This Week Health

Interview in Action Fall '23 - Pranay Kapadia, Co-Founder & CEO, Notable

October 10: Today on the Conference channel, it’s an Interview in Action with Pranay Kapadia, Co-Founder & CEO of Notable. What pushbacks are common from organizations regarding AI implementation? How is Notable using conversational AI to help with scheduling, bill pay and more to improve the patient experience? What is the difference between a chat bot and utilizing an LLM to guide the patient experience?

This Week Health Subscribe

This Week Health Twitter

This Week Health Linkedin

Alex’s Lemonade Stand: Foundation for Childhood Cancer Donate

Transcript

This transcription is provided by artificial intelligence. We believe in technology but understand that even the smartest robots can sometimes get speech recognition wrong.

Welcome to This Week Health Conference. My name is Bill Russell. I'm a former CIO for a 16 hospital system and creator of This Week Health, a set of channels and events dedicated to leveraging the power of community to propel healthcare forward. Today we have an interview in action from the Fall Conferences on the West Coast.

Here we go.

Alright, here we are from Noteworthy, a notable conference, and we're here with the CEO for Notable, Pranay Kapadia. Thank you for the invite, by the way. I really appreciate it. Thank you for joining us. This is fantastic. I got to hang out the last two days with a bunch of digital officers, CIOs, talking about where we're going to be going with AI.

Those kinds of things. I think it's really fascinating. What are you seeing in healthcare with regard to AI? How are people approaching it? And what are some, what's some of the pushback that we're, we're feeling?

There's probably three camps or three trends that we actually see. One this crucial moment in time where health systems are actually eager to change their margin structure.

Eager to change how to get their employees and teams to actually operate at peak. Because of which, they're actually excited to immerse new technologies or ways of doing things. I think the inflationary pressures, COVID, has actually helped accelerate that. The second, incredible fear, around...

Something is going to

change, dramatically.

What does the seismic event mean?

Could there be an inflection point that's caused by the introduction?

The thing that is actually unique is this is the fourth wave of AI.

Interesting. I realized we were using AI back in the day, but what is,

what do we see? If you actually go back and you look at it, the concept of a neural network started in the 60s.

Right. With, we cannot just deterministically program things, we've got to actually create neural networks based on how the mind works. And so that started in the 60s, and then you actually had a leapfrog around building out neural networks for machine vision. And then you had a leapfrog for speech synthesis.

And now we've actually gone into a whole new era of large language models. Which is... It's no different than the three generations prior, and yet there's this trepidation around should we embrace it, should we not? We have to embrace it, but how do we go about doing it in a safe way, because usually healthcare is risk averse, and I say usually, but we're at this critical point and juncture that we see where when AI applied right, you can get humans to operate at peak.

You can get care to improve at a different cost level, and so that's actually really exciting to be a part of right

now. Well, that was the interesting thing. When we have a conversation around AI, and I do a lot of interviews with people around AI, bring up a lot of different things. Oh, we've got to get the data right, we've got to do this right, we've got to get this right.

But I walked through the demo center, and the thing I really appreciated is, by layering the different AI technologies. We're able to get past a lot of the traditional challenges we've had, where we know that 20 percent of the data is structured, 80 percent is unstructured. Some people will say it's 30 70, it doesn't really matter, 80%, 70 percent is unstructured.

Dotable Platform has found a way to pull that data out and make sense of it for various use cases, and we'll go into some of those use cases. And that really I think addresses one of the big pushbacks that we hear from organizations. What other pushback are we hearing?

The usual ones that we actually run into are, actually, anytime you start with, what is the problem, that's the right place to start.

Anytime you start with, our architecture isn't right, and we don't have our data in the cloud, or we don't have an AI strategy or AI governance structure. You have just wasted 18 months of your organization's time. Versus, what are the right problems that we should actually tackle that technology can assist with?

The ones that we have found in our experimentation with these larger language models you know, we're backed by investors both big and small, but all through Silicon Valley Roots, that actually founded these companies. Like the OpenAIs the IntuOpics. And so we've actually had access to these for the last 9 12 months.

To actually understand what works, what doesn't. And some of what we've seen is... It needs very little data from health systems. In order to drive exponential value... I'll give you one example of this. With one of our partners, we actually looked at utilizing large language models. To extract information from unstructured data.

Identification of risk of patients. Identification of care gaps that were actually in PDF documents. So, electronic health records digitized the paper, but didn't actually give you discrete data. And so you still have humans trying to extract that information. Right. We saw a 30x improvement in efficiency from what humans were actually doing for chart extraction.

to what LLMs could actually do to understand when was the last time a patient had a colonoscopy. What do the last lab results say? And

it's the layering, right? So, one of the things I saw was faxes coming in. In the prior auth, in the authorizations. And fax comes in, NLP looks at it, which is what we would normally do.

Then essentially LLMs, Gen AI models look at it and say, da da da da, let's do this. And then the whole thing gets organized the same way it does today, except it's a person who's in there going, Oh, there's that. I've got to do this. I've got to do that. Okay. But like an hour later, they think they've done one record.

We've clocked this. We actually saw some of our partners that we work with on the referral side as an example. And this is, again, we don't even need data from the health systems to actually power this in so many which ways. There's this thesis of, like, I have to give up all my data in order to avail of the benefits.

But it's similar to, I'm going to move to the cloud to actually not have my data center anymore. And you do that because of the benefits without actually giving up the control. And so the referrals is a great example of LOMs that we've actually seen tremendous success in. One of our earliest partners actually had, it took them 21 days.

On average, from the time a referral actually came in, till a patient was scheduled for that referral. What they didn't know was, was it the most important? What was the pay or mix of the acuity high enough?

And the health system could set those rules.

They could set those rules but they weren't setting those rules because the employees were just going in first in, first out.

Yeah. And not even being able to get through the entirety of the queue. What we've seen in the three months that we actually went live with them, 25 percent of them autonomous. From the fax to the patient self scheduling, we've been actually able to do without a human involved. What was 21 days is now 3.

6 days.

So it can look at acuity, it can look at, I hate to say this, but revenue, essentially. It can look at a lot of different factors. It always

comes down to three things. It's availability, acuity, because if those two don't match, and then the last one is profitability. And how do I actually balance those three?

And every CFO we talk to, every operator, is trying to identify how to make all of that work, because with all the goodness, we can only, if you just go after acuity, Right. you might not have a business to stand on. Right. And you can't solve the availability problem, because that is up to physician schedules, and there's ways to manage the templates.

But what you don't want to have is four and a half people. So we have a team of people per physician working on mining this unstructured data and calling patients and that's the exciting part of where we are right now with LLM is that we just get to be this inflect, we get to inflect collectively with our partners. 📍

  📍 We'll get back to our show in just a minute. Having a child with cancer is one of the most painful and difficult situations a family can face. In 2023, to celebrate five years of This Week Health, we have partnered with Alex's Lemonade Stand all year long with a goal of raising 50, 000 from our community.

We've already achieved that goal and we've exceeded that goal by 5, 000, so we're up over 55, 000 for the year. We want to blow through that number. We ask you to join us. Hit our website in the top right hand column. You're going to see a logo for the lemonade stand. Go ahead and click on that to give today.

We believe in the generosity of our community and we thank you in advance. Now back to our show. 📍

 I

just, I thought it was interesting. I just talked with Kristen, North Kansas City Hospital of Maritaza and it's as big as it sounds, right? It's North Kansas City. I mean, it's, it's not small, but it's not, it's not large. they're implementing the platform, they're working with NLP, they're working with LLMs, they're working with JIN, they're working with all these various technologies.

And I asked her, like, so did you hire, like, are you worried about the skill sets? And she's like, no. It's a platform, it's a no code platform that essentially, it's using all this stuff in the back, but I didn't have to worry about that. I just, what I have to worry about is... what's the next use case that would bring the most benefit to our system or the community, the people on social media?

We talked a little bit about outreach. Talk a little bit about the population health and the outreach. I thought it was amazing how automated that was as well. Yeah, so

the unique pieces about the population health is what we've seen is often about 30 percent of the time systems reach out to patients.

We found this out the hard way. But there's too much friction coming back. No, the first one was majority of the outreaches were having to have, you know, about a year and a half ago, majority of systems were actually doing the outreach with paper and nobody was responding. Then we started to digitize that portion on the platform and what we discovered was either there were no slots available, and so we had to then build out intelligence to figure out who to reach out to only when slots are available.

then we identified with our partners, with, with Kristen and team that about 30% of the time when we were reaching out to the patient, they already had that information somewhere in their chart, or it was in a record that they got from another provider. And so we were now outreaching to a patient that really didn't need to be reached out to, and that experience actually compounds, right?

How many times are you get you, you get pinged again and again and again. You're like, I'm gonna turn this off. I wanna opt out. Stop. Texting me, stop messaging me. And so then we start to use LLMs with them around how can we actually chart scrub before we do the outreach to identify who can actually be available with that care.

Who's already closed it somewhere else. And now they've extended it to tie in with the registration, which is actually really unique. So if Bill is coming in at Kansas City Emeritus at NKCH Emeritus, Bill will not only be able to do his registration, I'll just walk you through it. Thank you. AI that's actually determining when to ping Bill, determining what forms Bill actually does need to provide, is determining if he provides his insurance card, what are the fields on there that actually need to be extracted, all the way down to the P.

O. box and the 800 number on the back, because that then ties to the payer plan matching that goes into the EHR to prevent the downstream denial. That then actually looks at his historic charts to say, Hey, it turns out your last colonoscopy was 18 months ago. You will be due for another one.

Would you like to schedule it now, three months out? So you've actually now done all of that in one personalized experience without the work queue, without the fax. And so the same things are now being applied to referrals. So if you did have a referral coming in as well. That is sitting there. Through that personalized experience, Bill can now also schedule his MRI all in one seamless flow.

So that's actually the, and what we've had with partners like Kristen is allowing us to work with her operators. Right. To showcase, to understand workflows, and then bring the very best of technology, bring the very best of experience. In a way that I don't believe the industry has done before. I

want to talk about a system, but before we get there I mean, I love the fact that you're bringing a platform for AI.

So there's people who are trying to figure this out. It's already there, but I've referenced this now a couple times. You had your phone out yesterday when we were talking and you were going through a Slack channel and it was all the responses from patients. And, like, I don't know, a ton of them were green, but the red ones, you were like, that's something, we're gonna work on that.

You're getting that instant feedback, which is really interesting, plus the positive feedback. I mean, hey, this was so easy, I just scheduled my four kids in like a couple minutes. I mean, that kind of stuff is, really powerful to get that kind of of feedback.

It's why we do what we do. At the end of the day often we'll, we'll debate internally is what we're doing.

Best for us as mere mortals as people. You could call us consumers, you could call us patients. We're just people. And are we doing it in a way that is safe? That we would want our data utilized? I'm a patient on our platform and through providers that are partners with us who I've actually seen all of my data is in the system.

How do I want that to be utilized?

I just love it. I mean, so it pops up and it knows, , this is my carrier, this is my payer. So these are my potential providers. It's figuring all that stuff out. Yep. Which I used to have to figure out.

Or, anyway, there's... And so that feedback though, Bill, which has been paramount.

Just the connectivity back to patients, because often when you go into the technology and you understand how LLMs work, and AI is cool and sexy, and you're dealing with, you know, health system problem of the day, we've served over 32 million patients in this country that have gotten care, we've powered their care in so much way.

And all of their feedback, as they use the system, flows into Slack for the entire company to see. And through Flow Studio, for our partners to be able to see and then benchmark against their peers. Which just makes it that much more real. And holds us accountable, because often, we'll get something negative.

Right. And we get to actually look back, was it a configuration issue? Was it an operational issue? Because you have to break ties. Sometimes it's web cycle versus clinical versus operations. And now you have the data from patients to say, Hey, it turns out you were actually asking for this consent form every single time and your patients hate it.

a way around it. Can we work together to configure the platform slightly differently to actually suffice it? Streamline things for everyone.

I don't want to give people the impression that you're just working with small health systems. You have some very large clients. Obviously, 35 million is a significant number of patients.

Talk about Assistant. It's interesting, so we've trained this model essentially on the health system and on some of the payer information and all this other stuff. Assistant is something you're rolling out now. Yep. Talk about its inception and

where it's going. So, at a very core, our Our mission has always been, how do we simplify and optimize healthcare for humanity?

And with that has been, how do we eliminate the mundane? If we're spending over a trillion dollars on the mundane, phone calls, faxes, work queues, how can we distill that? How do we eliminate that? With large language models, we actually saw the opportunity to kill chatbots. We saw the opportunity to kill phone calls.

65 percent of all engagement with every health system is still over a phone. And that's probably the lower end for most systems. The

call center at our health system against 16 hospital system, we had multiple call centers and they were significant

in size. And it's complicated. Like one of the CIOs I just talked to recently of the largest non profit in the country who's a partner of ours.

It's just like nobody wants to test their phone system. And I cannot go to my board and say, hey, I need 30 million to go change my phone system. But, if I can go from 90 percent calls to 10 percent calls, can you pave that path? Or what does that look like? And with the advent of LLMs, what we've actually seen is the ability to really set the bar for engagement.

We no longer ever follow airlines do this, and FinTech does this. I spent 15 years in FinTech. I'm tired of saying I used to do this in FinTech. I actually want to talk about what we can do here in healthcare. And so we designed these systems.

So the LLM is a natural language front end to everything behind it.

And you can put it on your website, on in your mobile app, in your mic chart, on your Cerno portal, what have you, in a way that is seamless for patients to actually engage.

What's the distinction between a chatbot?

So if you actually think about chatbots, historically, chatbots are more like a decision tree, hard coded.

They took about six months to actually stand out. With the Assistant, we've actually set it up in a way where, to your earlier point, as the AI platform, you can utilize any of the best in breed LLMs. We've created some of the plugins that can actually integrate with your EHRs. To help with scheduling, to help with routing, to help with bill pay and financial assistance.

And then we provide the tools for our partners to actually extend them themselves. The craziest thing about this, and the most exciting thing about it, is we can now program in English. You can take the call center manuals that you have, the care coordinator manuals that you have, and the instructions that you would actually utilize for them.

And load them in as plugins. We actually have our internal operators think. Folks that have actually come from the UCSFs, comes from the HSSs, that ran Access, that ran all of these operations, whether it's Revenue Cycle or Patient Access or Call Centers, and had them actually build out these plugins, and it's blown their minds where you can actually start to upload.

Here are the instructions of how I would tell my call center reps to actually triage patients to send them to a certain acuity of care. Or here's the way that I would actually check to see which providers I want to showcase through a digital navigator. They would have concierge members doing this, the assistants now democratize it.

So we, we created the assistant as the fastest way for patients.

It's

interesting how much time we spent on that homepage. But it still required, like, the person to navigate. We tried to make it as simple as possible. We reduced the number of boxes and things. But if you look at most health systems homepage, you're like, Am I this? Am I this? Or do I click on this? And you don't really know what to do.

I assume there's An authenticated experience and a non authenticated experience.

That's exactly right. And one additional piece to that if you go to any of our partners that are already live with this we proactively steered away from the chatbot bubble in the bottom right. And... You want it to

be the...

It has to be the way to navigate. The world is going to move from search links and trying to figure out where to go to answers and actions. And that's what the assistant actually helps you do. And so we worked with them on how do we start to coach and train your patients, your people, your consumers to actually access this.

We built it in a way that's ADA compliant. It actually works with voice and natural language and has prompts to actually tell you the types of things in over a hundred languages. Out of the box. And our partners at first were like, oh, but I want the chatbot. And it's important, we believe it's important, to educate on this is the next generation.

If you ask for the chatbot, we can figure out how to resuscitate Clippy, but we're past that era.

But this is why it was a code read at Google when they saw these large language models, because they were like, hey wait a minute, this is going to give people the answer. What I used to do is type something in, then I'd click on a couple different links, then I'd read the article, and now I just ask it a question like this.

Hey, I'm, looking for somebody for, a colonoscopy in my region that's going to take my insurance. And essentially it just comes back. It comes back with pictures of the physicians that I could potentially go see with.

Ideally, if they've actually opened up their schedule. The ability to actually click, yeah, click, and then you authenticate.

To make sure that we know who you are in a safe and secure way.

other part of that... So you're doing OAuth to, right now I assume,

the EHR providers? Yes. So we're, we actually support a whole bunch of formats from SAML to, But it's actually with EHR providers, but also sometimes they might have your own identity providers if you're actually moving in that direction.

what we've actually seen often... About 40 to 45 percent of all of health systems patients don't have a portal account. And so we've actually built algorithms to help patients match using personally identifiable information such as your last name. Because you came from FinTech. That's exactly right.

That's exactly right. Way more of your time. I appreciate you being so gracious. Thank you for your time.

Bill, thank you for joining us.

  📍 Another great interview. I want to thank everybody who spent time with us at the conference. I love hearing from people on the front lines. It is phenomenal that you shared your wisdom and experience with the community and we greatly appreciate it. We also want to thank our channel sponsors who are investing in our mission to develop the next generation of health leaders.

They are CDW, Rubrik, 📍 Sectra, and Trellix. Thanks for listening. That's all for now.

Contributors

Want to tune in on your favorite listening platform? Don't forget to subscribe!

Thank You to Our Show Sponsors

Our Shows

Keynote - This Week HealthSolution Showcase This Week Health
Newsday - This Week HealthToday in Health IT - This Week Health

Related Content

1 2 3 242
Transform Healthcare - One Connection at a Time

© Copyright 2023 Health Lyrics All rights reserved