New research suggests that people with diabetes can be diagnosed using a voice recording from their phone.
Experts have long feared that millions of Britons with the condition, which causes blood sugar levels to become too high, are going undiagnosed.
But audio clips of patients’ voices lasting just 25 seconds, combined with basic health data such as age, gender, height and BMI, could determine whether someone has diabetes.
Scientists who tested the AI model on more than 600 adults found it was accurate more than two-thirds of the time.
Charities and experts today welcomed the ‘innovative’ tool, but warned more research is needed.
Experts have long feared that millions of Britons with the condition — which causes blood sugar levels to rise too high — are going undiagnosed. But audio clips as short as 25 seconds, along with basic health information such as age, gender, height and BMI, can determine whether someone is diabetic
Nearly 4.3 million people were living with diabetes in 2021/22, according to the latest figures for the UK. And a further 850,000 people have diabetes and are completely unaware of it, which is worrying because untreated type 2 diabetes can lead to complications including heart disease and stroke.
Abir Elbeji, a digital health expert at the Luxembourg Institute of Health and lead author of the study, said: ‘Most current methods for screening for type 2 diabetes are time-consuming, invasive, performed in a laboratory, and expensive.
‘The combination of AI and speech technology can make testing more accessible by removing these barriers.
‘This study is the first step towards using voice analysis as a first-line, highly scalable screening strategy for type 2 diabetes.’
In the UK, more than 90 per cent of adults have type 2 diabetes. But many go unnoticed for years, as symptoms can be generalised or absent.
To be tested for the condition, people usually need to visit their doctor and have their urine and blood tested.
In the study, researchers analyzed 607 recordings to identify acoustic features that distinguished non-diabetics from diabetics.
Everyone was asked to read sentences straight from their phone or laptop.
Using signal processing, they were then able to detect subtle changes in pitch, tone and intensity that are invisible to the human ear.
One advanced AI technique captured up to 6,000 detailed vocal features. Another focused on 1,024 key features.
The model had an accuracy of 66 percent in women and 71 percent in men.
The model performed even better in women aged 60 or older and in people with hypertension, the researchers said.
Dr Lucy Chambers, head of research at Diabetes UK, commented on the findings: ‘Rapid and accurate diagnosis is vital to prevent the serious and life-limiting complications of type 2 diabetes.
Unfortunately, symptoms can easily be missed and there are currently around 1.2 million people in the UK living with undiagnosed type 2 diabetes.
‘By using artificial intelligence to develop convenient and cost-effective screening methods for type 2 diabetes, we can identify more people who need treatment and support, ultimately improving their quality of life and reducing the risk of long-term diabetes complications.
“We look forward to further research into innovative AI speech analysis tools for diabetes screening.”
The research will be presented tomorrow at the annual meeting of the European Association for the Study of Diabetes (EASD) in Madrid.
Type 2 diabetes occurs when the body does not produce enough insulin or when the insulin the body produces does not work properly.
This hormone is needed to lower blood sugar levels.
High blood sugar levels can cause heart attacks and strokes in the long term, as well as problems with the eyes, kidneys and feet.
Patients may need to adjust their diet, take daily medications, and have regular check-ups.
Symptoms of the condition, which is diagnosed with a blood test, include excessive thirst, fatigue and increased urination. But many people have no symptoms.
According to the latest figures for the UK, almost 4.3 million people were living with diabetes in 2021/22.
About 90 percent of diabetes cases are type 2 diabetes, which is associated with obesity and is usually diagnosed in middle age. This excludes type 1 diabetes, a genetic condition that is usually diagnosed at a young age.