Naji Shafi, director of healthcare AI at AWS
Photo courtesy of Amazon Web Services
Amazon Web Services announced today the launch of Amazon Connect Health, an agentic AI that integrates with a health provider's EHR to perform tasks associated with administrative work, such as patient verification, ambient documentation and more.
Naji Shafi, director of healthcare AI at AWS, sat down with MobiHealthNews to discuss Amazon Connect Health and how he sees artificial intelligence enhancing healthcare in the future.
MobiHealthNews: Can you tell our readers about Amazon Connect Health?
Naji Shafi: Earlier last year, the team and I looked at healthcare, and what's been super interesting is that healthcare workers are drowning in administrative complexity. It's tough – the burnout, the pajama time, all of that. Eighty-nine percent of patients tell us care navigation is a core challenge, and that's why they're switching providers. Staff tell us they're spending 80% of their time on calls, just compiling data, trying to figure out what to do.
And so, what's been interesting is when we looked at this, what was fascinating for us is that all those challenges across the entire care journey, it starts from the first time they're trying to get care, trying to figure out, how do I see a doctor, all the way until after the visit when they're getting discharged. And then, if you think about it, all of the administrative tasks that need to happen across that entire journey.
So, the thing that was interesting for us is that we really focused on the administrative tasks. And so, we tackled patient verification, appointment scheduling and management, patient insights, ambient documentation and medical coding.
Really, the key pieces here are that these are all in one. With Amazon Connect Health, you're getting that entire suite of capabilities all in a single solution, and it's really focused on nontechnical audiences, zero implementation, you get value from day one, and it's really building on top of the decades of experience that Amazon has had in healthcare.
MHN: Is it an offering that is integrated in a provider's network?
Shafi: So, there's actually probably two primary use cases that I would segment, and it's actually very natural when you think about a patient journey.
As a patient, you're trying to get care; you're trying to essentially contact your health system or your clinic, and so, that's where the patient verification [comes in]. When you call in, if you think about it, the first thing you always do is authenticate who you are. You give your social, your last four digits, or you give your date of birth or your address. So now an AI agent will agentically authenticate you and verify you, and then, essentially, what I joke and say is, unlock the rest of all of the capabilities.
So, the next step is appointment management. I want to see my primary care provider, and the traditional experiences in that scenario were really very rigid if you think of, like, press one, press two, or it's kind of like the more traditional chatbot experiences. Here, these are all generative AI-based, LLM-based, so these are very natural.
So you can say, "I want to see my pediatrician. My daughter's sick, and I need to book an appointment with my pediatrician." And it will think through as to who is your established relationship with? What time slots do you typically book, and look at all of that existing history in the EHR, and then pull that and that's all in, if you think of it from patient engagement experience, that's really patient focused.
But also if you have a contact center, like, for example, our customer UCSD [University of California San Diego] has contact centers, so that's really where we're building into Amazon Connect.
The second set of scenarios, which is like, well, now you've booked an appointment and you're coming in to actually see a clinician, or, let's say, a PA or a nurse. Now, typically, the PA or the nurse or the clinician has to essentially look at your medical history, try to figure out what's happening, what's your chief complaint, and why you're coming in.
And then, very often, even from your own personal experiences, they probably just give up and just ask you, what brings you in today? And so. what gets interesting is this is where patient insights come in. And so, at the point of care, the patient insights, ambient documentation, medical coding, those are all integrated into your EHR. And so, we have our launch partners, Netsmart, Paradigm and Greenway that we're working with. Specifically, Netsmart we've been piloting for over the past six months, and they've had pretty dramatic results.
So, taking the patient insights piece, as an example, I think what's super fascinating about this is that everything we're building is grounded in terms of data, and so clinicians, physicians, their staff, they're pretty, I'd say, finicky – like if you don't earn their trust, they're not going to use your products. So, what I mean by that is that when we generate insights – like, okay, Naji is coming in, he has a history of hyperlipidemia, he has sleep apnea, etc., chronic conditions – those insights that we generate, whether they're chronic conditions or tied to the provider specialty or the chief complaint (the reason why they're coming in), all of those insights that we generate and show to the user in the EHR, we actually have evidence mapping.
So, then, as a clinician, if I'm like, "Well, I don't remember this about this. I know Naji, and I see him all the time, and I don't remember this," they can actually click and see where that insight is being pulled in from. And now, when you go into ambient, and this is where we get super excited, that context gets passed into your ambient clinical documentation service. So, then your documents are actually more accurate and better.
And so now, when you're, let's say the conversation is over, the patient and doctor had a very natural conversation, now, when the doctor turns back to their PC or laptop and is trying to now, let's say, finalize the note, they get the codes generated right there and then. And then dynamically, based on the codes that they select, that note can actually get updated.
What's super interesting is that if you think about large language models, like the foundation models, they all inherently understand FHIR. They all understand that format. And so the reason why we've built all of these capabilities on top of FHIR, of course, it's interoperable and we love it, but also because these LLMs understand it.
And so, we're actually building on top of HealthLake, our FHIR data store, but then we look at that as essentially our AI-ready data infrastructure. And so, for us, we are actually also, on top of Healthlake, and I'm happy to share with you that we're also trying to help our customers transform more of their data into the FHIR format so that they can unlock it and do more use cases with it.
I think a core part of this that we get super excited about is all of this is building on top of our FHIR service. So, we have a service called HealthLake, which is historically, if you Google it, if you look it up, it would be our FHIR transactional data storage. That's where you can store any interop data.
MHN: So, with Amazon Connect Health, the AI understands the natural way of speaking, like, "I have to see my doctor because of this reason" and it understands you?
Shafi: So, of course, the way we've designed it is we have to have guardrails. We don't want something, like, we don't want people chatting away about random things. So, the AI has been trained or has been guided to focus on extracting intent. So, the whole idea there is that, well, you're calling in. Well, if you haven't verified, okay, before we book an appointment for you, let's actually verify you first, right? So then, I can look up all your information. So it's actually a very natural interaction, versus if you think about traditional ones, it would be like press one for this, press two for this, and by the time you get to the department only then you can book the appointment. Here, it's much more natural, like human conversation that you have.
MHN: Is the AI updated in real time with the EHR?
Shafi: Yes. That was super key for us. So, for our patient verification and appointment management, we are integrating directly with Epic, and so, we are looking up your real-time information.
MHN: Tell us something else you're excited about.
Shafi: What's been super interesting is if you think about, just for me, I find it almost bizarre that in your interactions, my interaction with most health systems, more clinical organizations, is that it is just hard, like, while AI is so prevalent in so many different industries, like, you know, you talk to a leading airline, customer service line or Amazon retail, everything is typically self-service, super rapid, etc. But for healthcare, that technology has not really, I'd say, been democratized.
And so with UCSD, what's been super fascinating is that we brought Amazon Connect Health, and with the 3.2 million patient interactions across contact centers, we're able to save a minute per call. You extrapolate that, that's like 33,000 staff hours saved annually, and what's been also super cool about this is that they actually saw a 30% average reduction in call abandonment. Patients were sticking around because, wait, this experience is great. I'm getting to what I need faster, and they're not just frustrated and dropping off.
MHN: I'm interested in the medical coding aspect of it; it generates ICD-10 and CPT codes. With the ACCESS model that CMS released recently, the way that they are doing a payment system is much different. They are not going off of CPT codes or ICD codes; it's just a standard set of costs. Is there going to be any aspect of Amazon Connect Health that will allow for integration with how the ACCESS model works?
Shafi: So, right now, what we're focused on is really the ICD, CPT and E/M codes. For us, with the ACCESS model, with value-based care, those are things that we're always, always, like, as Amazon, if I don't say the words customer obsessed, I'd be doing us all wrong. I'm really keen to hear more about how we can better help in those scenarios and in those use cases, and especially with ... this space is also evolving with the regulatory changes. So, I'm super keen to hear more, but right now, really, we're focused on the ICD-10, CPT and E/M codes.
MHN: When is Amazon Connect Health launching?
Shafi: It is actually launching on March 5. So, patient verification and ambient documentation will be in GA [general availability]. The other features [appointment management, patient insights and medical coding] will be in preview.
MHN: This is a very fresh announcement, but I'm really curious. When you think about the future of AI in healthcare, is there anything that you really want to work toward at AWS, like, you want to make it happen? Is there something you feel passionate about?
Shafi: I mean, there's so many things. The biggest part for me is like, well, first of all, healthcare organizations, they have so much data. They have so much data – clinical notes, multimodal data, imaging, SDI, like there's a ton, and so, how do we help all these healthcare organizations unlock their data? That, for me, is one of the first steps.
The second thing is, well, if I think about it just from a patient perspective, we talk about appointment management. If you're trying to book an appointment, but, oh, guess what, the office is closed. You can't book an appointment, and you work a job and, unfortunately, you can't make phone calls on your job. Patient access is a challenge.
And you know, it's funny, but like doing something like patient verification, appointment management, and just having a 24/7 phone number that you can reach and just book an appointment and just get it done. It is such an unlock for so many patients, and so for me, that's really where, when I think about patient access, it's one of, like, my team and my kind of goals is to do more in that space.
