A voice-based algorithm can predict type 2 diabetes status in USA adults: Findings from the Colive Voice study

Area of Science:

• Biomedical Engineering
• Computational Health
• Endocrinology

Background:

• Undiagnosed type 2 diabetes (T2D) poses a significant global health challenge.
• Current screening methods may lack accessibility and scalability.
• Innovative, non-invasive approaches are needed for early detection.

Purpose of the Study:

• To investigate the efficacy of a voice-based algorithm for predicting T2D status in adults.
• To develop a scalable and non-invasive screening tool for T2D.
• To compare voice algorithm performance with existing risk assessment scores.

Main Methods:

• Analysis of voice recordings from 607 US participants (Colive Voice study).
• Development of gender-specific algorithms using hybrid BYOL-S/CvT embeddings.
• Evaluation via cross-validation (accuracy, specificity, sensitivity, AUC) and Bland-Altman analysis against ADA score.

Main Results:

• Voice algorithms demonstrated good predictive capacity (AUC 75% males, 71% females).
• Accurate prediction of T2D cases (71% males, 66% females).
• Improved performance in older females (AUC 74%) and hypertensive individuals (AUC 75%), with >93% agreement with ADA score.

Conclusions:

• Voice-based algorithms offer a potential accessible, cost-effective, and non-invasive screening tool for T2D.
• Promising results warrant further validation, especially for early-stage T2D and diverse populations.
• This technology could significantly aid in reducing undiagnosed type 2 diabetes globally.