, ML models trained on acoustic features from brief clinical conversations identified CI with high accuracy. These findings support the feasibility of passive, speech-based screening during routine primary care.