AI-Powered Speech Analysis Shows Promise for Rapid Type 2 Diabetes Screening

A novel approach utilizing artificial intelligence to analyze speech patterns could enable the screening of type 2 diabetes in mere seconds, according to a recent study. Researchers indicate that this technology offers a "new route for diabetes testing," with voice recordings being feasible through phone calls or mobile applications.

Millions in the UK are living with diabetes, and a significant portion remain undiagnosed due to the often slow onset of symptoms and limited access to routine check-ups and standard blood tests. This new tool aims to address these diagnostic challenges.

The technology, developed by researchers from the tech company thymia and RMIT University in Melbourne, Australia, leverages AI to identify subtle speech alterations linked to type 2 diabetes. These changes can include vocal strain, increased hoarseness, and impaired breath control. High blood sugar levels, common in poorly controlled diabetes, can damage the vagus nerve, which affects the vocal cords, potentially leading to a rough or scratchy voice. Additionally, higher rates of stomach acid reflux in individuals with diabetes can irritate vocal cords, while reduced lung function may affect airflow necessary for clear speech.

To train the AI model, researchers analyzed over 63,000 voice samples from more than 21,000 individuals in the UK and US. The model was subsequently tested using 20-second voice recordings of people reading aloud. In a study involving 7,319 participants in the UK, the speech model correctly identified individuals who reported having type 2 diabetes with 80% accuracy. The study noted that while the tool performed well across various ages and genders, its accuracy was lower for black patients, a discrepancy attributed to the limited number of black participants in the dataset.

A secondary analysis of 801 individuals who had undergone diabetes blood tests within three months of their voice recording showed the AI tool assigning higher risk scores to 75% of these participants.

Traditionally, type 2 diabetes is diagnosed via blood tests measuring average blood sugar levels over two to three months. These tests are typically offered to individuals experiencing symptoms or those aged between 40 and 74 during routine health checks.

Giedre Cepukaityte, a research scientist at thymia, highlighted the tool's potential impact on screening. "This is the largest real-world study of speech-based screening for type 2 diabetes to date which also checks the model's predictions against blood test results as well as against what people reported about their own diagnosis," she stated. Cepukaityte added, "A speech sample can be taken over the phone or through an app, so we can reach far more of the people who need a blood test than current pathways do, particularly those who never get to a health check."

Cepukaityte emphasized that the AI tool is intended as a screening method and not a replacement for blood tests. "Our next step is to test the model in clinical settings and to understand how well it works for every group of people, because a screening tool has to work for everyone," she concluded.