Readers of this publication probably have heard of the impressive advances in analyzing voice patterns for disease. These automated models can monitor so many people efficiently that they have potential applications in public health as well as diagnosis and treatment.
A simple app on a cell phone can detect unusual patterns in speech that suggest when a person is depressed or anxious, is losing cognitive function, or has pulmonary disease.
But there are several reasons that “prediagnostic” data from popular devices can’t be trusted: lack of regulation, user error, lack of information about the device’s operation, etc.
Nobody is relying on vocal markers for diagnosis yet, according to David Liu, CEO of Sonde Health. But the analyses are useful input to doctors’ diagnoses. I talked to Liu recently about what Sonde Health is doing in India and other parts of the world with vocal markers.
India’s second-greatest health problem
India surpasses the world in cases of chronic obstructive pulmonary disease (COPD). It is the second biggest cause of death in India, according to Liu. Sonde Health is trying to improve the tracking of COPD (along with other illnesses that can be detected through voice analysis) through a partnership with Koye, a pharma company in India. Koye in turn is promoting a Sonde app to clinicians.
The models used by the Sonde app to detect disease are based on data collected over several years in clinical settings, where each vocal sample is tagged with the diagnoses assigned by clinicians to the patient. Sonde’s reliance on high-quality clinical data contrasts with some experimental AI applications that take in lots of low-quality data and hope to find patterns that reveal the dimensions related to disease.
Another advantage of Sonde’s methodical approach to data collection is a truly diverse data set across socio-economic, geographic, ethnic, and linguistic lines.
Although the Sonde app can assess some health parameters of a patient after a single analysis, the real power of the app comes from repeated health checks over a long time period. The app can get to know the patient and can establish a baseline against which changes in the patient’s voice can be measured. Conditions tracked by Sonde include COPD, depression, anxiety, substance abuse disorder, asthma, cognitive impairment with other institutions, and some motor impairments.
The hope at Sonde and Koye is that clinicians can detect COPD more quickly and identify the patients that need attention. Naturally, because Koye works through the doctors, it does not reach many patients who lack medical care in India. But Sonde is also considering ways to work with public health authorities to identify at-risk areas. The app, if widely deployed, could provide important advance notice like checking waste water for disease.
Expanding the reach of health monitoring
Although fitness devices are popular, many people still can’t afford or don’t want to bother with them. Vocal tracking is an appealing technology because it can be loaded onto a common cell phone, reaching a much larger population.
Sonde is going further, talking to Qualcomm and other chip manufacturers to embed tracking directly in the chip so that cell phone users have access to monitoring without installing an app. To ensure privacy, the chip never uploads patient data to a third party. Instead, the phone downloads the models that were pre-computed on clinical data. The individual’s data and the analytics that the models generate remain on the individual’s device. A patient can choose to share results with doctors and researchers.
As another delivery mechanism, Sonde has a commercial partnership with a multinational company, GN Group, that manufacturers and markets hearing aids and earphones.
Sonde is a useful illustration of how a company can expand the reach of medicine, with potential benefits to public health, while finding a business model that works within the current health system. That’s the ethical and practical balance that every small health-related company has to find.
