Speech-Driven Digital Biomarker for Cognitive Decline Detection

Overview

This study demonstrates that automated analysis of spontaneous speech significantly improves prediction of cognitive performance in older adults compared to demographic data alone. The developed speech-based digital biomarker achieved a ROC-AUC up to 0.81 in identifying individuals below normative cognitive thresholds and generalized well to an independent Alzheimer's disease cohort.

Background

Cognitive decline is increasingly prevalent in aging populations, necessitating accessible and scalable screening tools for early detection and intervention. Traditional cognitive assessments can be resource-intensive and intrusive. Speech, as a natural and non-invasive behavior, offers a promising avenue for digital biomarker development. Leveraging machine learning to analyze linguistic and acoustic features from spontaneous speech may provide a low-cost, scalable method for cognitive evaluation.

Data Highlights

Key Findings

• Speech-derived linguistic and acoustic features substantially outperform demographic data alone in predicting cognitive domain scores.
• Machine learning regression models using speech features achieved a fourfold performance improvement.
• A binary classifier based on speech features identified individuals below normative cognitive thresholds with ROC-AUC up to 0.81.
• The approach generalized effectively to an independent clinical dataset of Alzheimer's disease patients and controls.
• Speech analysis offers a low-cost, non-intrusive digital biomarker suitable for large-scale cognitive screening and clinical trial participant selection.

Clinical Implications

Automated speech analysis can serve as an accessible and scalable screening tool for early cognitive decline, facilitating timely intervention. Its non-invasive nature and strong predictive performance support integration into routine cognitive monitoring and research settings. This method may enhance participant selection for clinical trials and reduce reliance on more burdensome cognitive assessments.

Conclusion

Automated speech analysis represents a clinically feasible digital biomarker for cognitive decline, offering improved prediction accuracy and generalizability. This approach holds promise for widespread cognitive screening and monitoring in aging populations.

References

1. Brito et al. 2023 -- Assessing cognitive decline in the aging brain: lessons from rodent and human studies
2. Nandi et al. 2024 -- Cost of care for Alzheimer’s disease and related dementias in the United States: 2016 to 2060
3. McDade et al. 2020 -- The pathway to secondary prevention of Alzheimer’s disease
4. Cummings et al. 2024 -- Alzheimer’s disease drug development pipeline: 2024
5. Petersen et al. 2018 -- Practice guideline update summary: mild cognitive impairment
6. Öhman et al. 2021 -- Current advances in digital cognitive assessment for preclinical Alzheimer’s disease