Building a Rare Disease Franchise with Machine Learning and Artificial Intelligence

Nitin Choudhary, Managing Principal of IPM.ai and Richard Wilson, Rare Disease Franchise Head for Kyowa Kirin recently spoke on overcoming patient-finding challenges through machine learning (ML) and artificial intelligence (AI) during a webcast hosted by Pharmaceutical Executive and moderated by John Seaner, Chief Marketing Officer of IPM.ai. Charged with building a global rare disease franchise for Japanese-based specialty pharmaceutical company Kyowa Kirin, Richard partnered with IPM.ai to uncover undiagnosed patients, link their treating HCPs and map the diagnostic and treatment journeys of pediatric, adolescent and adult patients. The goal of the partnership was to establish Kyowa Kirin’s Poteligeo as the therapy of choice to target CC chemokine receptor 4 (CCR4), which is frequently expressed on leukemic cells of certain hematologic malignancies, including cutaneous T-cell lymphomas (CTCL). The condition is estimated to affect only 12,000 patients worldwide, and IPM.ai immediately uncovered 350 patients that went on to be treated, with another 69 currently being evaluated as candidates for treatment.

Read on for the full interview between Nitin Choudhary, Managing Principal of IPM.ai, Richard Wilson, Rare Disease Franchise Head of Kyowa Kirin and John Seaner, Chief Marketing Officer of IPM.ai.

Richard Wilson (RW): Thank you all for joining today. I'm really pleased to be here and discuss our artificial intelligence and machine learning initiatives for patient finding. Kyowa Kirin is certainly not a well-recognized name in the US, but I want to give some background on the organization. It's a Japan-based specialty pharmaceutical company that has been around for over 70 years. Our footprint has been established in North America and, at the moment, it's a fairly small organization with just under 500 employees. Growth is fabulous, as the region is already almost 20% of global revenues. We have three locations in the US and Canada, and three first-in-class franchises. The mission of Kyowa Kirin is to bring innovation globally to the communities that are in need and make a profound impact on patients.  

Nitin Choudhary (NC): Thank you, Richard. It's a pleasure to be here with you and share some of our experience leveraging machine learning and artificial intelligence to solve commercial challenges and optimize patient outcomes. As some of you are aware of IPM.ai and others may not have heard of us, I’m going to take you through who we are and what we do. 

IPM.ai transforms real world data into real world insights that uncover undiagnosed and misdiagnosed patients as well as the key, hidden moments in their treatment journey. For example, when we are looking to find patients that are progressing through a line of therapy, an obvious question is “who are the patients who are ready to switch therapies? At the core, the ultimate objective for our clients is to accelerate their life saving therapies and create better patient outcomes quicker and with less risk.

We work with our clients throughout the product life cycle, whether starting from the inception of drug development, clinical study, product launch or commercial strategies. In the development stage, questions arise, such as “what does the market landscape look like?” or “what is the size of the patient population?” If you're running a clinical study, we can help you find trial sites and better understand the ideal patient, so that you can recruit and enroll them. As you approach product launch, there are always questions regarding the physicians caring for these patients. Through commercialization, as patients progress through their diagnostic and treatment journey, we continuously mine the data, leverage the business intelligence that we have built and drive the next best action, whether through a personal or a non-personal engagement channel.

If you look at IPM.ai, we fundamentally start with the ideal patient and their treatment journey, their complex referral networks and their courses of treatment. The key objective behind all of this is “how do we engage these patients?” and “how do we activate disease intercept?” By better articulating the key moments in the patient journey and by profiling the providers they are engaging for care, we help our clients make informed decisions regarding the right messages and what kind of education they need to provide to these patients and their physicians. 

At its core, there are three foundational data assets that we use. The first is health data, which consists of 300 million unique de-identified patients going back a decade and updated weekly. That means every week, we get new information and new transactions for the patients who are either already in our data pool or who are just entering the health system. 

The second pillar is integrated affiliation data, which becomes essential in rare disease and specialty oncology to understand where patients are receiving their care, because that care can be very different if a patient is in an academic setting versus in a community hospital. With this data asset, we help paint a better picture of where these patients seek care. For instance, is it in a hospital? Or is it in a group practice? How do the hospitals and group practices then roll up into the IDNs and the health systems? 

Last but not the least, we also have social determinants of health, which are psychographic, demographic and lifestyle attributes that can help us better understand what these patients look like as consumers and the socio-economic burdens that some of them may have. In the end, the goal is for our clients to be prepared to address these challenges. 

The way we are creating value is a three-step process. The first step in the process is uncovering these otherwise hard-to-find patients or hard-to-find moments in a patient journey. By hard-to-find patient, I mean the patients that are otherwise misdiagnosed and undiagnosed in the healthcare system and are bouncing from one specialist to another without receiving a proper diagnosis. The hard-to-find moments are times when patients could really benefit from treatment; to figure out the right time to intervene with healthcare providers so that patients receive the most benefit from therapies. We are helping uncover these hard-to-find moments and activating that intelligence in a timely fashion. We don't want to reach out to a physician when the patient has already started the therapy or when the diagnosis is already made. Where patients receive the most benefit is if we are further upstream because that's when we can drive the outcomes quicker. Ultimately, what we are doing is looking at the intersection of the patient, physician and the provider and trying to understand how best we can provide connected guidance to our clients.

John Seaner (JS): Richard, let's begin with you. You joined Kyowa Kirin in 2020. You're tasked with building a rare disease franchise. What were some of the primary challenges you faced, including the roadblocks, the bottlenecks, the things you needed to address right away?

RW: I think one of the primary challenges was having no infrastructure for a rare disease business. Every rare disease operates differently, but I wanted to determine where the patients and the physicians that are treating or diagnosing these patients were. Where are they located, because that's where you're going to put your resources and your bodies. We treat super rare diseases. The prevalence stats are all over the place in the US — it’s anywhere between one and 20,000 and 100,000. We take for our purposes, one in 25,000. We look at roughly 12,000 patients and figure out where they are. There's no ICD-10 code and while there are claims data out there you could still be missing patients. Obviously, that’s sort of the long-term challenge but there was also an immediate need because of COVID you've got fewer boots on the ground that can do this investigating. This concept of discovering patients through other means — artificial intelligence and machine learning — is certainly heightened. I brought both Crysvita and Poteligeo into the US to see if we could increase the business or make the business a little more productive, especially in these times. With IPM.ai you get really deep into the patient journey. We can shorten the journey for these patients that are basically left alone, that are struggling. I've worked in rare diseases where the time to diagnosis is 10 to 12 years, if you knock that down to three years, it's fantastic. 

JS: Now you’ve said it — machine learning and artificial intelligence. We've heard all the buzzwords, the new categories that are being created and the promise of all this. Bring IPM.ai into the equation. You're building this franchise and you clearly need to understand the patient diagnostic and treatment journey. What did IPM.ai do to overcome some of these challenges that you had with a “needle in a haystack” problem of only 12,000 or so patients?

RW: One thing I've always found impressive with IPM.ai, and why I enjoy working with you, is the core competency. It's all about patient-finding. There's a lot of companies out there that do this and do a great job, but they do many other things as well. IPM.ai is 100% dedicated to patient-finding. And as you go through a project, it always leads to something else. We can look at the journeys and segment the patients. I like how IPM.ai engages evolving challenges and solves the problem with the client.  

JS: Certainly, we've seen the promises of the marketed solutions. I think everybody thought that there was a Magic Eight Ball that was going to uncover everything and give you the answer. I think what you're emphasizing is augmenting intelligence; that there is still a human aspect to interpret, engage and activate these insights. Can you touch on that? 

RW: It was very interesting because in the US there was no rare disease knowledge and it was hard to track down the right patients. Bringing the IPM.ai concept to the organization at first was like, “Well, hold on, how does that work?” It sounds fantastic, but it's not going to be a magic bullet. You have to present this as, “it's not going to be the entire answer, and everything is solved and we're done.” It’s going to guide you better and you're going to be more targeted. 

JS: Speaking of sharing information, Nitin, how does the IPM.ai solution work? Let's talk about some of the advantages and how we do all this in a privacy-safe manner.

NC: The fundamental aspect is our people. While machine learning and AI are certainly key to our solution, we built a team around it. We know the secret sauce as to how to articulate the problem, feed it to the ML and AI engines and get the answer, which we can then analyze and relate back to our clients. Analysis of machine learning output is critical and essential, and that's where the people aspect comes into play. Ultimately, we bring together people, process, data and technology to help our clients get to these patients faster, determine how these patients receive a diagnosis, get on the therapies and optimize their outcomes.

The other piece you mentioned is privacy. Privacy is essential and it’s a part of everything we do. Our people, our processes and our tools are privacy-safe so that our clients don’t have to worry about that.

JS: Along those lines, talk to me about the role of first-party data and how it can be incorporated into our data universe. 

NC: First-party data is essential. What we have built is an ability to ingest any data set in a privacy-safe environment. Sometimes it's hard to find a starting position when dealing with rare disease patients. You can look into the data set and work with the clients, medical and clinical team, but it may still not be sufficient. If you are working in a disease where there is no ICD 9 or 10 code and there are no established therapies, it’s hard to have a starting point. What we do in this case is leverage first-party data inputs. It could be a genetic testing panel as an example. We can ingest that and then, in a privacy-safe environment, link those patients to our data universe. This leads to the prototypical ideal patient that our clients want to help. We then go in and use that starting point to apply machine learning and AI around it to help us find other patients.

JS: Richard, back to you, Let's touch on this data privacy issue as well. Certainly, it’s one of the questions you were asked internally when adopting this. In addition, what were some of the other internal hurdles you faced? 

RW: Privacy is critically important. I'm sure in any organization you're going to have those same questions from legal and compliance. It’s very important to bring IPM.ai into those discussions and see exactly how it's done. I introduced this concept of how to find patients in Tokyo or Korea globally and there was a lot of interest across regions.  

Obviously, every region is very different. You cannot do the things that you're doing in the US everywhere, and we tackled a lot of questions. I think the other hurdle is that all companies do this. Your data purchasing on an annual basis is huge in this business. We don't need more data; we just need to be better with the data we've got now. I approached this as, let's give this a try and see what happens because you may find going forward, that some of that data can actually be discarded going forward. I've found that to be true in the past. You’ve got to be engaged with the field, you have to listen, as this type of concept will continually evolve — and I hope it does because the end-result is to shorten the diagnostic journey for these patients.

JS: We seem to be in a data arms race where more is better. But folks really don't understand why. I think what ends up happening is they start sacrificing the present for the perfect, they're not experimenting, they're not getting started. It ends up being a discussion about the data and more of it instead of what they can do now. What are the typical hurdles that organizations have when we start one of these engagements? How can they best prepare for collaborating with IPM.ai?

NC: If we look at what happened within the last decade, the whole healthcare focus shifted towards more specialty oncology and rare disease. These are the areas where we have a huge unmet need. The traditional approaches don't work. What started happening is the patient-level data started becoming more mainstream. However, these data sets are not perfect. We will never live in a world where we will have a highly connected data set that is 100% accurate for 100% of the patients. I think that's euphoria. It's not going to happen. When we started this business a couple of years back, machine learning was our focus. We knew that machine learning could investigate and analyze the data in all sorts of different ways that are humanly not possible. Supervised machine learning is becoming more mainstream in the industry. We get the benefit of this in everyday life. Think about fraud detection as an example. That’s been done using ML and AI. 

What we have done is take the same technology and use that to help us find these patients and uncover these key moments in the journeys which are otherwise hard to find, using the data that we have and knowing that it is not perfect. We can create pattern recognition algorithms to help find these needles in a haystack. I think the industry is getting there, but people venture into this with a little bit of skepticism and are unsure of what they will get as an output. My assurance to those clients is to look at the level of success we have had with our clients. We've been able to use these technologies and drive better patient outcomes. For prospects that are still not fully on board, what we can do is show them the evidence and examples of what we have done and where it has worked.

JS: Let's talk about the noisiness of data and how we eliminate it. 

NC: There is noise in every data set, but I think that's the benefit of working with an organization like IPM.ai. We minimize the noise by leveraging control groups more efficiently. I know who I want to find, but at the same time, I can compare and contrast those journeys across every single journey that exists in my data set across 300 million patients and build an algorithm that can truly find these otherwise hard-to-find patients. 

Machine learning is a process that gets more mature the more you do it. Our sister organization, Swoop, has been leveraging ML and AI to create custom DTC audiences for cross-channel marketing purposes, uncovering consumers long before we started training machine learning models on in-house data. So we are a step ahead of the market because we've been able to experiment, test and scrutinize those algorithms. There is a lot of investment in the tools and technologies which helps us get better at reducing that noise.  

RW: I think an important point here is that it's a dynamic process. The algorithm will get better and better. You can feed and test the machine. As you get patients on your therapy or you've diagnosed the patient, the machine will pick it up so it's always getting better and better at evolving. 

 JS:  Let's talk about outcomes. Richard, can you let us in on some of the results that you were tracking, and what was unexpected?

RW: Across regions it has been phenomenal. We saw the specialties treating this and it answered some questions, as these patients see 4 or 5, 6, 7 different HCPs and we weren’t sure where to focus our efforts. The most immediate bump came in the Poteligeo business. There are about 350 to 400 patients at any one time on the therapy and it's grown in the last six months.  I just followed up leads for the franchise the other day and they have 69 new patients; that’s significant. The data is refreshed on a weekly or biweekly basis so you can get out there immediately, in front of the HCPs. That's been really good work and hats off to the IPM.ai team. It's a true partnership. It's about finding a problem and gap and getting to the best solution.

JS: Can you tell us a little bit about how IPM.ai works both on the clinical study side and assessing a market for the development of a therapy when you don't even know how many patients are candidates?

NC: The fundamental problems don't change but the application does. From a clinical trial perspective, many fail because they cannot meet their objective of finding participants for the trials. We know that that's a fundamental challenge, and more so in rare diseases than in some of the more prevalent conditions. Let's imagine a pharma manufacturer who is running a phase II clinical trial, they have their sites up and running, and they are falling behind in patient recruitment. In this particular case, not only can we help them find the patients, but also link their providers so they can advocate enrollment in the study.

In terms of understanding a market for a rare disease, the challenge in most cases is there is some broad understanding or literature that's out there, but the estimates are too high-level. We have helped clients size the market through primary search with the physician or KOL and this helps our clients with funding. 

JS: We talk about patient-finding and understanding their diagnostic and treatment journey, and you just started to touch on something that I think might be a little bit overlooked, which is the healthcare provider side. Tell me a little bit about the healthcare provider equation. Certainly, we understand specialty designations, but what happens when a primary care physician is actually serving as the specialist? How does IPM.ai handle this?

NC: The specificity is just a designation. The provider just has to register with the state based on license number. They have to mark their specialty, but that could be outdated. HCPs could have also gained new specializations, especially when they may be the only provider for 50 miles. We focus on how we can uncover the specialty of any provider based on what they are doing, who they are treating — that's the fundamental difference in how we are approaching it — instead of taking specialty designation at face-value. How an HCP actually behaves is apparent in the data. For example, a PCP can behave more like an oncologist, a psychologist or an endocrinologist. What our solution can do is take a provider, look at them through a comprehensive 360-degree view, and then put it in a bucket to say this provider looks more like an oncologist or a neurologist. We can then score the provider based on their resemblance based on the data to a designated specialty.

JS: Richard, we certainly engaged you post-launch. Tell me, is there any plan to take this to the clinical trial or market assessment side within the organization?

RW: I think it's a great concept. I know it's already happening in oncology and rare disease, especially for clinical trial recruitment. You get the right patient in the right study. The ends can be smaller, study duration can be shorter — it makes sense for everybody.

On the commercial side, it's the finding as well as the journey that converts into messaging. We definitely see it with an XLH. These patients will see several different physicians, so the next question becomes, who's going to be the most likely to manage that patient and that look-alike may not even be a patient. That's where you need the influence networks across those decision types. It's very disparate for XLH. These patients are seeing primaries, orthopedic surgeons, rheumatologists, nephrologists, endocrinologists, and most of the business is coming from endocrinology. You're managing those patients, but I could see down the road to expand the therapy by focusing on primary care and rheumatology.

JS: How long does this take?  What time periods are we talking about for the audience to try this out and benefit from it?

NC: Most problems can be typically addressed within eight weeks after kickoff and discovery.

JS: Is there anything that either of you would like to add or you think that we may have missed during the discussion?

NC: I would encourage the audience to reach out to see if we can help. If you have a problem statement, give us a call, I'm happy to set up one-on-one time and walk you through what we can do and the possibilities. 

RW: I agree, a hundred percent. IPM.ai has been super collaborative. It's the solution. 

 

 

 



Related Posts