April 24th, 2019
Welcome to The Future Labs, a Podcast series talking to people building the future, today. We speak to experts around the globe about their role in changing the world for the better.
In this episode, we are exploring The Future of Voice Biomarkers (and maybe coining the term “voice-omics”!) and the potential to diagnose medical conditions based on voice. We are speaking to Ed Kliphuis, Investment Director at M Ventures, the corporate venture capital arm of Merck KGaA. Ed spoke with us about their recent investment into Sonde Health, a digital medicine company developing diagnostic tools based on voice biomarkers.
Biomarkers are measurable indicators that are associated with specific biological states and diseases. The world of medicine has traditionally focused on either on physical biomarkers such as body temperature and blood pressure or on biochemical biomarkers such as cholesterol levels in the blood, protein levels in urine or presence of a genetic mutation. These markers are typically measured by a healthcare professional as part of the diagnosis and management of patients. However, rapid improvements in consumer technology have led to the development of a plethora of diagnostic and monitoring devices and Apps for use away from the doctor’s office.
Sonde Health is a Boston-based digital medicine company that is developing technology to monitor and diagnose physical and mental health conditions based on analyzing “voice biomarkers” present in people’s voices. They recently raised $16m in a Series A financing led by M Ventures, which they will use to conduct clinical studies with their diagnostic software.
The core concept behind their technology is that there are subtle features present in the voices of people with certain medical conditions that are distinguishable from the voices of people without those conditions. Sonde has developed software that can measure a range of such features, including vocal elements such as prosody, melody, pitch slope, intensity and frequency, and identify subtle alterations in people with Major Depressive Disorder (MDD / Depression).
Sonde has collected voice samples from people with confirmed MDD and those confirmed to not have MDD to determine the global baseline voice characteristics. Analysis of voice samples from over 4,000 people across the US, India and Australia confirmed that they could predict which people had MDD based on a 6-second voice sample collected via an iPhone App. Importantly, the differences in voice biomarkers are independent of the language spoken and no patient-specific baseline is required for correct diagnosis. Sonde’s analysis was performed alongside a traditional mental health questionnaire, allowing them to estimate its accuracy. They found that they correctly identified MDD in over 80% of patients with this condition and the rate of false positives was below 10% (i.e., identifying MDD in people that medical professionals did not identify as having MDD).
"That [thermometer] has been one of the most valuable preventative medicine tools that we've had for the past 200-plus years. And it doesn't diagnose anything. It just cues us when our body's normal function if out of range - and at a level that tells us how significant that change is" - Jim Harper, interview with STAT News
Discussions with FDA suggest that this level of sensitivity and specificity is sufficient for controlled use in a known population suspected of having MDD, but would probably not be sufficient for use as a screening tool in the broader population. As the technology evolves, the hope is that the false positive rate would come down, reducing the risk of incorrectly identifying someone as having MDD when they do not. Additionally, the data collected to date do not provide a mechanistic understanding of why such biomarkers may be associated with MDD and it is not yet clear whether the machine learning algorithms they use for biomarker identification are picking up correlative or causative relationships.
In an interview with STAT News Jim Harper, Sonde’s founder and CEO compared their technology to the thermometer: “That has been one of the most valuable preventative medicine tools that we’ve had for the past 200-plus years. And it doesn’t diagnose anything. It just cues us when our body’s normal function is out of range – and at a level that tells us how significant that change is.”
Data protection, especially that of sensitive medical information, is increasingly a point of concern in the Information Age, and Sonde is acutely aware of the need to address these concerns. Their technology is designed to listen passively (i.e., running in the background and continually sampling audio input), but to only listen out for key vocal biomarkers which then trigger a full sampling of audio (for a short duration). This is similar to Amazon’s Alexa, which listens constantly but only for the word “Alexa” which then triggers the device to actively listen to the following command. In addition, the technology relies on local processing on the device, meaning that no raw voice samples are sent to the Cloud for processing, further ensuring data protection.
Sonde is currently focusing on taking its MDD program through clinical studies and potentially seeking FDA approval in late 2020. However, they believe there is further potential in a range of other disease areas beyond psychiatry, including respiratory, cardiology and neurology.