Klick Health screened diabetes from short voice recordings

India alone has more than 101 million people living with diabetes, and more than half of cases in the country go undiagnosed, driven by limited healthcare access in rural areas, cost barriers to routine blood testing, and gender gaps in healthcare access. The standard screening path for Type 2 diabetes requires a clinic visit and an invasive blood draw, a process that's costly, logistically difficult, and simply never completed by a large share of the people who need it, particularly in regions with limited healthcare infrastructure.

Klick Applied Sciences, the research arm of health-marketing company Klick Health, had already been investigating vocal biomarkers, measurable characteristics of the human voice linked to underlying physiological states, as a research direction. The team recognized that with more than a billion Indians owning mobile phones, a smartphone itself could potentially serve as a screening tool if a genuine diabetes-linked signal could be found in something as ordinary and already-occurring as a person's speech.

Piggybacked screening on speech: analyzed smartphone voice recordings for 14 acoustic markers of diabetes, so a 6-second talk replaces the blood draw and clinic visit.

Type 2 diabetes alters vocal characteristics imperceptibly, so voice carries a diagnostic signal.

People already speak into phones daily, so screening requires no extra effort or cost.

Machine learning on 18,000 recordings identified 14 vocal markers with 86-89% accuracy.

Removing the clinic visit and blood draw eliminates the main barriers to screening uptake.

When a diagnostic step is a barrier, find a passive signal in behavior people already do, and make screening a free byproduct.

The underlying research, led with Mayo Clinic investigators, was published in Mayo Clinic Proceedings: Digital Health in October 2023, reporting 89 percent accuracy for women and 86 percent for men in distinguishing diabetic from non-diabetic voices.