AI Cough-Monitoring Can Change the Way We Diagnose Disease

How many times do you cough a day? Do you cough more when you’re indoors or outside? Or more often after you eat? Or at night?

Chances are, your cough memory might not be that accurate. But all of that information about your coughing patterns could be an untapped resource to better understand your health. Coughs may be benign ways to clear a little extra phlegm, or they could be early signs of more serious conditions such as asthma, GERD (gastroesophageal reflux disease), or even lung cancer.

“In the era of precision health, it’s ironic that such a problematic symptom is simply unmeasured,” says Peter Small, chief medical officer at Hyfe AI, a company that boasts a database of more than 400 million cough samples, the largest in the world, that researchers are analyzing for potential medical use. [time-brightcove not-tgx=”true”]

There is power in cough information, and the ubiquity of cell phones and wearables with recording devices mean that apps can increasingly capture and analyze that data. Companies like Google Health see even basic information such as getting an accurate count of the number of times a person coughs a day as a useful resource, and part of a larger need to collect and chronicle more health information to refine the way doctors diagnose disease and manage treatments in the future. “It’s a sea change to have a common device, the smartphone, which everyone has sitting by their bedside or in their pocket, to help observe your coughs,” says Jamie Rogers, product manager at Google Health.

More from TIME

Rogers and his team are taking advantage of improved recording technology and powerful algorithms to distinguish coughs from background noise, and to isolate one person’s cough from another’s. That ensures that the data collected—all via the microphone on smartphones or wearables—is specific to an individual and actually means something for that person. In the not too distant future, Google is aiming to capitalize on cough monitoring in three ways: helping people determine how environmental factors such as allergens and pollution can affect their health; giving them a heads up when there are spikes in the number of coughers in their area, which could be a sign that the risk of picking up an infectious disease like flu or COVID-19 is higher; and alerting people with chronic conditions like asthma or chronic obstructive pulmonary disease (COPD) when their cough patterns change ahead of an episode of distress.

“The science is moving very fast,” says Joe Brew, CEO of Hyfe AI. “A couple of years ago, nobody could tell you anything about coughs. Now there are several dozens of publications, and companies working to use cough information in diagnosis.”

What coughs can reveal about your health

Even collecting basic information on the number of coughs people have in a day, and when and where they cough most, could reveal a lot. Hyfe AI is conducting studies to confirm what doctors have mainly inferred about those connections to this point—that people with asthma are more likely to cough outdoors during allergy season, or indoors in poorly ventilated spaces, for example, and that people who cough more at night or after meals are more likely to have GERD. “There is a tremendous value in simply empowering patients and doctors with this information,” says Small. “It’s going to transform the whole clinical approach for this common and chronic symptom. Patients will come in, have the data on how much they are coughing, and the physician can suggest a treatment based on that information to see if it makes the coughs better.”

With artificial intelligence known as acoustic AI that’s now able to parse apart information from hundreds of thousands of cough samples, it’s becoming possible to think of coughs in the same way we think about blood pressure, blood sugar, and body temperature. Collect enough samples, gather enough data, and patterns begin to emerge.

Hyfe AI has several products built from its AI-based algorithms that can distinguish cough sounds from other noises with up to 99% accuracy. One is a free demonstration app, CoughTracker, that users can download to track the number of their coughs. The coughs recorded on this app are not stored, unless participants join a research study and consent to having their data collected. Using CoughTracker, for example, researchers led by scientists at University of Navarra in Spain are testing whether tracking coughs through digital apps can predict incidence of respiratory diseases such as COVID-19, influenza, asthma, and COPD.

The company is also studying how such data can help predict the prognosis of respiratory illness. In a recent study conducted with COVID-19 patients hospitalized at the University of Florida Health Shands Hospital and at the Montreal University Hospital Center, for example, scientists used Hyfe AI’s research app and learned that the more a patient coughs, particularly at the beginning of their infections, the better their outcomes; the less a patient coughs, the more advanced their respiratory illness might be, alerting doctors that they might need more intensive interventions.

Hyfe AI is also developing a wrist wearable loaded with the cough-detecting algorithm, called Cough Monitor, that it hopes to submit for Food and Drug Administration approval in early 2024. Early data from the company show that the device can accurately detect cough sounds from background noise, and track the frequency of coughs from an individual wearer.

For now, Hyfe only collects cough data from participants in such research studies who consent to participate and provide their cough data to the scientists. The free app records cough audio but only stores it on a user’s phone, and does not upload any of the data to the cloud or share it with the company, so data from the app is not being used to improve Hyfe’s algorithm. Google’s cough monitoring, available through the Digital Wellbeing app for Android phones, takes the same approach. The audio data only exists on the phone itself, and is not uploaded to the cloud.

But going forward, Hyfe, Google and others with cough-monitoring apps will want to capture more data from users in order to analyze it more deeply; those versions will have to address privacy issues, and request consent from users if the companies want to collect their cough data for purposes of improving or developing new algorithms. In order to improve and develop additional cough monitoring technologies, Rogers says Google Health may consider partnerships with researchers to license its technology to build new models based on data collected with users’ consent.

Using coughs to diagnose disease

Beyond the initial utility that chronicling the frequency of coughs can provide, the ultimate goal is to mine cough data to see if distinct signatures materialize that would distinguish, say, a flu’s cough from a COVID-19 one, or lung cancer from allergies. Hyfe is currently developing another device, Cough Watch, a wrist-based wearable that is only available in research trials, which uses deeper AI algorithms designed to test whether it’s possible to use cough data to help diagnose different conditions.

Google Health also sees potential in using cough data for diagnosing disease, beyond recognizing the differences between wet and dry coughs, or between hacking coughs and wheezes. “If you asked me a year ago, I would have said I was skeptical about [using coughs for screening or diagnosis,]” says Dr. Lawrence Cai, a medical specialist in mobile sensing at Google Health. “When I was in medical school, never ever did they teach us that we could listen to somebody cough and identify whether that person has TB (tuberculosis), COPD, or a tumor. But I keep seeing more and more studies of people coughing into a microphone, and an algorithm can detect whether somebody has TB with 95% specificity and sensitivity, or if someone has pneumonia or an exacerbation of COPD.”

More work needs to be done before such diagnostic uses are clinically viable, but that possibility could be especially powerful in remote parts of the world with limited health resources, where people have less access to technology such as X-rays to diagnose TB. Using a smartphone app to screen the coughs of people who live far from health care resources to determine if they are likely caused by TB, for example, could identify those people who could benefit from the long and costly visit to a health care facility to receive treatment. In places like India, for instance, where the government hopes to eradicate TB by 2025, a cough app could identify areas where infections are flaring up so health authorities can direct resources such as health care personnel and treatments to those regions.

If digital cough monitoring became more available, the information could even help people with COPD, asthma, or allergies to prevent severe episodes that require costly and intensive medical treatment. In several years, says Rogers, understanding changes in coughs could help everyone become more proactive about our health as we rely on cough apps to alert us when pollen and allergy counts are high, or when we’re coughing more and might be infected with a bug. “Our phones are perfectly capable of offering that insight to us,” he says.