Overview
- Researchers from thymia and RMIT presented results at the European Association for the Study of Diabetes in Milan showing a speech model trained on 63,283 recordings from 21,129 people.
- In a UK validation of 7,319 adults the model assigned higher risk scores to people who reported having type 2 diabetes 80% of the time.
- In a subgroup of 801 participants who did home HbA1c tests the model matched blood-test‑based diabetes status 75% of the time with sensitivity of 82% and a false positive rate of 47%.
- Performance fell for recordings from Black participants and for people with heart disease, high blood pressure or obesity, highlighting representativeness and confounding issues that need fixing before rollout.
- Authors disclosed multiple financial ties to thymia and say the tool is intended for remote triage to flag people for confirmatory blood tests rather than to replace laboratory diagnosis.