How AI can help detect and monitor coughing in infants and young children

Disclosure: Dr. Jaganath has a grant evaluating the role of cough sounds for TB treatment monitoring.

Transcript [Edited for clarity]:

Contemporary Pediatrics®:

Thanks so much for visiting Contemporary Pediatrics®. I'm editor Joshua Fitch.

Devan Jaganath, MD, MPH:

My name is Dr. Devan Jaganath. I'm an assistant professor of pediatrics at the University of California San Francisco.

Contemporary Pediatrics:

Thank you so much for taking the time. Today we're talking about artificial intelligence tools related to cough monitoring, potentially playing a role in RSV patients. First, can you explain to our audience how these AI algorithms are used, and how it can be incorporated into the pediatric healthcare setting related to coughing in infants and young children?

Devan Jaganath, MD, MPH:

First, to take a step back to really think what is a cough? We all do it, and we all recognize it, but how do we actually characterize it? If we all cough, we think about this kind of short, explosive sound, that we all say, 'oh, that's a cough.' And what a lot of these algorithms do is detect those short, loud noises and make a decision, is this a cough or is this not a cough? The way they've done this is by collecting lots of sounds, thousands, if not more, and have manually had an expert say this is a cough, this is not a cough, and then they make an artificial intelligence algorithm that says, we think this is a cough with this amount of confidence and we don't think that this is amount of cough. After making those and doing it again, over thousands, and thousands [of times], they've created these tools that allow us to determine if a sound is a coffin and track it over time. And so when we think about its role in a healthcare setting, it there's a variety of applications. In its current form, when we think about how we ask about cough, it can be quite subjective. We ask someone, do you have a cough or not? And if we treat someone we ask someone, did you get better or not? From a diagnostic standpoint, that can be very nonspecific, because the many things cause a cough, it's one of the most common symptoms that a child might present in a routine practice or definitely for from an infectious standpoint, and so getting more objective data with these tools really could be quite valuable, because for the first time, we can more objectively assess what is the pattern of cough that led to these symptoms that could help maybe differentiate different types of diseases, and then also any treatment or supportive care that we provide, will that be able to allow us to objectively monitor whether they've improved, which kind of in our current approach, we really haven't been able to do that before.

Contemporary Pediatrics:

Can you explain how these algorithms are administered to the patients, how are they recorded, and it sounds like real time?

Devan Jaganath, MD, MPH:

In the current way that we've utilized these and with this particular application [Hyfe] is utilizing a mobile phone. So it could be really, most phones that can download the application and basically, it just is on and it allows it to capture, again, these loud noises. I know often there's concerns around privacy and recording, and so the way that this application works in particular is that it really is only trying to capture that specific loud, half a second noise, and then determine whether it's a cough or not. As one goes throughout their day, or in an office setting, for example, you may just have it recording, and then it captures, you know, the sounds and then says is the cough or not, and then you're able to track that over time.

Contemporary Pediatrics:

Coughing is a symptom largely associated with RSV in infants and children. Obviously, RSV is a big topic these days, a potential Pfizer vaccine just got FDA support recently. How does this algorithm monitor severity and can it lead to a diagnosis easier? Are there clear cut indications for certain diagnoses using this algorithm?

Devan Jaganath, MD, MPH:

Yeah, that's a great question. Yeah, it's wonderful to see that there's advances in preventing RSV, but it continues to be a large problem that faces young children and being able to detect which children will go on to have more severe disease and need more additional care really is quite important. In terms of the role of role of cough, I think, I would say broadly in the field of acoustic epidemiology as kind of thinking about how we can use sound to detect and monitor diseases, continues to grow and so I think that's an important question, which was what you're highlighting, that researchers and clinicians, you know, should partner with groups such as Hyfe and others in order to help answer that question, because it could be that cough frequency or certain patterns with a cough may be able to predict whether a child will have more severe disease. I think working together and starting to independently evaluate the role of cough and these diseases, hopefully, we can determine is there a threshold that could maybe raise the level of concern that a child has a risk of having a severe disease and be able to provide care earlier? That'd be really great.

Contemporary Pediatrics:

You know, in the way of the world AI is sort of dominating being discussed in the healthcare setting more and more frequently. With your experience with Hyfe AI using these algorithms related to cough, was any kind of skepticism taken away? Can you speak to the skepticism surrounding this relatively new technology and how it's sort of being implemented now in healthcare settings?

Devan Jaganath, MD, MPH:

Yeah, I think for sure, whenever there's any new tool, especially artificial intelligence, I think, as clinicians and in healthcare in general, we need to be skeptical in some form because we expect that any new test really should have a higher level of reliability and interpretability, because that's really going to have implications and how we care for our patients. We want to make, you know, make sure that we're providing the highest quality of care, and utilizing the tools that are the most accurate. I think the key for Hyfe and other cough applications, and other kind of I would say AI tools more broadly, is how can we really ensure that they're structured in place to allow for independent evaluation and research? At the same time, strong partnership with clinicians in order to assess what is the utility of cough sounds and how can it be implemented in the clinical setting and that it's not on its own going to be able to solve every issue, but it's really kind of in partnership and in collaboration with clinicians and researchers. I think, in terms of taking any skepticism away, I think what's been encouraging is really these close partnerships that we've had with Hyfe and other groups in order to really assess whether cough is valuable, and how to really guide how to use it. I think, ultimately,what we've found to date is that cough tracking seems to be able to reliably detect a cough and monitor that over time, as we've been talking about. It's really the question next of how can that help guide whether it's severe disease or any specific intervention. I think that'll be really an exciting next step to continue to partner with these groups in order to care for RSV and other respiratory conditions.

Contemporary Pediatrics:

I know you touched on these algorithms being done in a healthcare setting, can this also be done something left on throughout the night that parents of young children can take to their health care provider and they can interpret it? How was it kind of utilized? Is it utilized in both settings?

Devan Jaganath, MD, MPH:

Yeah, I think that's an important thing to recognize, is this a tool that is just for physicians to kind of remotely assess, or I think what you're highlighting and is an exciting aspect is really, how can we enable caregivers, children, and others to really have a larger role in their own health to help guide that, you know, the next steps in terms of seeking additional care or communications with their providers and I think that is an advantage of these tools, like you're suggesting, whether it's keeping it on at night, and then having this data that they can themselves receive feedback on or share it with their provider could be valuable. Having it integrated into something like a cell phone really enables that to be able to be used by a wide range of people.

Contemporary Pediatrics:

Anything else you would like to add?

Devan Jaganath, MD, MPH:

I think we've covered a lot of it. I think the role of cough sounds is has a lot of potential, as we've as we've mentioned. I think how we've thought about cough broadly, to date has really been quite basic, that it's 'do you have a cough or do you not have a cough? Is it getting better? Is it getting worse?' I think there's a growing recognition that there's so much rich data in the sound that's could potentially differentiate different diseases and also be used for assessing severity, for example. So, you know, I think there's a lot of potential roles to assess how these cough sounds could be used in the future, for RSV, and other conditions. And so, I think, again, kind of steps forward for cost tracking and, and classification in general will be kind of an important role for pediatricians and pediatric researchers to really consider how these tools would be best to use in our practice, and really take a leadership role I think and research and evaluation and, and and leading kind of how how we feel like the best care for our patients.