Sonde Health’s Tech Can Help Pre-Screen Acute Respiratory Infections

Sonde Health is sharing new research that demonstrates the ability of its respiratory responsive vocal biomarker (RRVB) machine learning model to differentiate patients with COVID-19 from healthy individuals with about 70% accuracy.
 
The peer-reviewed study, published in the Journal of Medical Internet Research (JMIR), suggests the RRVB tool could serve as a pre-screening tool for acute respiratory infection and pave the way for the development of voice-based tools for future disease detection and monitoring applications.
 
The RRVB tool had already shown strong performance in differentiating patients with asthma, chronic obstructive pulmonary disease (COPD), interstitial lung disease, and cough from healthy individuals. In the new study, conducted in collaboration with Montefiore Health System, Brigham and Women’s Hospital, UC San Diego Health System, and Deenanath Mangeshkar Hospital in Pune, India, the model achieved 73% sensitivity and 63% specificity for the entire COVID-19 population (97 patients), and it detected 66% of asymptomatic COVID-19 subjects (46 patients) using only a six-second recording of an “ahh” vowel sound on patient smartphones. These findings suggest the tool could help uncover respiratory conditions before symptoms arise.

“We have shown that the same technology originally developed for asthma and COPD can be applied to pre-screen for COVID-19 with meaningful sensitivity and specificity,” Sonde Health Senior Vice President of Clinical Development & Customer Success Erik Larsen said. “This study demonstrates the robustness of our tool across conditions, geographies, and languages, paving the way for broader respiratory disease monitoring and surveillance efforts going forward.
 
The study enrolled 497 participants across four clinical sites in the United States and India, including patients who were COVID-19 positive, patients who had other acute illnesses, and asymptomatic volunteers who did not have an acute illness at the time of the study.
 
Dr. Sunit Jariwala, M.D., professor of Medicine and director of Clinical Research and Innovation in the Department of Medicine at Einstein College of Medicine and Montefiore Health System, served as principal investigator for the study. “This study highlights the potential of vocal biomarkers to improve access and outcomes for diverse and varied populations with respiratory diseases,” he stated. “By utilizing a digital tool that is non-invasive and can be easily scaled and distributed, we can effectively monitor respiratory health and identify individuals’ levels of symptoms and risk.  Based on the promising results from this study, we are working with Sonde Health to study the RRVB tool for respiratory monitoring in patients with moderate-to-severe asthma, and we are at the beginning stages of an Agency for Healthcare Research and Quality (AHRQ)-funded study to incorporate the RRVB tool into our own ASTHMAXcelmobile platform.”
 
The RRVB tool was originally developed and tested using a diverse dataset that included more than 3,000 patients with respiratory conditions, including asthma, COPD, and interstitial lung disease, as well as healthy individuals. The data was collected from more than 20 hospitals across India between August 2018 and January 2020 and encompassed multiple languages.
 
The COVID-19 validation study tested this tool for its ability to differentiate patients with COVID-19 from healthy individuals with data collected from September 2020 through April 2021.

Sonde Health develops voice-based health monitoring technology. Sonde serves top health companies, providers, pharmaceutical, and device OEMs through its vocal biomarker platform. Leveraging a best-in-class voice data set with more than 1.2 million samples from 85,000-plus individuals on four continents, Sonde uses advanced audio signal processing, speech science, and AI/machine learning to sense and analyze subtle vocal changes due to changes in a person’s physiology to provide key insights into health and well-being.