Calum Yacoubian, M.D., director of NLP health care strategy at IQVIA, joins us for the latest episode of RISE Radio, our podcast series that focuses on issues that impact policies, regulations, and challenges faced by health care professionals responsible for quality and revenue, Medicare member acquisition and experience, and/or social determinants of health.
In this 21-minute podcast, Dr. Yacoubian discusses natural language processing (NLP) and why it’s vital to use in health care for risk adjustment, quality, social determinants of health, and health equity.
About Calum Yacoubian, director of NLP health care strategy, IQVIA
Dr. Calum Yacoubian is director of health care product & strategy at linguamatics, IQVIA. He is a medical doctor who trained and practiced in the UK before moving into medical technology. He has worked with NLP in clinical data for seven years and is passionate about the potential for data to drive improved patient outcomes.
Dr. Yacoubian’s work has focused on the use of automatic deep phenotyping from the EHR to support a variety of use cases, from population health to rare disease diagnosis. He works closely with users in leading academic medical centers–and other areas such as diagnostic labs, to ensure IQVIA’s NLP toolkit remains at the cutting edge of augmented intelligence.
About IQVIA
IQVIA is a leading global provider of advanced analytics, technology solutions, and clinical research services to the life sciences industry. IQVIA NLP unlocks and transforms data, at scale, from bench to bedside. Its leading natural language processing platform is used by 19 of the top 20 pharma, as well as health care organizations and government, to uncover previously hard to reach information or replace manual extraction.
Calum Yacoubian, M.D., director of NLP health care strategy at IQVIA, joins us for the latest episode of RISE Radio, our podcast series that focuses on issues that impact policies, regulations, and challenges faced by health care professionals responsible for quality and revenue, Medicare member acquisition and experience, and/or social determinants of health.
In this 21-minute podcast, Dr. Yacoubian discusses natural language processing (NLP) and why it’s vital to use in health care for risk adjustment, quality, social determinants of health, and health equity.
About Calum Yacoubian, director of NLP health care strategy, IQVIA
Dr. Calum Yacoubian is director of health care product & strategy at linguamatics, IQVIA. He is a medical doctor who trained and practiced in the UK before moving into medical technology. He has worked with NLP in clinical data for seven years and is passionate about the potential for data to drive improved patient outcomes.
Dr. Yacoubian’s work has focused on the use of automatic deep phenotyping from the EHR to support a variety of use cases, from population health to rare disease diagnosis. He works closely with users in leading academic medical centers–and other areas such as diagnostic labs, to ensure IQVIA’s NLP toolkit remains at the cutting edge of augmented intelligence.
About IQVIA
IQVIA is a leading global provider of advanced analytics, technology solutions, and clinical research services to the life sciences industry. IQVIA NLP unlocks and transforms data, at scale, from bench to bedside. Its leading natural language processing platform is used by 19 of the top 20 pharma, as well as health care organizations and government, to uncover previously hard to reach information or replace manual extraction.
Hello and welcome to the latest episode of RISE Radio. I'm your host, Ilene MacDonald, the editorial director at RISE, and today we'll be discussing how you can use natural language processing, or NLP, to empower risk, quality, and health equity in payer organizations. And joining me today is Dr. Calum Yacoubian. He's the director of NLP Health Care Strategy at IQVIA, a leading global provider of advanced analytics technology solutions and clinical research services for the life science industry, and IQVIA's leading NLP platform uncovers previously hard to reach information or replace manual extraction. And, Calum, I know you're going to be helping me understand what all that means. Thank you for joining me today.
Calum Yacoubian :Yeah, no, no problem. Thanks for having me. Great to be here.
Ilene MacDonald:Maybe we should just start with the basics. Can you explain what NLP is and how it all fits in with the health care journey?
Calum Yacoubian :Yeah, of course. So NLP stands for natural language processing and it's a form of artificial intelligence that enables you to take human language so such that would exist in many different forms, like what we, how we speak, or what's in the Wall Street Journal, but I guess very much in the health care space, what's written or dictated by doctors in medical records and converts that data into kind of a normalized, structured form, so removing the kind of all of the complexities of the unstructured data and translating it into something structured. So, and it's very simple, since it's taking messy, complex free text data and converting it into structured data. There are lots of different applications and areas that it's applied in everyday life and lots of new applications and with kind of generative AI. But yeah, at its very essence it's about getting value from free text and it has existed in many forms since the 1950s, but lots of exciting things happening recently.
Ilene MacDonald:So can we talk about that,
Calum Yacoubian :Yeah, so I mean, I guess, like I say, natural language processing itself is actually a form of AI.
Calum Yacoubian :You're trying to replicate human understanding by extracting information from text. So that's been around, as I say, in various different ways since the 50s. There've been a lot of developments in the last kind of 20 years about in how we can do that, the different techniques that we can use and their kind of accuracy through things like neural networks and deep learning, to kind of, more recently, large language models specifically and you know obviously a lot of the buzz that has existed around some of the general purpose large language models, I think everyone's heard of or had a go playing with, Chat GPT. Chat GPT is using different types of, or applying different types of, natural language processing in order to kind of take human input to understand what we're asking and then to mine the internet essentially to provide us with human, sensible, seemingly humanly sensible answers, and so that's created a huge buzz across all industries and, of course, people are quite excited about the potential of it in health care.
Ilene MacDonald:Why is it so important do you think in health care today for the purposes for many of the listeners today who are involved in risk adjustment, quality, social determinants of health?
Calum Yacoubian :Yeah, yeah. So I think the main, the biggest reason why it's really important in health care and it's a reason that has existed for a lot longer than this recent buzz is that 80 percent you know of member data, patient data that health care organizations generate, that health care insurers receive, is unstructured and, despite the efforts of EMR to capture everything in a structured format for the importance of the revenue cycle and for billing mean, that's been the big driver of digitization. The information that clinicians extract from their patients, either through the history or through the investigations that they order for them, is always captured in free tech narrative form. There's so much more to differentiate between patients and the granularity that exists in ICD-10 codes, and so so much of that information that we are processing as health care plans, as health care providers, is unstructured. So having a means by which to aid in the kind of interpretation and use of that data is really important, and you know the kind of fundamental technology that helps us do that is natural language processing.
Ilene MacDonald:And it sounds now like it's really a need to have.
Calum Yacoubian :For sure. Yeah, I think, I think it's been a must- have technology now for a couple of years and there have been a few trends in the industry that have made that clear. And then I think, coupled with this kind of exploded interest because of generative AI, there's a broader appreciation of the importance and value of this data and of the technologies that can be applied to it.
Calum Yacoubian :So a couple of really interesting pieces I guess Micky Tripathi, the national coordinator for the ONC, when he was being interviewed about a year and a half ago around what was happening with the Cures Act and the Patient Access Rule and information blocking, so you know, this idea of full health care data interoperability in October and what kind of, as of last year, electronic health information has included all of that free text data is not just administrative data, it's all of that free text medical record data, with a very few exceptions around some sensitive conditions. And then in his press release, he said that health care organizations must turn to artificial intelligence to machine learning and to natural language processing to get the value from that data and to confer that value back to their members by you know, they must kind of enable themselves to analyze it properly, to release those insights, to then kind of impact on their impact on the patients and the members you know that they're caring for.
Ilene MacDonald:How does this all fit in? You mentioned some of the rules that have been introduced, but how does it fit into compliance? So, if I'm an organization that maybe has never thought about this, how does it fit in and why should they also be considering it, not just for the good it can do, but with the regulations that require it?
Calum Yacoubian :Yeah, absolutely. I think for health plans there are some very well known applications and use cases for NLP and then there are some less well known and there are some areas of kind of active exploration. So they're really well known areas, and the most well known area is in risk adjustment. The use of NLP, particularly in Medicare Advantage, risk adjustment is probably the single biggest use of the technology and a payer in the payer kind of ecosystem and i t makes sense because the payments are tied to the accurate coding and demonstration of the comorbidity of the population that you're covering and therefore a technology that can help you give as more, as accurate as a picture of that comorbidity is really important. And NLP does that. It will mine the, the millions of medical records for your, say, your hundred thousand or, depending on the size of the health plan, your Medicare Advantage population and present all of the potential diagnoses to your risk adjustment team. So it's very much human in the loop of mental intelligence, AI doing the heavy lifting, letting subject matter expertise do the coding. What we've seen most of the time is that that's been used kind of in the first pass coding, but not really by the compliance teams and the compliance teams will be doing their spot checks, but they are not using the software vendors who are responsible for the first pass coding. I think that's going to change significantly because of, as you mentioned, a couple of changes in the market. So for any risk adjustment, there's kind of a I say a couple of changes. That's kind of a three- pronged environment that we're in just now, the way that CMS is going to audit plans, which is called the risk adjustment and data validation audit, RADV, is changing or has changed, and for many payment years from 2018 onwards, they can extrapolate erroneous coding, up coding as it's called, for higher financial penalties. So if organizations are getting this wrong and are over coding, then the financial penalties are potentially pretty huge and obviously the bigger the organization you are, the more risk you have, the more risk you carry. So that's the first problem. The kind of relates prong to that is that if you look back at the last where I recently read back to the last eight OIG reports for recent audits, all of them reported over coding.
Calum Yacoubian :So the trend is that when you're audited, you're not found to be under coding. I mean that doesn't necessarily mean that coders themselves for organizations are deliberately doing things incorrectly. But obviously, if you don't have a technology, it allows you to prove an audit trail of why you accept and reject every single kind of condition that's been coded, it's quite hard to stand up when the audit comes around. So preparing for audit with technology to help you see, you know, Mrs. Smith, was coded as complicated diabetes and click the button, this is exactly why, and here are the medications, the supporting evidence, all within a software platform, you're much better. Compliance teams are going to have much more confidence that they're going to pass those audits and that actually they're doing things right and they can remove any conditions that are not substantiated.
Calum Yacoubian :So we have that as the first two problems. We have the the RADV audit change also, just this kind of picture of lots of over coding being found in, you know, lots of litigation, if you can follow any of the press, it seems every week there's a new, a new case coming up. And then the third, and potentially the biggest, is that the model for risk adjustment is changing. We're going from Version 24 to Version 28. It's the biggest change in the risk adjustment model that there really has been for MA.
Calum Yacoubian :That means that 2,000 ICD- 10 codes that were previously risk adjustable are going away. They're not going to be risk adjustable anymore. So if organizations are just relying on how they coded last year and accepting that was right, they are automatically at risk of losing appropriate reimbursement for patients. Patients aren't going to stop having the diseases. You just have to make sure that they're being accurately and appropriately coded. These three things are all representative of this climate and this move from CMS recognizing that there have been overpayments and it wants to claw back, you know in the region of sort of nine, 10 billion dollars over the next five to 10 years of overpayments. So yeah, I think that we're going to see these technologies move from just supporting coding and getting it and coding in the first instance to compliance teams, because audit resilience and making sure you're doing things correctly is going to become super important, particularly with the kind of RADV extrapolation ramifications.
Ilene MacDonald:Well, it sounds crucial, you have to. W hat are the common challenges or obstacles that organizations are facing? So if I haven't used an NLP yet, or maybe not to its full extent that it could be, what are some of the obstacles that organizations might face and do you have any tips on how they can overcome those?
Calum Yacoubian :Yeah. So if you have a very well defined use case like risk adjustment, typically the obstacles that you're going to have are that, if you've never done it before, you may not have your data, your member data, ready or in a format that can be pushed to an NLP engine. So thinking about, as you're going, if you're deciding to do this, thinking about, like you know, do I actually have the data or do I have to partner with an organization to support in that kind of EMR extraction. A nd there are many organizations that do that and that partner with organizations like ourselves and to kind of help not just do the risk adjustment, coding, and workflow management, but also getting the data. Now, if you have the data through partnership, for instance with your HIE, or you already have the data, that's less of a concern. But you obviously if you want to use NLP, you have to have an idea which members and which data you're going to push through it.
Calum Yacoubian :I think for certain teams, particularly if we're looking at kind of users like compliance, who may not have used this before, one of the barriers is many kind of tech vendors offer their services kind of completely in the cloud, with a kind of payer.
Calum Yacoubian :Cost can be based on just how many records you process through and obviously if the record had no risk adjustable information in it, that's potentially cost for something that didn't have any value.
Calum Yacoubian :So I think, like navigating the kind of cost and the pricing models can be something that's important to consider. Actually oftentimes having a solution that you could deploy within your own infrastructure to give you that flexibility to run your audit program more frequently and more within your command. A lot of organizations are looking more for that to which they can use in house, and that's also something that can be challenging, that not all vendors will offer. But I think you know for use case like risk adjustment you're pretty, you know it's a very well established use case. You know you just want to consume the output of NLP. In fact to the end user, if it's AI or any other way, if it's presenting with useful, accurate information, you don't really care how it's done. If you want to look at kind of using NLP in other areas, then there are kind of different approaches that you can take because, as I say, there's a lot more value to health care data for payers than just the specific use cases like risk adjustment.
Ilene MacDonald:I wonder if we could talk a little bit about one of those use cases might be for health equity, as that becomes much more important, it seems, not only through CMS but NC QA, and wonder if you can touch on that a bit.
Calum Yacoubian :So health equity is something that we've been doing a lot of work recently, on the provider side, and it's becoming very increasingly obvious that social determinants of health are hugely important. They drive a lot more kind of health outcomes than we probably previously thought. We're seeing, I think, a lot more investment in health equity initiatives to try and kee p patients out of hospital, keep them more well, less sick for longer, and, surprisingly to many, but I guess not to us, because we work in this unstructured member data day and day out, there's actually a lot of patient level, member level, social determinants of health documented by clinicians, by nurses, by doctors, when patients kind of come through the four walls of hospital. When you kind of state it seems obvious, but a lot of us don't typically think of unstructured data as a place to find insights, you know. So a patient comes in, be very common for the doctor to write something along the lines of, came in by bus and you know as doesn't have a car, or is being discharged home where he lives alone, or has recently tried to quit smoking but still smokes 10 a day.
Calum Yacoubian :All of these bits of information are typically captured in a social history, not by every clinician but by many, and so NLP, just as it can be used for coding, for risk adjustment, it can be used for identifying these types of information like we ourselves have a particular algorithm tuned for that that we make available to our customers that has over 14, I think it's 14 different domains of SDoH, kind of over 60 different models, like transportation issue, financial insecurity, joblessness, homelessness, etc. One of our provider clients actually is using really effects by just now. And we recently won an award for best AI solution in health care, which was great. Actually we, I think we announced in the last RISE quarterly newsletter, so but yeah, great to see social determinants being actioned in the health care setting.
Ilene MacDonald:Is there any other advice or final thoughts that you may have, things that I didn't ask that you think is important for listeners to take away?
Calum Yacoubian :Yeah, I think when we talk about natural language processing and about the use cases, there are many different types of persona or stakeholder in a health plan for whom it might be interesting. So there are the risk adjustment managers and risk adjustment coders, very kind of use case specific use. The value is pretty obvious there. There's the compliance teams within the risk organization. We talked about SDoH.
Calum Yacoubian :Obviously that spills into quality, increasingly with for either the hybrid measures where we're doing the kind of the annual chart chase, or for the electronic measures, there's an increasing burden on being able to identify relevant patients in both the nominator and numerator for these different kind of quality measures.
Calum Yacoubian :So we often see NLP being applied there. But increasingly health plans within a kind of data science kind of set up are looking to identify patients at risk of disease progression, at risk of kind of readmission. All of these things which are quite important to health plans were typically they're doing modeling based on the structured data they have. NLP can be applied to any of those use cases and a real strength that we've had over the years working with payers and providers is that our kind of approach is that whilst we have these very specific use cases, what we're expert in and what our technology is expert in is translating free text medical records to structured data for pretty much kind of any endpoint or any condition or any factor that is of interest to the health plan. So when you think about this capability, it can be used far beyond the kind of use cases where you might be thinking of it traditionally like risk.
Ilene MacDonald:Thank you so much, and people can reach you at IQVIA if they have any questions further.
Calum Yacoubian :Yeah, so I think NLP@iqvia. com ultimately comes my way. So that's a great way to get in touch. Or I'm always happy to connect with people on LinkedIn. There are not many Calum Yacoubians, so you should hopefully be able to find me. And yeah, always happy to discuss and very passionate about this as an area of health care. So always happy to learn what other people are doing and to hear their thoughts. And definitely, rest assured, not every conversation needs to be a sales engagement, just really enjoy talking about this topic and sharing thoughts with others.
Ilene MacDonald:Thank you so much.
Calum Yacoubian :Thank you so much.