Tell us about the company, how did it come about?
Hyfe was born from the simple realisation that cough is important but ignored and quickly grew to be a small team whose ambition is nothing less than transforming the world’s approach. Although it is an important symptom for every respiratory condition as well as many non-respiratory indications, no one has figured out a way to unobtrusively measure cough. To this day health care providers rely on patients to tell them `how much are you coughing?”, but they don’t know – how many times did you cough yesterday?
So we went and built an acoustic AI that can run on any smartphone and detect coughs as they happen. When I tell my friends who are pulmonologists “You guys are like cardiologists without blood pressure cuffs” they agree and get super excited by the technology we have developed. The number, pattern and sound of coughs contain an incredible amount of information that is currently ignored.
What have you learned so far since starting at Hyfe?
I have had the privilege of spending the last 35 years of my life in academic medicine, clinical research, and philanthropy trying to solve one of the world’s toughest health challenges. I’m proud of what the teams I’ve run have accomplished, but I’ve learned that there is a completely different style and pace in the start-up world. It’s invigorating to see how much can be accomplished when you combine a great idea, a diverse team and some great technology.
I’ve learned that one of the biggest advantages of a startup is the constant learning mindset. We are always looking for ways to rapidly distil insights and test assumptions. After we released the first version of our model into the wild, thousands of people downloaded it and they started reaching out to us and telling us their stories. We were blown away. So many people suffer from persistent coughing that they cannot explain or treat. All of a sudden, these people could quantify their coughs. They could see objectively how their cough is evolving, and they could identify correlations as they brought changes to their lifestyles. They were able to get their health care provider to recognise the magnitude of their coughs and how disruptive they are in their lives. They also told us what they did and did not like about our app, allowing us to quickly modify it to meet their needs.
We have learned that there are many researchers in many different fields who need a way to collect and analyse objective cough data at scale – something that has been impossible until now. And I’m relearning how fast science can progress when technology unites divergent fields around common challenges.
From a commercial perspective, we’ve learned that there is a strong, latent unmet need in the market for a straightforward, unobtrusive way to track cough – like a Fitbit, for cough. And this need spans many domains of healthcare: pharma, clinical trials, telemedicine, device manufacturers, behavioural modification approaches, and environmental health. We also believe this is foundational – meaning that once we have a reliable, convenient way to track cough at scale, that will unlock layers upon layers of innovation, as patients, providers and payers find ways to create value around this new data.
What challenges have you faced?
As a category-defining innovation, we do not really have a playbook that we could follow – the way a SAAS product might have in a well-explored market. Because of this, we have to be open to opportunities and always learning and be ready to pivot. As a “zero to one” technology we face the challenge of unlocking demand for capabilities that our customers are not aware of having.
The need for our models to run on almost any smart device creates technical challenges. That means that there are endless possible combinations of variables that are outside of our control. Permutations of hardware, firmware, operating system, software and user behaviour will all have an impact on how our model will work in the real world. Add to that all the noises that could be coughs but aren’t: a barking dog, someone banging in a wall, a slamming door, a crying baby. The way we dealt with these challenges is we threw our model in the world and have been committed to improving very fast. As an outcome, all of these challenges are now becoming strengths – and a powerful competitive advantage. By having exposed our models to the real world and continuously improving, we have ended up with the world’s most extensive dataset (140M samples!) and the most sophisticated acoustic AI models, forged by real-world data collected from real people in the real world..
What sets Hyfe apart from other MedTech companies?
We are a science-based company with a laser-focused cough product mindset. Hyfe is a collection of cough nerds – clinical investigators, data scientists, anthropologists, programmers and engineers who are obsessed with cough. Our product mindset is reflected in our process – we built a basic version of our model and instead of trying to optimise it in a laboratory, we went and tested it in the real world. That was brutal. Yet, this is the best way to forge a real-world-proof product.
As scientists, we started with the biggest challenges, overcoming what experience told us would be the two fundamental “traps” in building high-performance acoustic AI models:
The first one has to do with the original dataset used to train the model. The history of AI has a long list of models trained on datasets containing inherent biases, usually related to how these datasets were put together. In acoustics, teams based in universities would collect their first samples by asking campus residents to contribute sounds. Or by setting up “citizen science” websites inviting people to contribute sounds. This allows significant social-demographic biases into the original dataset, which will affect the outputs of the model years down the line. By putting an app on the market and making it available globally, across all types of devices, Hyfe has managed to activate a very diverse, global user base that spans geography, age, and demographics and amass the world’s largest relevant data set comprising over 120 million cough like sounds. This sets us up to create solutions that will perform well in equally diverse settings.
The second “trap” has to do with the nature of the sounds collected traditionally. By asking people to “contribute” coughs, the typical datasets contain a very high percentage of “elicited” or “artificial” coughs from artificial settings that may not contain the same signals that are in the intended use setting. In contrast, Hyfe’s users contribute organic, naturally occurring coughs. Additionally, by tracking cough longitudinally, we unlock additional signals: do the coughs come in clusters or are they isolated? Is there a time of the day correlation (after meals, while lying down), etc? We are learning the pattern of cough is as rich as the sound itself.
More from Interviews
- A Chat with Andrey Suzdaltsev, CEO & Co-Founder at Brightside AI
- A Chat with David O’Hearns, Founder and MD at Creative Agency: Dawn
- Meet Rachel Hartley, Co-Founder of Nutracheck
- Meet Matteo Penzo, CEO and Co-founder of Social E-Learning Platform: zick learn
- Meet Jordan Bucknell, CEO of eCommerce Agency: Upbeat Marketing
- Meet Natasha Morrison, Chief Operating Officer at WeFlex
- Meet Dan Awais-Dean CEO of Property Platform: EverKnock
- Meet Emma Davey, Founder of MyNARA.
How does the Hyfe technology work and how does it improve healthcare?
The Hyfe app runs continually in the background on smartphones. Importantly, we do not actually listen, but merely monitor sound levels. When an explosive cough is detected the app sends a half-second snippet to the cloud – which is not enough to contain any personal information but enough for our AI to determine the number and pattern of cough.
This backend has three different user interfaces which allow patents, providers, health systems and researchers to track coughs in real-time, conveniently, cheaply and at any scale. These products are: Hyfe Cough Tracker – a “Fitbit for cough” – empowering anyone, but particularly those with chronic cough, to unobtrusively monitor their cough; Hyfe Patient Monitor- a “Holter for cough” providing health care providers and systems real-time awareness of when there are significant changes in their patient’s cough; and Hyfe Research – a platform for monitoring cough in a cohort of subjects in research and regulatory quality drug development trials.
We think the ability to passive monitor cough will greatly improve the lives of patients with cough. In the United States alone, before Covid, there were more than 25 million outpatient consultations for cough, costing a staggering $23.5 billion US dollars. Over half of these visits were from the 6% who have a refractory chronic cough – most of whom get little relief despite multiple doctor visits and numerous therapeutic interventions.
The inability to accurately measure cough impedes doctors’ ability to understand their patients’ respiratory needs and Pharma’s ability to develop better treatments, culminating in a vicious cycle of unmet need and under-investment in solutions. We expect that Hyfe can flip this to a virtuous cycle where coughers benefit from newly emerging therapies. Improving their lives and demonstrating demand for even better solutions.
What can we hope to see from the company in the future?
We expect to drive a new era in respiratory care in which quantifying cough will become the norm for prevention and treatment, both at home and in the clinic. The ability to unobtrusively monitor steps (Fitbit) precipitated a change in the fitness industry. Similarly, Hyfe’s ability to monitor cough will transform respiratory health. We anticipate Hyfe will play a comparable role in health care. This will be manifest in traditional care settings but will be central in the booming virtual of diseases such as COPD, refractory chronic cough, bronchitis, congestive heart failure, pneumonia and asthma.
But we will also stimulate rapid advances in cough science. Already, our collaborators are divining basic insights into the epidemiology of cough. For example, they have brought rigorous statistical analysis to the adage that coughers have good days and bad. It turns out cough is not really random, rather it occurs stochastically in statistical distributions that will be used to allow us to promptly alert individuals when their respiratory health improves or deteriorates.
Finally, Hyfe will be central to the respiratory care ecosystem. Empowering patients to quantify the magnitude of their cough will drive sales of new antitussives, quantitative endpoints in clinical trials will accelerate the development of better treatment and the availability of effective therapy will stimulate more patients to seek treatment.