Biomarkers.

Area of Science:

• Gerontology
• Computational Linguistics
• Artificial Intelligence

Background:

• Geriatric physical frailty is a reversible condition associated with increased risk of neurocognitive disorders like Alzheimer's disease (AD).
• Early detection of frailty is crucial for preventive interventions, yet non-invasive screening methods are limited.
• Speech markers have potential for neurodegenerative pathology identification but their role in frailty detection is unexplored.

Purpose of the Study:

• To evaluate machine learning models using speech-based features for scalable, non-invasive screening of physical frailty in older adults.
• To determine the effectiveness of acoustic, temporal, and linguistic speech features in classifying frailty status.

Main Methods:

• 271 participants (≥65 years) underwent physical and cognitive assessments; frailty was determined using the Fried Frailty Index.
• Spontaneous one-minute speeches were recorded and analyzed for acoustic, temporal, and linguistic features.
• Random Forest, XGBoost, and Support Vector Machine (SVM) models were trained to classify frailty status.

Main Results:

• The SVM model achieved the highest classification accuracy with an Area Under the Curve (AUC) of 0.93 and Average Precision (AP) of 0.95.
• Speech analysis, incorporating acoustic, temporal, and linguistic markers, proved effective for frailty detection in the elderly.
• SVM demonstrated robust performance, highlighting speech as a valuable marker for identifying frailty.

Conclusions:

• Speech-based machine learning models offer a novel, accessible, and scalable method for early frailty screening in aging populations.
• This approach can facilitate timely interventions to improve robustness and reduce the risk of neurocognitive disorders.
• Implementing these classifiers in clinical settings can enhance preventive care strategies for older adults.