Q&A: AI-developed algorithms could change chronic cough monitoring

Key takeaways:

• There is a disconnect between cough scientists and the FDA in what represents a meaningful outcome for chronic cough drugs.
• AI-developed algorithms could allow for continuous cough monitoring.

In November, all but one of the members of the FDA’s Pulmonary-Allergy Drugs Advisory Committee voted against recommending the approval of gefapixant for adults with refractory or unexplained chronic cough.

Key points of the panel’s discussion prior to voting included the small reduction in cough frequency (number of cough events in 24 hours/total recording duration) seen with gefapixant (Merck), concerns over cough frequency as an endpoint and the lack of correlation between patient-reported outcomes and cough frequency.

Despite a majority voting no, all the panelists acknowledged that chronic cough is a severe illness that greatly impacts a patient’s life. At this time, there are no FDA-approved therapies for these patients, suggesting a major need for new treatments.

To gain more insight on what was discussed during the committee meeting and how cough counting/monitoring systems are evolving, Healio spoke with Peter Small, MD, acoustic epidemiologist and chief medical officer at Hyfe.

Healio: The two clinical trials that evaluated gefapixant for chronic cough used the VitaloJAK system (Vitalograph) to count coughs. How is this system different from other cough technologies?

Small: There is no FDA-cleared cough counter. The VitaloJAK has been the workhorse for recording coughs. It’s essentially a belt-worn device that has wires going to a contact microphone and a lapel microphone. It records everything that is said and experienced by that individual for 24 hours and then it is listened to by humans who count the coughs over that day.

In this meeting, the FDA was very clear that they had issues with the device. Although the device is FDA cleared, it is essentially a tape recorder, which is not the same as being a cough counter.

In contrast, we at Hyfe and others are working on inconspicuous wearables like a watch that monitor sound levels rather than recording. When there’s an explosive sound, it grabs half a second, runs it through an artificial intelligence (AI)-developed algorithm that recognizes if it’s a cough or not, and then counts the cough. All of this happens on the user’s device, so there’s no loss of privacy. Because it’s automated, it runs continually, and you can run it for weeks or months throughout the full duration of the study.

The technology of VitaloJAK is really different than this new wave of cough monitoring. The new wave has brought out a lot of insight into the statistics of cough. Monitoring cough is not like hypertension or heart rate; patients with chronic cough have good and bad days. When you grab a random day, it is unclear whether you’re catching them on a good day or a bad day. For many patients, they have wildly variable cough over the course of a day, week or month.

Twenty-four-hour cough counts before and after are very imprecise in comparison with measuring virtually every cough over the entire duration of the study, and that’s the power of new technology.

Healio: Several committee panelists suggested a need for clearer understanding of what it is about cough that bothers people. How could studies assess this? What tools are needed to measure things such as cough clusters and bouts?

Small: During the meeting, part of the disconnect between the cough scientists and the FDA was fundamentally on two different questions: “How much do you cough?” and “How does your cough affect the quality of your life?”

The first question is a fact that can be counted with a monitor. In terms of the second question, we recently published a study showing that if you take 25 people with chronic cough, ask them, “How bad is your cough on a scale of one to 10?” and measure their cough, they really don't compare very favorably. People don’t really know exactly how much they’re coughing, so if you want to know, measure it. Now if you want to know how it impacts their life, you ask them. The FDA was very clear that they consider none of these patient-reported outcomes tools to be validated.

We don’t know what it is about the coughing that matters to people. For example, two people have the same total cough count — they cough 100 times a day. One person coughs 25 times an hour at four different points of time, and the other person just coughs four times an hour all day long. They’re going to have the same cough count, but they’re going to experience it differently because during coughing bouts, many older women have urinary incontinence and people may even break their ribs if they’re coughing hard. This relationship between these two fundamentally different questions, both of which the FDA needs to pay attention to, was a source of a lot of confusion during the meeting.

Healio: What are some of the limitations of cough counting approaches used by pharma companies?

Small: First and foremost, cough is not measured in many studies for drugs that treat diseases for which cough is a common problem. We’ve gotten into this catch-22 because we can’t measure cough very well and drug companies aren’t focusing on it, but cough is a big problem for many diseases, including asthma, bronchitis and COPD. Because cough is hard to measure, drug studies often do pulmonary function tests instead. Cough is not measured or used as an endpoint despite its prominence in a lot of conditions.

The second issue with cough counting is that cough is measured for 24 hours even though cough is stochastic. If you enroll a patient on a good day and do your follow-up test on a bad day, you can completely miss the impact of the drug.

The third issue with counting coughs is that the intrusive cough monitor can create the Hawthorne effect. The cortex in your brain controls your cough; you can cough or you can suppress your cough right now. When I strap you with a couple of wires and a thing on your belt and say, “Just cough normally,” it’s a bit like saying, “Don’t think of an elephant.”

Healio: How will AI and technologies that use AI change current cough-counting approaches?

Small: In general, the impact of AI on health is dictated mostly by data, not by mathematics, so you need really big data to do AI well. Images are very data rich, and that’s why health care in general, and specifically radiology, dermatology, pathology — are all radically changing because of visual AI.

Sound is the next frontier, and it’s really where a lot of interesting work is happening because sound is a very rich, easily accessible signal. In the cough space, we’ve collected more than a billion explosive cough-like sounds. We’ve had humans listen to about 20 million of those and classify them as cough or not. Then we’ve used AI to train these convolutional neural networks to recognize the difference. We gave it lots of coughs, and importantly, lots of non-coughs, from a very diverse background.

We used AI to develop the algorithms, but then we locked the algorithms. That’s what we’re bringing to the FDA this month, and we’ll ask for their clearance on our software as a medical device.

Healio: What cough-counting devices/technologies are currently being researched and developed?

Small: Trying to measure cough has been an effort that’s been ongoing for more than 25 years. People have tried everything from belts that measure your chest expansion to electromyography. The lack of a really fieldable FDA-cleared cough counter is not for lack of trying, but anybody who started before AI was basically wasting their time.

We can now use AI with the high-quality microphones that run with sophisticated chipsets in phones, smartwatches, smart speakers or other wearables and monitor cough continually and actively. This is not only going to be helpful for drug companies, but it’s also going to be a big game-changer for clinical care.

There are some good studies that show that patients with COPD exacerbations develop a worsening cough 3 days ahead of when they arrive to the ED. Cough is a lead indicator, but people don’t really know how much they’re coughing. Without objective data, many patients with COPD get super sick and go to the ED, but if they and their health care providers were able to unobtrusively and continually monitor their cough, then you would be able to prevent those exacerbations admissions with all of the morbidity and cost to the health system.

Healio: How did you feel about the future of treatments for chronic cough after the meeting?

Small: It’s a complicated time for the cough community, as there are other chronic cough drugs currently in the pipeline. The lack of a validated tool to move from molecules to approved drugs is adding murkiness where we need clarity.