One look at ClinicalTrials.Gov and you will be inundated with the sheer volume of different clinical trials currently underway to tame the rapacious coronavirus that has claimed lives globally.
Of those, a study neither looking at treatments nor the development of vaccines is a central IRB-approved virtual study where participants will record the sound of their cough and send the data to a startup. San Francisco-based HealthMode, which is sponsoring the study, aims to use this data to help researchers understand whether a cough’s sound can indicate – among other things – whether the person has Covid-19.
“We are just trying to collect natural cough samples from folks who have and don’t have the disease and then track them over time to see how their cough changes over time along with their other symptoms,” said Daniel Karlin, co-founder and CEO of HealthMode, which was incorporated in December 2017.
HealthMode is backed by seed funding from DCVC and Promus Ventures, Karlin said, though the startup does not appear to be listed as a portfolio company in either venture capital firm’s website. The goal at inception was to create digital measurement tools for clinical trials that could inform the development of drugs, medical devices and combination products. The original research was in counting coughs to help drugmakers develop better cough suppressants or other drugs where coughing is a symptom such as COPD and idiopathic pulmonary fibrosis, Karlin said. But when Covid-19 hit, that research was repurposed to measure the sound of coughs to help researchers understand the disease in the hope of ultimately being able to make predictions.
“So at this point, we are not trying to make any predictions. We are just getting the cough sounds along with the other information so we can try to train an AI or ML model to determine what they can predict.” Karlin said, in a phone interview. “This is the science to find out what we can predict.”
However, there is a certain hypothesis for what can be predicted and once enough data is collected and the models emerge, Karlin hopes to work with the FDA to get regulatory clearances for making such claims.
“The claims would either be the likelihood that this cough is a product of Covid or not the product of Covid, so a positive claim or a negative claim,” he said. “The other set of claims we would like to make… is a probability claim about severity. I do think our highest value will be outside healthcare settings — the data might be able to help in decisions on returning to work or monitoring a workforce to make sure they are safe.”
Other efforts to understand the particulars of Covid-19 cough are also underway — for example, the city of Jackson, Mississippi, has created a Covid-19 symptom collector form to allow residents to record their cough sounds.
HealthMode’s study aims to have 1,000 enrolled and collect 10,000 independent episodes of coughing. As of earlier this week, 200 participants have signed up. They answer both demographic and health-related survey questions and send their cough sounds using an iOS app, through the web, or from a landline for a study period of eight weeks. Aside from the initial onboarding information gathered, surveys are also sent weekly to monitor the patient’s cough and any changes they notice.
Karlin said that the background information collected when the study participant comes online involves asking them about whether they have been exposed to Covid-19, whether they have been tested for it and whether they are currently experiencing symptoms like loss of taste or smell, shortness of breath and GI symptoms and other questions. Although people with hypertension and diabetes are among those that experts believe at most risk for Covid-19, HealthMode’s survey questionnaire deliberately avoids asking too many questions about those illnesses.
“We actually don’t pry on chronic conditions and so we ask about respiratory-related chronic conditions,” Karlin said. “I don’t think there is any reason to think that diabetes will change cough sounds.”
However, the app also doesn’t ask about smoking and vaping, which Karlin acknowledged would be something that would be important in the overall understanding of Covid-19.
“It’s something that’s on the list for the next iteration because I think you are right — smoking and vaping are probably likely to change symptoms,” he said. “Depending on what we are seeing from the data that we get back, we can always add additional questions and ask the IRB for permission to include the questions.”
So far the 200 people who have signed up are all based in the U.S. according to the limits of the study design, but people in other countries can signal their interest. Depending on where a critical mass of interest is seen, HealthMode has the ability to seek regulatory approvals to conduct studies there, Karlin said.
U.S. participants can either use their iOS device to record and send cough data, use a browser-based application to do the same on a computer, or a dial-in number that people can call from a landline, Karlin said.
“We don’t expect things like microphone quality to substantially impact our ability to detect the quality of audio collections, but that is a hypothesis that we are testing over time,” he said.
The company is using HIPAA-compliant back end systems to ensure that the personal health information collected — including the cough sound data — is protected and can only be accessed by HealthMode’s internal research team.
“We also process the sound data so that it’s cut into just over a second-long chunks and the chunks are not stored in order, so that even if someone were to accidentally say something sensitive while they were recording, we wouldn’t be able to piece that conversation together because they are split up into those chunks,” he said.
Karlin said that he would love to be able to set up a data consortium so that academics and any other efforts to review cough data would be able to pool their efforts together.
