Exploring the Role of Clinical-Grade Augmented Intelligence in Enhancing Health Care Data Management

Read the full transcript:  

Welcome back to PopHealth Perspectives, a conversation with the Population Health Learning Network where we combine expert commentary and exclusive insight into key issues in population health management and more. 

In this episode, Colin Banas, Chief Medical Officer for DrFirst, discusses the challenges in health care data management, the importance of semantic interoperability, and how clinical-grade AI can improve medication safety and patient outcomes. 

To start out, really easy, can you please introduce yourself? Just your name, your title, your organization, and your professional history. 

Yeah. Colin Banas, the Chief Medical Officer for DrFirst. I joined the company about 5 years ago now. But just by way of history, I'm in an internal medicine position by training. I was a practicing hospitalist and the Chief Medical Information Officer for Virginia Commonwealth University health system, which is a large academic health system in Richmond, Virginia. I was there for 17 years practicing medicine and informatics before joining the team at DrFirst.  

Can you elaborate on the specific challenges in health care data management with clinical-grade AI aiming to address these issues and then how its implementation can help fill data gaps specifically for clinicians? 

Yeah, it’s the billion-dollar question for the past 2 years. It seems like, ever since generative AI hit the scene, that's all we've really been talking about. But what's interesting is that some of us have been up to this for quite a while. You know, AI is not necessarily brand new, but it's definitely making some big leaps and bounds and splashes as of late. And it's really accelerating so I think it's deserved it. But always to know where you're going, you have to remember where you're from. And you know we've been at this for a while now, in terms of the digitization of health records and the application of technology at the point of patient care. 

You know, what's interesting is my career was fortunate enough that I've actually seen it all transpire. So when I was originally training and originally in residency, we were on paper. Paper charts, paper prescriptions. If you wanted to know what a consultant said, you had to climb up to the tenth floor to go pull the red chart to see if they've come by yet, it was crazy. And pagers and so forth. And then within a few years of that, we went hybrid. We had an electronic medical record system. A lot of it was optional, but we would use it for order entry and eventually for prescription writing, but a lot of the charting was still analog, it was still on paper.  

And then when I took over in a leadership role for informatics, I was fortunate enough to help lead a lot of the transitions to say, okay, we're really going to do this, we're going to go fully digital. And so I bring that up because I remember, I have this sort of longitudinal view of how things used to be. And there was a period of time where getting data was nearly impossible. It was really, really difficult. I am not exaggerating when I would say I would admit a patient to the cardiology intensive care unit and I would need to basically break into the records room of where the echocardiograms were stored on paper so that I would have that information to take care of the patient, and also to be prepared for the next morning when we were rounding. 

With the advent of interoperability we're getting better. Some people might even say there are moments in patient care where there's now that the clinicians have too much data. There's too much stuff that they have access to, or too much stuff that's being thrown at them. And so, we sort of went from one extreme to the other in some cases.  

The interesting thing is that despite decades of standardization and the digital explosion of digital technology at the point of patient care, we've gotten good at moving data from point A to point B, but it's not necessarily good data. It's not necessarily usable data. A lot of the time I call this the semantic interoperability problem. So, I've got a whole bunch of blob text that I've been able to consume from a health information exchange or from another electronic medical record. But it's not, it doesn't land in the right spots in my record. And I can't see the longitudinal history of the patient without having to do a whole lot of what I call dumpster diving, or a whole lot of taking that data and transcribing it back into my system. And I think that is a pivotal role for AI and technology to help basically with the semantic interoperability problem.

I do want to go back and say I actually hate the term artificial intelligence. You'll hear me say augmented intelligence a lot. In no way are we looking for AI to replace people, we want them. We want AI to be that digital co-pilot. It has the massive potential to help me do my job. Better to relieve the cognitive burden of some of the, what I would say, menial tasks or administrative tasks to free people up to actually practice at top of license. And I think you could apply that to almost all professions. So yeah, I think it's a disservice to call it artificial intelligence.

Based on research, the use of this clinical-grade AI (or augmented intelligence) may reduce the risk of adverse drug events and readmissions. So, could you provide any examples or case studies where AI has been pivotal in improving patient outcomes and preventing such events? 

Yeah. So at DrFirst, when we talk about our clinical grade AI, it's in the context of medication safety. We are a medication management company, actually started as an E-prescribing company. And as you can imagine, over the course of almost 24 years now we have a lot of experience with the world of medications, the world of medication routing. We have a lot of medication data.  And one of the use cases that I'm most proud to work at in DrFirst is in medication reconciliation. So, you know, the 2 min overview is when you show up in front of a clinician they need to gather a history of the medications that you're taking or supposed to be taking, and then start to make decisions about whether to continue them, to change them, to stop them, etc. The very first part of that process is gathering that history. And I just described the semantic interoperability problem where I might have access to a lot of that data now. but it certainly isn't being presented to me, the clinician, in a way that's imminently usable. And so that's where I have to start dumpster diving. That's where I have to start re-inputting data, that is, you know, quite frankly, in the year 2023 this stuff should be, we should have solved this by now with standards, etc., and we still haven't. 

What our clinical-grade AI does is takes that glut of medication data, that medication history data, and starts to reconstruct it in the nomenclature, in the structure that the receiving system is expecting. So you know, too, there are many drug compendia out there. Maybe system A uses something like Multum and system B uses something like First Databank, and they don't always talk. They usually do, but a lot of times they don't. And when they don't talk that stuff comes over as free text, it's not usable. I can't use it for drug interaction checking, etc. 

So let's codify that. Let's make sure that at minimum we're going to get that normalized. And then let's take it the next step. The instructions for a prescription, commonly known as the “sig”.  That's the part that says, to take one tablet by mouth daily.  A lot of times that data is missing from that initial data poll that the clinicians are looking at, or that data is coming over also as free text and someone's going to have to re-input that data into their system. We're able to take that with our clinical-grade AI and do 2 things. One, when there's free text there we can translate it. So, it's a combination of natural language processing some proprietary, you know, almost secret sauce. And we'll turn that free text back into structure. And, more importantly, the structure that your system is expecting. You know, system A might say  “take it by mouth” but System B says “take it orally”. You know it seems it seems silly, but these are the things that actually are not always talking to one another.  

The other thing that AI will do is when there are pieces of the medication data missing, when it is clinically safe to do so, we will fill in those gaps we will. We will basically infer the missing pieces of data to put back into the string, so that when the clinician is reviewing it, they can say, “Yep, that looks good. I accept”. One of the most important things about what we do, and this goes back to the co-pilot analogy, none of this is happening without the oversight of a human, without the oversight of a clinician. Nothing gets committed to the database or to the record without somebody looking at it and saying, I accept that. We're just giving you a massive head start on doing the med rec by bringing this data over in a structured and codified way.

Some of the case studies, we've taken initial med review for some pharmacy techs at some pretty large health systems. You know, they used to average in the 45-minute range to do a complete medication history for a patient. Now they're doing it in 20 or less. You know, so it becomes what I call a force multiplier; I've made you more efficient, you can now see more patients. From a safety perspective, and these are the ones that really get me excited, we have some large health systems in Boston and using their own data after implementing our tools they had a 25% reduction in adverse medication events and a 25% reduction in those events potentially reaching patients. That's their own data. That's, you know, that's a pre and a post study that said, “Hey, we made those clinicians more efficient. And not only that, we made them safer”.  And that's through the leveraging of that augmented intelligence. 

In what ways does clinical-grade AI eliminate the need for manual data entry, and how does this optimization at the point of care impact staff productivity and the quality of patient care? 

It might be it might be fun to dive into why we are very intentional about using the word clinical-grade AI. It's sort of 2 sided, right. On the one hand, I just opened by saying, “Hey, AI has been around for a long time. This is nothing new”. But, on the other hand, there is so many people using the term AI, and there's not necessarily good definitions out there of what is what type of AI. And so when we use the term clinical grade we're being very intentional in that because we want to make sure that people understand the distinction between using our tool which has been around for close to a decade now and, you know, plugging in a medical record into something like Chat GPT, or Bard, or something that's, you know, widely publicly available. When we say clinical grade AI we're saying this is something that's been intentionally, narrowly focused on this use case, this medication safety use case, and it's been trained on clinical data. In fact, it's been trained on clinical data for, like I said, over a decade. And also, it's being constantly overseen by a team of clinicians. So the algorithm is processing over 25 million transactions per day. 

And then we're also looking at the output and continually tuning. Cause there's new drugs that get introduced. there's new indications, there's new ways to write for those drugs and so we're constantly accommodating for that. Whereas, you know, something like the initial version of Chat GPT, it stopped. It was like the Internet up until, you know I can't remember, the year like 2021 or 2020, and then it, you know, it stops. One of our engineers likes to use the analogy of like your orthodontist having a very specific you know drill that they're going to use for your root canal versus going down to, you know, Lowes and getting, you know, the Dewalt 12 volt. You know they're both drills, but one of them is very specifically designed for a very specific clinical purpose. 

And there's other things in the in the realm of clinical grade AI that are important such as transparency. Not all AI is perfect. We actually, our particular algorithms, they will fill in those gaps that I refer to about 90% of the time. The 10% we don't do is actually intentional. We would rather put no data in front of a clinician doing medication reconciliation than put bad data in front of a clinician. Because when you get something like a medication wrong, you get the frequency wrong, or you get the dosing wrong, or you omit one, or re-add one that's supposed to be stopped- that is precisely how patients get hurt. The blood thinner that that doesn't get restarted, or the seizure medicine that wasn't on the med list when the patient got admitted. You know I have countless stories like that from my time at an academic health system, you know, leading medication safety committees, etcetera. 

But your your question about efficiency is a good one. That is, you know, back to that copilot analogy again. that is what we're looking for. I'm looking for the computer to help me construct the right medication list in the right format in a very efficient manner so that I can do other things. I can spend more time on my interview or spend more time on my physical exam or reach out to family members rather than being huddled at the computer. 

I heard a statistic recently. For every full day and clinic, a clinician is spending 2 hours in the electronic medical record. And that's what's leading to the, you know, the burnout equation. The fact that now everybody is heard of ‘pajama time’ or this notion of doing work after hours. This is precisely the sort of thing that we should be looking to technology to relieve the burden for. So that's why I'm really bullish on augmented intelligence and specifically those clinical grade applications.  

The 3 unique efficiency improvements you know that we’ve mentioned include eliminating manual data entry, igniting adherence, and supercharging value through continuous analytics. Could you provide some more details on how these aspects enhance overall health care delivery and then, again, patient outcomes? 

We've definitely spent a lot of time talking about the medication reconciliation side of the prescription journey. But that's actually just one piece of it. If you actually trace from the point of a clinician deciding to prescribe to a patient, you know, deciding whether or not they're going to pick it up to the renewal process to, you know, getting admitted to the hospital, there's actually a sort of a loop that you can draw as it relates to the medication journey, and DrFirst is proud to have a an intervention or influence in a lot of those points in that journey. And one of the things that we haven't talked about yet is the ability to show prices. Prices to providers, prices to patients, at the moment of making those prescribing decisions. If people go to our website and look up our fusion platform it really is talking about combining a lot of these tool sets that we've been alluding to here into one platform. But the idea behind the price transparency is if I can have more data in front of me at these pivotal moments in clinical care, I can choose better. I can have you in front of me and say, “Wow! Your copay for this one is $200 and it has a prior auth. But the system has shown me that this one over here is 5 bucks, and it's got no prior auth and it's just as good”. I'm going to make that switch, or I'm going to have this conversation with you and find out you know which one do you want to take. 

I think that's another example of putting data or putting information in front of people to make these informed decisions. And then none of this is worth a hoot if you can't have the data to back it up. If you can't provide analytics that prove the impact that you're having. And so that's another part of our offerings is a robust analytics package. We're going to show you who's using the tool, who's using it correctly, who's not using the tool. And therefore, that's an opportunity to educate. You know, perhaps the general medicine nursing unit is knocking out of the park and I can tell from analytics they're accessing the medication list, and they're using the tool to import those meds. But the folks over in, you know, neurology, they're not even touching it at all so they're doing everything from, you know, brute force, or just from the patient interview. Wow! We need to go have a talk with them, or we need to have a in service education with them. And that's how you get that continuous improvement cycle that you know health care is so fond of. So it's putting more information, not just data, but information in front of people at the appropriate times and then reinforcing it with robust analytics. 

The integration of prescription price transparency, medication fill data, and automated reminders with educational resources is highlighted as a method to improve patient adherence. Can you discuss the role of AI, or these tools, in delivering this information and how it positively impacts patient behavior? 

Yeah, so back to that medication journey loop that II had everyone visualize earlier, there's also reinforcement points. After I've written the prescription and after I've seen how much it's going to cost, I need to be able to, or the system, or somehow, we need to be able to engage the patients to inform them that a prescription has gone out, and this is what that prescription is, and this is where that went, and here's some education about that prescription, and here's some price information about that prescription. And, in fact, if it's appropriate, we can line you up with the coupon or the copay assist card directly in that experience. And what I'm describing is entirely mobile.

Back to when I was writing paper prescriptions. You, the patient, actually had something physical in your hand that you would take to the pharmacy, there was this tangible reminder. You know a call to action was, I need to take this piece of paper, do something with it in order for me to, you know, get my prescription. And of course we went to e-prescribing, and that disappeared. It's sort of poof out into the ether. And people forget, or people don't have that call to action that they had anymore. And what our technology does, is in my mind, it replaces that physical piece of paper with a digital interaction on your smartphone. So you get a app like experience over SMS. We don't want you to go get a new app. We don't want you to get a new login, etc. In fact, not everybody even has a portal or a portal account. But everyone has text messaging and like, 95% of people interact with text messaging. So let's, at least for this crucial touch point, let's meet them where they are with a text and let's give them an app like experience without making them go to the app store. And in that app like experience we're going to help you overcome what I consider the barriers to getting your prescriptions filled cost. I just said we're going to show you what it costs and show you coupons if we can line you up on.  

Education is the other bucket. You either don't know why the medication is important, or you don't know the side effects, or you're scared. Well, let's educate you. Let's give you more information about that drug and why it's important while your doctor wrote for it. So we have a library of content that we also deliver there and then. The third one is the life just gets in the way. Like I forgot that my prescription is waiting for me down the street. Well, now you're going to get reminders, you're going to get nudges. And you know, of course we give people a way to opt out. If people say well hey, don’t bother me with this. You know, you’ll opt out and never be contacted until you opt back in. But its all about meeting patients where they are and giving that personalization.

A lot of this is powered by algorithm, a lot of this has opportunity for even more intense personalization as we learn more about these tools and as we have access to more patient data that is safe and appropriate. That's another pillar of clinical grade AI is patient safety, patient transparency. So all these touch points can be super-powered by technology, by augmented intelligence, to increase the efficiency, to increase the personalization, to improve the experience. That's why it's a really exciting time in the world of AI. 

Is there anything else you’d like to share? Is there anything that we have missed or that you would like to touch on more? 

It is fair to have a healthy skepticism of this technology. People are running at breakneck pace at a lot of the newer stuff. But there's also, there's also folks that have been at this for quite a while. And there is narrow use cases that are tried and true, that are proven. You know, over the course of a decade we can show that technologies such as ours in the medication safety space is a game changer and makes a big difference. And of course, we're out there looking for the next thing as well. So I think it's a really exciting time. And let's retire artificial and replace it with augmented for augmented intelligence.

Thanks for tuning in to PopHealth Perspectives. For similar content or to join our mailing list, visit populationhealthnet.com.