People could be diagnosed with diabetes using little more than a short voice recording from their phone, according to a new study.
Using an audio sample of just six to 10 seconds, along with basic health data like age, sex, height and weight, scientists created an AI model that can determine whether someone is diabetic with almost 90% accuracy.
Klick Labs recruited 267 people for the study, including some already diagnosed with type two diabetes.
Each was asked to record a phrase on their phone six times a day for two weeks, and the team used AI to analyse more than 18,000 samples seeking acoustic differences between the diabetics and non-diabetics.
These included changes in pitch caused by type two diabetes that cannot be perceived by the human ear.
The model had an 89% accuracy rate for women and 86% for men.
Study author Jaycee Kaufman said the results could “transform” how people are screened for diabetes.
In the UK, more than 90% of adults with diabetes have type two – but many go for years without realising because symptoms can be general or non-existent.
To be tested for the condition, people usually require a GP visit and urine and blood tests.
Ms Kaufman said: “Current methods of detection can require a lot of time, travel, and cost.
“Voice technology has the potential to remove these barriers entirely.”
Previous research has found voice recordings – in combination with AI – can be used to diagnose other illnesses, including COVID-19.
Klick Labs believes the technology could also diagnose conditions like prediabetes and hypertension.
The peer-reviewed study has been published in the Mayo Clinic Proceedings journal.