Kyowa Kirin Uses AI/ML to Uncover Undiagnosed Patients

By: John Seaner, CMO, IPM.ai/Swoop

The rare disease patient journey can be arduous. Many patients experience multiple misdiagnoses or go undiagnosed before finally figuring out their true health issue and receiving treatment. For Japanese-based commercial rare disease business, Kyowa Kirin, there was a need to find patients for a commercially available drug but low internal awareness of how artificial intelligence and machine learning solutions (AI/ML) could be utilized for patient finding at scale.

Richard Wilson, Vice President and US Rare Disease Franchise Head at Kyowa Kirin, Inc. originally became interested in predictive intelligence for patient finding when using IMS data to map patient journeys six years ago – while he took advantage of the yields from this endeavor as best he could, the output was understandably unsophisticated. Time passed and Wilson moved from leading Takeda’s endocrine division, which included the NATPARA^® brand for the treatment of hypoparathyroidism, a rare hormonal condition in which the body is unable to regulate calcium and phosphorus levels, to Kyowa Kirin with a focus on X linked hypophosphatemia (XLH) – a condition also marked by low levels of phosphate in the blood. Both diseases have similar prevalence rates in the US and are characterized by a fragmented patient population. 

XLH is a hereditary mutation that is easily identified through genetic testing, and most often detected by pediatricians when very young patients first become weight-bearing. For adult patients, however, symptoms can lead them to multiple specialties and result in many varied claims. Even if they have been previously diagnosed, de-identified adult patients still need to be re-discovered in the field. Based on the confusing nature of the disorder, lookalike patient data generated through AI/ML solutions are necessary to refine XLH patients. 

Richard found out about IPM.ai through word of mouth by networking in the industry. He decided to use IPM.ai instead of another provider after realizing that AI/ML for patient finding is the company’s focus and area of expertise; the organization is built around this capability, whereas many other firms work in research or analytics and claim to have AI specialization but lack a devoted program. Another reason he was attracted to IPM.ai is the company’s expansive real world data universe, which consists of 300 million de-identified patients. Despite wanting to jump on the technology, Richard found one of the biggest hurdles was familiarizing others within the organization about these capabilities. Although the team was aware of partnering with AI providers for development and in clinical trials, there was a lack of understanding on the commercial side for patient finding at scale. Ultimately, the best way to prove the concept internally was to put it to the test. 

With the help of IPM.ai, Kyowa Kirin uncovered 18,000 XLH de-identified lookalike patients. The results were so impressive that IPM.ai’s machine learning platform was used to source lookalike cutaneous T-cell lymphomas (CTCL) patients for the company’s oncology division. Returns on this front were equally exciting, with 13 out of 260 de-identified patients, or a total of 5% of the confirmed patient population, uncovered through their HCPs within mere months. 

A net-positive outcome of AI/ML based patient finding is a significantly shorter diagnostic journey, which has the potential to benefit many and save lives. On the clinical side, an abbreviated journey also means getting drugs to market faster. In any instance, delineating de-identified lookalike patients to clients so sales reps can conduct outreach to treating physicians, who can then recommend a therapy to a patient, is a transformative way to overcome one of the greatest challenges facing the pharmaceutical industry. 

Read on for the interview between John Seaner, CMO of Swoop/IPM.ai and Richard Wilson, Vice President, US Rare Disease Franchise Head at Kyowa Kirin, Inc. - U.S. 

John: Can you tell us a little bit about yourself, the company, its mission and an overview of the good work you guys are doing?

Richard: I joined Kyowa Kirin in August of last year, mid-COVID, and was tasked with building the company’s rare disease business. At the moment there are two franchises globally, oncology and neuroscience. The company’s brands include POTELIGEO^® in oncology, NOURIANZ^® in neurology, and in April 2023, will gain back the rights to Crysvita^® in the North American market. I'm putting the infrastructure in place and getting ready for the rare disease business relaunch, as we're already approved on the market. 

John: What are the primary business challenges that you overcame and how did IPM.ai fit into your vision for the company? 

Richard: That's a great question because going back to my prior role at Shire, which was obviously then acquired by Takeda, I worked in the endocrine division. This business is now part of hypoparathyroidism which is very similar to hypophosphatemia – there are the same targets, same prevalence in the US and, to be honest, the same issue of finding the appropriate patients for therapy. With Takeda’s NATPARA^®, patients with hyperparathyroidism were accessible but the challenge was finding the appropriate patients who weren’t managing their treatment. With hypophosphatemia, it's a hereditary genetic mutation. Pediatricians can identify patients quite easily because parents would notice something was going on when their children become weight-bearing, and a diagnosis is apparent following a genetic test.  

For adult patients, it’s a different story. Even if they have been diagnosed years and years ago, and they're on supplements, the issue then becomes, how do you refine them? How do you go out there now and get these patients in the marketplace? The claims are all over the place and there's many claims that can point you in the right direction, but in this business, as in hyperparathyroidism, we need predictive lookalike data.

I've always found that the representatives and salespeople will focus the leads they feel are the most lucrative. Over my time working in three different rare disease businesses at Shire, these are the sort of leads that the good reps jumped on because they were pointed in the right direction. The data that IPM.ai has provided is excellent for hyperphosphatemia. Many of these patients would see four, five, six, seven different specialties. The reps would then focus on maybe four of the seven doctors, or even all of them, and eventually find the most likely physician managing a patient. I think it’s not just about the initial data outputs, it's also the changes that patients will go through in their diagnostic journey. We did some work and I brought IPM.ai to Kyowa Kirin and one of the obstacles was that no one in this organization was aware of artificial intelligence/machine learning specifically for patient finding on this scale. 

One area where it's been useful as well is in the oncology business for POTELIGEO^®. Cutaneous T-cell lymphomas (CTCL) patients have a very interesting journey because they'll experience sudden changes in their condition, whether it's skin or blood – and this will lead to immediate real time triggers that indicate a medical event. This information is very valuable for representatives out there because it alerts them that now is the time to outreach to a health care provider (HCP). One particular challenge is that a HCP may only see a patient once a year, so the timing aspect is hugely important because the physician may not remember a patient they met once, months ago.  

Perhaps the greatest obstacle was really convincing the internal team to try a different approach to patient-finding, because it requires a different mindset. You have to point out that this area is evolving, and that there’s nothing to lose by trying it. Obviously, what’s standard practice three years from now is going to be a lot different from what was happening three years before, for example, and especially with evolving regulations.

John: Yes, you really have to go through a lot of organizational mind-bending because it's not the normal way of doing things – it’s a new and emerging thing so a lot of people don't understand it — some people think it's a magic eight ball and it's going to give an instant answer, while others realize that it's augmented intelligence, which will make you a little bit smarter. As humans, we still have to interact with the data and find attractive leads. What was the company’s process before working with IPM.ai? Did Kyowa Kirin sourcing through somewhat artificial intelligence and machine learning or was there a more traditional way of going about finding these patients? 

Richard: I think it boils down to when you're doing salesforce optimization. I remember taking over the lead for the NATPARA^® business and I just could not figure out why we had so many reps at the time. Traditionally, a vendor would come in and basically divide up the country into the number of endocrinologists in each area. This actually creates confusion around opportunity, while AI/ML analytic technology allows companies to act much more efficiently. 

We got to a point at NATPARA^® where we would get down to the territory-level IC plans. Initially, there’s an internal question about how many patients reps can add, but I think once they become more familiar with the system, they recognize that it's pretty accurate. I've always found you've got to try it to prove it, but the hurdle is getting that trial in place. The beauty in rare disease is that the payouts are great – you don't need to do much. Once you’ve found several patients you've paid for the program.

I think a lot of people initially think IPM.ai can predict all patients and all you need to do is send a rep, but actually, you're taking a hit rate from 0.1% to 1% — which is great because it’s ten times what you had. I think there’s a lot of education that has to happen and people have to mature into the idea, which was my experience. 

I’ve had many discussions with a lot of firms doing this and I thought this was a pipedream. COVID pushed the industry into digital and that has created a lot of noise. I’ve found the best approach is to test organizations out for yourself and prove the value internally. Within Kyowa Kirin, no one was really familiar with AI/ML for patient finding. There was some understanding for study recruitment and on the development side, more so than the commercial side. Throughout the industry, the same hurdle of patient finding has presented itself. The problem across regions is the access to data. However, I think the strategy and way of thinking is correct. 

John: How did you find out about IPM.ai, since you mentioned that nobody was really looking at artificial intelligence machine learning solutions? 

Richard: It just came about from networking and being in the industry; it's a close-knit community and it makes sense that in rare disease, we would be the first to jump on this because the biggest hurdle in advancing a drug is access to the appropriate patients. There’s a need there that is lacking in other areas of the business.

Following that, I think an even bigger question is, which firm do you pick? I was attracted to IPM.ai because AI/ML patient finding is your area of expertise. That focus is what set IPM.ai apart, coupled with the data sources and how much coverage you get across the universe.  

The end benefit is aiming to shorten the diagnostic journey. I think the majority of people in rare disease and in oncology truly want to make a difference, and if you can take a journey from ten years down to three years, that's a pretty good story to tell. You've changed a lot of lives, which is nice. 

John: Did anything surprise you when you started working with us?

Richard: Not specifically, but I think the only surprises were positive and centered on the power of machine learning and the way it can be constantly optimized. That's the other positive with working with a company like IPM.ai – and sticking with one partner as the data stream evolves.

The trigger that got me first interested was when I first joined Shire in late 2015 and a person on my marketing team had looked at IMS data and they tried to overlap that and map the patient services data. This was sort of the first step into the predictive nature of the data. We tried to get the results published, but patients had only consented to internal use so we couldn’t release it even though it was all de-identified. That was my intro where my interest was first piqued. Looking back at that six years ago, it seems pretty primitive, but let's see what's happening ten years from now. 

John: Do you predominantly use this for commercial effectiveness and commercial operations?

Richard: Yes, but only because my remit is the commercial side. I would certainly do it if I were working on any clinical programs. I would definitely advocate for IPM.ai on all levels of uncovering patients. Kyowa Kirin is definitely focused on specialty – they're not bringing drugs to the mass market and we’re always working with orphan indications with small patient populations. So, absolutely, if you can take the recruitment down from two years down to eight months, or something like that, it's fabulous. 

John: Can you share any returns you’ve seen from working with IPM.ai so far?

Richard: POTELIGEO® has some good data. I think 13 patients have been uncovered because of the work with IPM.ai. I always think that there’s a great benefit when there’s a positive response from the sales reps. 

John: Can you measure if patients are getting in earlier at the time of the responsiveness?

Richard: I would assume with these data, you’re getting the triggers at the right time, which lends itself to an abbreviated timeline, so from engagement with a representative to beginning a therapy is much shorter. It would be an interesting thing to mirror that up against the traditional way of doing things, which relies more on crossing your fingers and hoping for the best. If you can prove you've gone from six months to a month, that's an incredible result. 

John: You mentioned that 260 had been previously found. What was the estimated total patient population of this condition? 

Richard: I can't remember exactly but there are two indications there. It's not a big brand and it's within a competitive market. POTELIGEO^® has its place within the patient journey but there is noncompliance. This differs from the XLH world, where patients stay on the therapy for life and have fantastic compliance rates.  

I think it just took a little bit of time to realize how this actually works and win over the skepticism. Ultimately, you can’t expect you’re going to find every single patient. I've had people say to me, because you did this predictive model for XLH, and we have a prevalence that's probably way off – we say it's one in 25,000 in the US, but let’s imagine there are 12,000 patients. The initial outputs are a different level of lookalike. It’s crucial to remember that identifying 18,000 tokenized patients doesn’t necessarily mean those 18,000 patients are confirmed, but rather 18,000 people that could potentially be patients. Again, it comes down to understanding what the AI/ML is doing. 

John: To a large extent market noise has tainted expectations because the market has become so oversaturated. Even if you have great performance gains, people’s expectations are still unrealistic. What do you think we can be offering more of?

Richard: There's a lot of players now but what I love about IPM.ai is that patient finding is your main focus. I think the problem with a lot of these other companies is that they've built something onto their current capabilities and they’re trying to do everything. What I love about IPM.ai is the customer service and availability of staff who are always checking in and asking if there’s anything additional needed or further explanation. For example, I asked if we could map physicians around the de-identified patients to figure out the most likely to prescribe and manage the patient, which illustrated the influence network.  

John: The triggers and targets that we're delivering right now are to the field, but what are your thoughts on having a trigger directly integrated into the physician’s electronic medical record (EMR) system?

I think it would be great. I heard about this idea a lot when I worked in hereditary angioedema. Physicians would love to partner with industry on this because a shortened diagnostic journey betters the collective good. If industry can fund some of this work and then the physicians could use it directly, if there’s a way to do that, I believe it would be successful.  

John: Thank you, Richard. We appreciate your time. 

Richard: Thank you. 

About the Author

John Seaner

Chief Marketing Officer

John leads a team charged with amplifying our brand, communicating the organization’s business value, and earning the loyalty of our clients. He has a 30 year track record of driving innovation, operational improvement, profitable revenue growth and sustainable competitive advantage for technology companies ranging from early-stage startups to global billion-dollar public organizations. John previously served as Chief Marketing Officer of 1010data, provider of analytical intelligence, consumer insights and data sharing solutions; Chief Marketing Officer of Signals Analytics, an early pioneer in the emerging Decision Science as a Service space; and Vice President of Global Growth Marketing at Medidata Solutions, the world’s leading clinical development lifecycle platform utilized by over 1,500 life sciences companies. Throughout his career his teams have earned notable awards and accolades, most recently for pioneering new approaches in account-based marketing, including the use of digital ethnography to uncover the customer decision journey, the utilization of data science to optimize demand activation and the application of signals intelligence to enhance customer empathy.