Ep. 17: Abboud Chaballout, CEO and Founder, Diagnoss. Topic: AI and Medical Coding
Kathy: Welcome to Episode 17 of the Smarter Healthcare Podcast. Our guest today is Abboud Chaballout, CEO and founder of Diagnoss. Abboud is here to talk to us about artificial intelligence and how it’s being used to improve medical coding and reduce physician burnout. We also talk about the future of the AI market in healthcare.
I hope you enjoy our conversation.
Kathy: Hi Abboud. Thanks for joining us on the podcast. Could you start by talking a little bit about your background, and about why you started Diagnoss?
Abboud: Thanks for having me on the show Kathy. I’m an attorney by education but I’ve founded a number of companies in the healthcare space, and the reason I started Diagnoss is because I noticed a few issues that were just prominent in clinics throughout the country based on some of the other projects that I’ve launched. And to be more specific, one of the issues that I discovered is in the area of medical coding and billing, where providers spend a lot of time and effort but also find that there are a lot of errors and mistakes that cost them real dollars and cents at the end of the day. So when I launched Diagnoss, the idea was to help providers pick better medical codes using artificial intelligence. But what that has evolved into is a platform from which we’re helping and offering EHR productivity for clinical teams. And that’s what we’re doing at Diagnoss now.
Kathy: And what is that link between medical coding and EHR productivity?
Abboud: Coding is one of those interesting things in healthcare, because it kind of feels like it was just thrown on providers. So the way I think about coding is that it’s a translation of what the provider did, and what the provider is seeing as being the patient’s issue or ailment, into a language that third parties can understand, and the most prominent of those third parties is an insurance company that’s eventually going to reimburse the provider. So codes will either turn into money or they’ll turn into information that third parties will use to manage whatever it is that they need to manage. Another example here is the government using codes to manage population health. When providers go to medical school, they’re not trained in this translation process. They don’t necessarily know how to do it. But in many instances, it’s just thrown at them and they’re expected to do it, and therein lies a lot of issues. And I’ll go into a few of them now, and many of these issues directly impact productivity. So one of those issues is – it might be actually better if I just bring it to life first. So as a provider, you’re going to see the patient, and you’re probably going to spend around 15 minutes documenting that patient encounter into an EHR, so you’re going to write out their history of present illness, you’re going to write out the review of systems and physical exam, and at the end of all of that, you’re going to want to document the diagnosis, or series of diagnoses that are relevant to that patient encounter. And many of the EHRs have you pick those from a drop-down list. So what you as a provider end up having to do is at the end of this 15 minutes, going to a specific page in the EHR, finding the search bar for those diagnoses, and then you just start searching for them. So if you’re searching for asthma, you’ll type ‘ast’ and then you’ll get a bunch of asthma-related codes, and you’re probably going to get a few dozen codes, and you’ll scroll through them hunting for the right one and then you’ll pick it. And if you’re a primary care provider and you’re managing a patient with many chronic issues, you’re probably going to have to pick ten of these. So you’re rinsing and repeating this process ten times if you don’t already have the code memorized in the back of your head. So as far as I’m concerned, that’s a very redundant process, because if you think about it, the provider’s spent 15 minutes documenting. So this asthma was already addressed in the documentation, but now they’re having to go a second time to search for asthma in order to document it in the way that the EHR wants it to be documented. And the reason I think that’s redundant is because why search for asthma if you’ve already addressed it? Why can’t the documentation be the search? And essentially the solution for that to be is to have a machine be reading the notes and giving you those results. So rather than make you search for it again, just give you those results based on what you’re searching and spoon feeding you suggestions that you can pick. And that is the type of solution that we’ve envisioned at Diagnoss and that we’re deploying. But the main point here is that redundancy produces inefficiencies, and those inefficiencies can be made less efficient with predictive technology. Another important part of the story of why medical coding is inefficient is the way that the EHR has you input the information around the patient encounter is very much siloed from page to page. So to bring this to light: Let’s say the meat of the encounter is in the history of present illness. That’s where you’ve documented all of the patient’s issues and all of the surrounding information around that. Where you put the diagnosis is going to be in the assessment page. But by the time you get to the assessment page, you may have forgotten some of the things that you’ve addressed, so you’re now toggling back and forth between EHR screens. So the obvious solution to that is well, rather than force you to toggle from page to page, why not give you your diagnoses or give you the search bar to find your diagnoses on every page, or have it follow you from section to section in the EHR and therefore avoid the toggling and therefore avoid many unnecessary clicks. So those are just two in-depth areas where coding itself is adding to the amount of time that it’s taking a provider to sign off on a note and get to the next patient.
Kathy: I think one of the really important things that you’ve just brought up too in terms of this time and efficiency, I mean I know that there’s a big conversation going on in the healthcare world right now about physician burnout. And it seems like all of this contributes to that burnout and the time that physicians spend on the EHR.
Abboud: Absolutely. 100%. When you look at the statistics around burnout and the studies around burnout, they’re often centered around two big issues: Administrative work that clinics and hospitals are piling onto the provider, and the EHR. So two different things – they’re often conflated, but two very different things. And coding is one of those things that’s both administrative and EHR-related. And that’s one of the reasons why I’ve particularly - I’m focusing on it, but I’ve also seen it be a problem in some of the other companies that I’ve started and managed. Another interesting point about coding is you do have medical coders who are certified in medical coding. So they go to trade schools and they get the certification and they do the work. So clinics and hospitals will often have in-house medical coders or outsourced medical coders, and they have technologies that help them. They have technologies that help them get to the right answer. But providers don’t. They’re almost left to their own devices. Really at best all they have is that search bar in the EHR. They’re not really provided much support. So we – I personally view it as a gap in the spectrum of solutions that are out there and the spectrum of technologies that are out there in the world.
Kathy: And it seems like medical coding is just – it’s one of those things that sounds mundane, but really has a huge importance in healthcare.
Abboud: Oh, absolutely, and just to go back to a thought – I didn’t really finish it, which is that if you look at the studies of burnout, coding- and billing-related issues are directly cited as sources of burnout, and the EHR itself is cited as a source of burnout. So, the way I look at it is, as consumers we have a lot of technology at our disposal to help us be more efficient, but really what it’s doing is it’s helping us save time and it’s helping us reduce our personal burnout. But providers don’t exactly have that today.
Kathy: Great. Now can you talk a little bit more in-depth about how exactly Diagnoss works?
Abboud: Absolutely. So what we’ve launched with is an AI assistant. And it’s basically a sidebar that integrates with a provider’s EHR. And it integrates in such a way that it feels like it’s a part of the EHR, so they’re not dealing with two disparate systems, two separate windows that they’re toggling back and forth from. And what it does is it breeds and it analyzes various metadata surrounding a patient encounter, and/or the doctor’s note itself to start recommending codes to them. And it will not only recommend codes, it’ll recommend documentation improvement suggestions, and one note here is – depending on the type of work that a provider will do they either look at codes as codes, or they look at them as diagnoses that they’re going to chart in the EHR. So there’s a little bit of an importance of how language is used, but it essentially will predict the codes or the diagnoses, and with one click a provider can add that diagnosis both to the chart and to the billing screen within the EHR. And our hope is that providers will be able to pick codes faster and more accurately given the level of intelligence that the system is providing them.
Kathy: Sounds interesting. Now I know there’s been a lot of talk about AI in healthcare over the last several years – what do you think we’ve gotten right about AI, and likewise what have we gotten wrong about it?
Abboud: Well, what I think we’ve gotten right is an understanding that AI will play an important role in our healthcare system. So I think that is spot-on. And the reason for that is because we see the role that it’s playing in every other industry. What I think we’ve gotten wrong is our understanding of what AI is and how it will play the role that we want it to play. And there’s a couple of things where I think we’re just, generally speaking, we’re going to get wrong in terms of our understanding. First of all, AI itself is such a broad term and it’s such a broad technology. And it’s not new. AI is really synonymous with pattern recognition and the packaging of that pattern recognition in a way that’s useful to a user or an organization. And when you think of it that way you can imagine there are some very basic patterns that we can focus on recognizing, and there are some very difficult and complex patterns that we can focus on and recognizing and that the AI itself is not going to recognize everything. It’s very much about pattern recognition and furthermore the scoping of that pattern recognition. What is the pattern that you want it to recognize? So that’s the first thing. The second thing is, and it kind of dovetails with that, is that we think of AI as this black magic and it’s not. Again, it’s very much an iterative process that helps a human either predict something or recognize something without necessarily putting in the effort to do it themselves. But what it’s not going to do is think on its own. Although we speak of it in human terms, at least in this stage and at least for the foreseeable future as far as I’m concerned, it’s not thinking in a way that a human is thinking, but what it is doing is recognizing patterns at scale in a way that a human won’t be able to do, if that makes sense. So that’s another important point in my opinion. And the last thing here, just dovetailing off of these two points, is that we often think about AI and the algorithm that’s crafted as the important element, and it is, so I don’t want to undermine that, but really the most difficult and the most important aspect of AI is the infrastructure behind it. All the data piping that’s behind it. And more importantly is your ability to match the right type of data to the right type of question. And that infrastructure is the hardest part to build out. The algorithm is arguably the easiest, and there are a lot of open-sourced algorithms that people use, and even if you’re writing your own algorithm, that’s as I mentioned arguably the easiest part. The hardest part is the infrastructure around it. And another way of saying this is that AI is not necessarily a thing, it’s a system. So if – and I often explain this to healthcare organizations because – the need to explain this often comes from the conversation that I’m having with them where the expectation is hey, the system is going to do everything. But really it can’t do everything, it’s only going to do something that’s as good as the data that is at our disposal, and the ease of access to that data, and our ability to narrow the scope of what that system is going to accomplish.
Kathy: Now do you think that COVID-19 impacted the AI market in healthcare, and if so, how?
Abboud: Absolutely. The first reason why is, as the old adage goes, necessity is the mother of all invention. And COVID-19 really put pressure on the healthcare system in a way that it hasn’t for a long time. I’ll give you an example. So after college, I had the opportunity to go do telemedicine research abroad. And this was in 2008 and at that time there was a lot of discussion about telemedicine, telehealth, and how awesome it was going to be, and then you had your naysayers and you had people who were prophesizing about the future, and if you think about it, we’ve had a slow and steady adoption of telemedicine over the past decade, but it’s really – COVID forced us to really depend on it 100%, and people are realizing, hey, we could have been doing this all along at this scale. But many people just didn’t really want to adopt it for a variety of reasons. So the pandemic really put pressure on cost for the system, and it really put pressure on the rigidity – it really put a question mark on the rigidity of our healthcare system. We realized, look, we need to be nimble at moment’s notice. Whether it’s because now we need to treat patients from home, or it’s because now we need to figure out how to use our EHR to roll out administration of vaccines, and we sort of kept running into these issues. And definitely, we’re seeing now the investment in healthcare technology across the board from organizations, which is really cool to see.
Kathy: Now often when we’ve thought of AI in healthcare, we’ve thought about these sensational applications such as robots performing surgery, but it seems like that’s going to shift now to AI that addresses issues like we’ve talked about – physician workflow and burnout. Why do you think we’re going to have that shift?
Abboud: That’s a great question. I think first and foremost it’s just time and interaction with this thing…right now the narrative is sensational, but just like with anything these sensational goals are – they take time. I’m sure you’ve read about IBM Watson and their missteps and how the feeling was is that the performance was going to be much much greater than it actually was. Now personally I think that the goals of Watson are awesome. I mean, that is – these are the ideals that we need to set in place, but truthfully, I think there’s so much more lower hanging fruit that can have an impact today in the world of AI. So I think definitely the promised land is predicting disease and that’s what we all want, but there are so many things before predicting disease that are much easier to accomplish and will almost guarantee value add in the day-to-day of both patients and the caregivers and those are coming out. There are so many start-ups and technology companies working on these things as we speak. I’ll give you a very basic example of something that – just to make it a little bit more consumer-facing. Right now, as consumers using cell phones, and as you’re texting a friend, the technology behind whatever texting app you’re using is probably giving you predictions about the next word that you’re going to use, and as a consumer it arguably makes your life that much easier and that much better. Now was that technology the sensational thing that was all over the internet? No, it kind of just happened, like as consumers – I don’t know how your experience was, but my experience using my cell phone is it kind of just happened and it was just there. And I don’t really think about it as AI but it is AI. We’re interacting with AI technology. And that I believe is going to seep into our healthcare system, and the more that it seeps into the healthcare system, the more we stop thinking about AI as this sensational thing and we realize hey look, this is a – there are much more day-to-day use cases that we’re going to be interacting with.
Kathy: Now, if we’re looking to the next year or two, what do you think we’ll realistically expect to see when it comes to AI in healthcare?
Abboud: Exactly what I just mentioned, is multiple use cases where you don’t even – you probably won’t even know that you’re working with an AI system. But what you probably noticed was hey, this thing that I’m used to doing just got that much easier. I’ll give you an example. I personally haven’t seen a doctor in a while, so I’m thinking – so I thought to myself, hey…or actually my wife told me I need to get a check-up. So I ended up scheduling two appointments with two different clinics as a new patient. That process took me, believe it or not, 30 to 45 minutes. By the time I got ahold of the right person to schedule my appointment, by the time I spelled out my name and the person on the other line not getting it right, by the time I spelled out my e-mail address, by the time I gave my insurance member ID, and then by the time I was asked to flip the card and give them the phone number on the back of my ID card, right, that was time-consuming. It took a long time just to get an appointment scheduled. Now I get it. There’s a lot of issues. You have to get your insurance eligibility verified. You have to – the provider has to get a variety of things right in order for them to see you in the first place. But all of these little things AI will play a role in – or, sorry, I should say many of these little things, AI and a mixture of other technologies like robotic process automation technologies, will play a role of in the back end making it just more elegant. And I think that’s what’s going to happen in the next one to two years is these smaller applications are going to enter the marketplace with this more accelerated adoption of technology in the marketplace and it will help make our interaction with our healthcare system that much more elegant.
Kathy: I was just thinking as you were describing that process of signing up for a new doctor, wouldn’t an AI chatbot be great in that case.
Abboud: Yes. Yes, absolutely. Absolutely.
Kathy: Now, if we were to look a little bit further out, let’s say five to ten years, now where do you think the AI market’s going to be in healthcare?
Abboud: Now, that’s where I think there will be the more sensational disruption. Where certain parts of clinical workflows could depend on – and when I say clinical workflows I’m talking about the actual treatment of a patient, where there are AI systems that we can depend on. We’re starting to see this in areas like radiology. Still not there, but we’re making progress. And I think in the next five to ten years we start to see real use cases there and that’s where I also start to see different parts of the healthcare system that are dependent upon heavy labor be impacted as well. Where organizations that are producing certain type of skillsets now have to rethink where is that skillset going to be important? Now in the five-to-ten-year horizon I still don’t believe that you’re going to have mass replacement of humans in our healthcare system, because I think our healthcare system is in such need of help right now that we have a shortage of good professionals across the board. Whether you’re talking about a provider, or you’re talking about a medical coder, there is a shortage, in my opinion, of really good professionals. And where I think the disruption in the five-to-ten-year horizon is to become dependent on some sort of automation, leaving our really skilled, and the people who are really good at what they do to cater, to the things they should be catering to. Not to the low-level things, and I think that’s what we’re going to see in the five, ten horizon.
Kathy: Well, Abboud, thanks so much for joining me, this was really interesting.
Abboud: Absolutely. Thanks for having me.
Kathy: Thank you for joining me for this episode of the Smarter Healthcare Podcast.
To learn more about Abboud’s work at Diagnoss, you can follow him on Twitter @abboudchaballo2. Follow the company on Twitter @diagnoss.
You can also follow me on Twitter @ksucich or @smarthcpodcast. Feel free to get in touch with comments or guest suggestions.
To listen to more episodes, visit our website at www.smarthcpodcast.com or find us on your favorite podcast app. I’d appreciate if you would subscribe, rate, and review.
Thanks for listening!
