Hudis spoke with Paul Goldberg, editor and publisher of The Cancer Letter.
Paul Goldberg: Thanks for agreeing to make an effort to get this through my thick skull.
Clifford Hudis: It’s kind of you and overly harsh. But anyway, I am happy to talk today.
How would you summarize what is happening? Is this a sale? Is this a licensure?
CH: This is a licensure and a strategic collaboration. This leverages the expertise of several organizations to allow us to deliver real value to our members and subscribers faster than we would have been able to otherwise.
First of all, who is involved?
CH: There are two parties for this. One is Tempus, a technology company that provides genomic sequencing services and structures and analyzes molecular and therapeutic data. It is based in Chicago and run by Eric Lefkofsky [founder and CEO], who is one of the founders of Groupon.
And the other is Precision Health AI, which was founded by Romesh Wadhwani.
They each have expertise and experience in complementary domains, but they came together to talk about licensing access to our data.
And that’s what this deal is.
In the simplest terms, this deal consists of three things:
Number one is that our collaborators gain license to access our data for a period of time. We obtain a revenue stream, which defrays a substantial—but not total—cost of CancerLinQ.
And they also bring curation expertise, and that’s something that all real-world data needs. Plus, everything they do to curate the data for any purpose also simultaneously feeds back to enrich and improve the CancerLinQ data.
Finally, they bring exceptional technical expertise, so they can help us deliver vital offerings back to our members faster than we were going to without them.
Can you give me the numbers?
CH: Which part of the numbers?
We could start with how much ASCO has invested in CancerLinQ, and how much it costs, to maintain it?
CH: ASCO has made a steady and growing investment in CancerLinQ, because we truly believe in the vision that better data can lead us to better cancer care.
This investment includes not only revenue, but also the intellectual resources of most parts of ASCO so it is hard to fully quantify. Even with this deal, we do not foresee a positive operating margin from CancerLinQ, but if its services enables improved care, then it will be a success for our members.
How do you report the CancerLinQ expenditures?
CH: There is an annual budget for CancerLinQ, but one of the nuances of CancerLinQ is, it is a wholly-owned non-profit LLC that’s completely contained within ASCO, so it’s not a separate business financially.
It’s a wholly owned enterprise within ASCO, so ASCO’s budget, it contains CancerLinQ. And there are some shared services, so it’s always a little tricky to get very specific about CancerLinQ revenue and expense.
It’s not a line item?
CH: It’s many line items, that’s right.
One of the problems or challenges we have faced, of course, is that the entire operating expense for CancerLinQ significantly exceeded its revenue, and it has had to be backstopped by ASCO funding.
We are cutting the deficit from CancerLinQ through this transaction. Our goal is to get CancerLinQ to break even.
We’re not quite there, but we’re closer and we’re sustainably close enough. It’ll become a modest enough loss that we could afford to do it, if it’s delivering value for a long time.
How many people do you have working on CancerLinQ?
CH: I think it’s about 55 right now. It may go up this year of the transition.
Do you have any estimates of what it costs to deliver on the promise of CancerLinQ—what the total investment level might need to be?
CH: My guess is over the long run it’s hundreds of millions of dollars if not even more. By the way, if you think about what drug developments numbers are, what I just said isn’t so shocking, is it?
No.
CH: Right. It’s not feasible for a nonprofit.
What about the deal with SAP? (The Cancer Letter, Jan. 23, 2015)
CH: We have a contract with SAP, where we are the customer.
They provide technical support, warehousing, and additional services that are not directly or immediately affected by what we’re doing here.
What is the data they’re licensing from you?
CH: Let’s just back up a half step.
The way that the CancerLinQ works is distinct from the registries that some other professional societies have created.
I actually think one of the goals of our discussion should be to clarify the distinction. These registries are, generally speaking, purpose-built and used to answer a specific question. It’s easiest to describe with an example.
Imagine that the FDA needs surveillance on outcomes with an implanted device. Maybe the relevant society would develop and run a registry and their members would record all the device implants along with certain relevant fields. Then, when you’re done, you have a pretty good snapshot of what’s going on in the world with those things.
And that’s something we’re used to. However, CancerLinQ is very different.
It represents a download of the totality of all of the recorded data in the electronic records of participating doctors and practices and hospitals. In that regard, it’s deeper and richer than any other focused dataset, but it’s also less structured and it’s often less complete. And its fitness to answer any specific question is variable. The data is sometimes there, but not in the same places in every chart.
Again, it’s easier to describe with specifics than generalities. You could imagine that for one patient, there is a staged description of their cancer, T2, N1, and M0 tumor in a data field labeled TNM and that’s structured. You can imagine the very next patient had their procedure done outside of a health system, and a physician had a paper note in front of him and it’s literally written into a progress mode with the TNM in it.
So, while both are functional for the patient and doctor and contain that information, in one situation, a computer program can recognize the labeled field and say, “That’s T, that’s N, and that’s M.” But in the next, it requires either a real or virtual reader to parse it out of the written notes.
So that’s curation. And the only way that you can take existing records like we have and reliably find the data outside of the structured fields is to curate them.
And that’s expensive.
CH: Yes, and we’ve been doing it already. We’ve learned a lot and we know what it takes. Our new collaborators bring in complementary curation skills to the mix here for us. And as one example, if you go back to Groupon before, which is what Eric founded, you know Tempus obviously has deep experience looking at data inside and outside of medicine.
On the other hand, Romesh has tremendous experience with medical data and finding the relevant components of it, and also with importing external data to enrich what we have. So, we’re going to be on a journey together.
We’re going to be developing natural language processing, we’re going to be accelerating curation, and we’re going to be generating new and better data, so our users that will have a richer data resource than before.
And what they’re going to be getting is access, ultimately, to the de-identified data, and they’re going to be curating it, and sharing that with us while they’re able to commercialize the process.
How much data is it and what kind of data is it? How many patients, how many practices?
CH: We have more than 100 distinct entities already signed up for CancerLinQ with more to follow.
Of those who are signed up, we have been able to onboard around 40. Now, I want to pause here, because it’s not that we’re not trying, but one of the ambitions that we had was to serve all members, and one of the consequences of that ambition is we never said we would only use name brand medical records.
We said we’ll take any. And we didn’t say we would inspect your record and only take it if it’s above a certain threshold for completeness or quality. Adding even more complexity is the fact that each installation of each electronic record system can be unique to some degree. They are often customized, and they then get upgraded at varying rates.
We have to constantly adjust our intake system for each version and each installation of each medical record. The data we bring in is widely dispersed, it has to be homogenized and normalized, which is a CancerLinQ function now, and then it goes into a database. Then we have a series of databases that are increasingly clean and increasingly de-identified leading to the ones that our partners can access.
I see. But what are the numbers? Is it 1,000 patients, 500,000 patients?
CH: So, of the roughly 100-plus entities that have signed on, we have about 40 that are onboarded. We have about one million cancer patient records right now from about 600,000 individuals.
One of the wrinkles is, we’ve got, I think, it’s just shy of 3 million patients potentially right now, and rising.
On the other hand, they don’t all have cancer. For example, some of them are hematology patients or even general medicine patients that might be in a practice for various reasons. But when you get down to diagnosed cancer patients, we can analyze right now, the number is about 600,000 patients.
So, it’s 600,000 patients?
CH: Yes. But this is the beginning. We’ve been onboarding them for a year-and-a-half, and we have more than half of our potential patients yet to be onboarded, and we have steady growth in our market penetration right now. So, there’s a lot more out there ahead of us than behind us.
So, ASCO is continuing to do this?
CH: Oh yeah.
And you will continue, even after this licensure deal?
CH: Absolutely. There are two aspects of it.
We will always need to be updating this because, of course, treatment changes and modernizes, and we always will need to get current data. The second issue is that the real reason we did this in the first place and what this deal allows us to focus on even more is to support our members delivering high quality care.
That means tools and applications that run against their records so that they can improve their care. It also means that we’re going to be hooked up to the ones we have and hooking up more so that we can learn from more and distribute more knowledge and tools.
What about the rapid learning system? Is that still going to happen?
CH: Well, that was the beginning of all of it. You’ll remember the discussion back at the beginning of the decade. And on an aspirational level, I would say that’s still very much out there for us.
But we are not yet able to fully realize this vision today. We are, however, going to be building towards this as we develop tools and services.
I see. How does monetizing look in this case? What is the structure?
CH: So, again, we are a non-profit, a 501(c)(3), we have a societal mission, and we could neither afford to develop this rapidly on our own, nor could we obtain all of the technical skills to accomplish our vision quickly and efficiently internally. That’s why we need collaborators.
We needed to find mission-aligned or compatible entities who saw a way forward to a successful business model so that they’d be willing to invest and would be able to support us in delivering on our mission.
It was an interesting journey to find the right structure. What they’re going to do, and we certainly will support them, is go out to pharma, payers and other for-profit entities, and they will be able to sell them analytics running off of curated, enriched and de-identified data. They see a business model that is obviously profitable enough to justify their long-term investment.
Can you think of specific questions that they might be able to answer?
CH: Well, they’ll be able to answer questions about real-world use and outcomes for treatments.
Any other clinical questions? Like how are people using drug A?
CH: That’s what I mean. So, a typical example would be you take some modern therapy, newly approved in phase III, where a group of very qualified patients have been used in a study to prove their treatment has the given effect.
And that provides the label. It is typical to find in the real world that the patients who get treated of course don’t completely match that pristine cohort in both specific or general ways.
Patients in the real world have hypertension, diabetes, and other medical problems that may have been exclusion criteria in the formal studies.
As a consequence, it is common to find that the use of the drug in terms of, let’s say, duration and its benefits, is modulated in the real world. That’s the kind of information that we will get here. That’s a typical, real example.
What about a question like, what is the prevalence of a specific condition?
CH: Sure. Natural history, outcomes and more. Remember, at the same time, some of these are the kinds of questions that would be of great interest to academic researchers and wouldn’t be equally so to for-profit businesses.
And ASCO continues, therefore, to provide access to our ever-improving database for the cancer community. That’s one of the direct benefits of the deal.
And you continue to co-own? Do they own it? Do they license?
CH: They license. There is no sale of anything here. They are paying us an annual fee, and you could really consider that fee as dollars and curated records. And based on that, they are able to license the data and sell it to the markets that are theirs.
And how long is this for?
CH: Well, I hope for a long time. It’s a ten-year deal in structure.
There is an offshoot of CancerLinQ. CancerLinQ Discovery I believe it’s called. Is that what is being licensed?
CH: CancerLinQ Discovery is our offering to the academic community and others for research. And that’s one of the beneficiaries of this deal, because the data in CancerLinQ Discovery will be more rapidly improved from their curation.
The CancerLinQ Discovery and research and publications committee, which we call R and P, that will continue to function reviewing all requests to access data from noncommercial customers. So, all the usual academic researchers who come through CancerLinQ will follow the data access policy, and they will go through R and P review, and they will gain access to CancerLinQ that way.
They (Tempus and Precision Health AI) will, you know, be able to pull into their data whatever CancerLinQ data is necessary for them to address their commercial needs. But de-identified. This is really important. Nothing identified ever goes through.
The part that I never understood is the value of this. Correct me if I’m wrong—I’m probably wrong—does the data come from community practices, as opposed to academic institutions?
CH: Not as opposed to, in addition to. That’s actually what makes this deal valuable, I think. It is not a selected cohort of people who go to referral centers.
Nor is it only the people who go to private practices. It is the full range of government-supported, low-income centers, to high end and socio-economic settings. It is small to large, it is private practices through most highly respected academic centers in the country. It is a full range of practices and settings.
That’s good to know. But of those 600,000, is there an overall balance in this picture?
CH: Well, that’s a really interesting point. To make this data most valuable, one of the places where we will be collaborating with them on an ongoing basis is to make sure that the curation is balanced across these different types of practice settings, as well as different diseases and demographics so that, over time, the data set is broadly reflective of all these elements.
So, you have to really grow the dataset to be reflective of what’s actually going on?
CH: Well, it’s not just grow the datasets, which we do have to do, but we also have to be thoughtful about the priority we place upon curation of different subsets.
For example, if we only curated data from a tertiary referral center, our improved offerings of CancerLinQ Discovery would be skewed. Instead we have to make an effort to make sure that as we curate, that we’re curating from all of the practice settings so that the data that researchers and our partners have is truly real world and not a skewed subset.
Do you see a possibility of using this for drug approvals, running clinical trials?
CH: I think it will contribute. I also think we have to be humble about real-world data for lots of well reported reasons. Let’s just talk about this for a couple of minutes, and you start to see how this fits into the broader canvas.
In the past, drugs would get approved and labeled and you’ve covered this many times in the last few years under different stories. It would get approved, and then they would get used by the community and they would typically get used off-label.
And that’s because there was little incentive for companies to go and file for a label in every possible indication. Also, there was an expectation after the FDA said a drug was both safe and effective, that the community would identify all of the uses, continue to adjust it.
After all, the FDA did not, in a restricted way, prescribe the practice of medicine. What they did was enable it. And this is an important difference there, because this comes up all the time when we talk about high drug prices and the role of the FDA.
Their job was never to control the practice of medicine exactly, it was to make sure that drugs are safe and effective when they’re put into the market, and allow the marketplace to continue to refine their use. This is our history.
And to prove it, you can go back and look at any of a dozen chemotherapy drugs from the 60s, 70s, and 80s, and then look at where they’re used first at the label, and you’ll remember there wasn’t really controversy about that.
It was expected. As drugs became much more expensive in the last years, many third parties began to restrict use in various ways to the label. And from a certain perspective, this was not the intention of the FDA label. It wasn’t established to restrict access or use, but it was applied that way by a third party, not by the FDA, but by the third-party payers.
So now we go in the modern era, and this will get us to CancerLinQ, but let’s just think for a second about what happened with ASCO’s TAPUR study. You’ve got next generation sequencing taking place all over, and this is even before the decision on the Foundation Medicine test last week.
And now you’ve got docs who have the following situation: they have a patient with a specific histology. In that histology, they have a specific genomic alteration, they have a reasonable, theoretical argument to give a targeted drug to that patient. That targeted drug is on the market because it’s FDA-approved for a different disease, and they yet find that they can’t do it, because now the label is being used to restrict your access by the payer.
The TAPUR study matches those patients, allows us to learn whether there’s something there or not. And in the latter case, if we treat a small number of patients with a histology and mutation or alteration and a targeted therapy and there’s no response, we can reasonably conclude that even though it made good sense, that’s not an effective therapy. And that’s helpful to patients, the community and the industry to know that.
The flip side is, we might see some evidence of activity, and that evidence of activity might lead to either a bigger, purpose-built study in some cases because the sponsor could build on the fact that there’s a bit of a promise already, there’s some evidence of activity. Or it might lead to actual label extension in some cases. Maybe the agency would accept the TAPUR data which is closer to real world because the eligibility criteria and the whole way we run TAPUR is closer to everyday practice.
So, now that brings me back full circle to CancerLinQ. We have real-world patient data in there, some of which might be convincing at least that there is the possibility of activity for specific drugs and matches.
Maybe that’s just enough to get a study launched, maybe in some cases it’d be more convincing than that. And then the other side is, the CancerLinQ data may provide for modern synthetic control arms, which would allow companies in some cases to either model more efficient studies, or even use the control data from CancerLinQ to support label extensions. I think this is a new world, and we’ll have to see how good the data is and whether it can be used for any of these purposes.
Maybe the question is how does this loop into CancerLinQ … Let me rephrase that. How does next-gen sequencing loop into CancerLinQ?
CH: One of the challenges we have, frankly, is only certain patients under certain conditions are getting sequencing.
Only some of that data is necessarily being recorded or fully recorded in the charts the docs have. And so, it is not the case that CancerLinQ is necessarily a genomic data set. It’s just the recorded data as it exists for individual patients.
To the degree that some patients will in decent numbers have accurately recorded, well done genomics and to the degree that they can be subset-ted and identified and studied, yes, it’ll be another contributor.
But, this brings me back to where I started at the beginning. One of the challenges of that it is a full data set versus a registry. It is “take it as you find it” data.
It would be more useful as a way to find signals of something you might want to try?
CH: Absolutely. Right, I think so. And also, there are so many things we don’t know. It’ll be a way of identifying natural history in the real world when there are other diseases present. It’ll be a way of identifying the size of populations, which are increasingly segmented by genomic and other factors.
It’ll be a way of identifying geographic distribution and access to patients for planning a research study. So, this data is going to have value in many ways, but I think that value may be in some cases different from what many traditional researchers imagined. It’s not a replacement for prospective planned, well done research. It isn’t. And we would not be doing ourselves or the community any favor by misunderstanding it that way.
One can easily imagine that it will provide insights that would allow us to develop more efficient prospective studies. It’ll contribute to a world where we don’t have to do quite as many studies that are dead ends.
Which of the three parties is responsible for developing clinical decision support? Which of the three parties is developing real world evidence capabilities?
CH: All three parties will be working, both collaboratively and in parallel, to develop clinical decision support tools and to evolve the clinical and scientific utility of real-world evidence. The primary role differentiation relates to commercial activities which are the exclusive responsibility of Tempus Labs and Precision HealthAI.
ASCO’s commercial partners are using CancerLinQ data to build enhanced capabilities based on CancerLinQ. Is there any aspect of this endeavor that only ASCO is doing? What is ASCO’s role in CancerLinQ now, as a result of this deal?
CH: CancerLinQ’s prominent role will continue in a fashion consistent with the founding principles of the initiative. CancerLinQ will continue to:
Maintain the relationship with the participating oncology care teams.
Maintain custodial control and responsibility for all personally identifiable data.
Manage the outreach to and ongoing relationship with cancer care centers.
Manage the technical integration with subscribers’ EHR systems.
Manage data ingestion and normalization.
Offer a de-identified data resource for academic researchers.
Manage the various committees of patients and clinicians who provide guidance to CancerLinQ.
Manage the Oncology Leadership Council, that growing body of sister societies and regulatory agencies that provide guidance to CancerLinQ.
How do revenues get to ASCO to get CancerLinQ to revenue neutrality and how soon will this happen?
CH: CancerLinQ will receive an annual royalty in exchange for a license to utilize de-identified clinical data. This licensing fee will make up the majority of CancerLinQ’s operating funds beginning in 2018.
How effectively are third-party payers using real world evidence now? How can real world evidence be used to inform their decisions—and do you see this actually happening?
CH: At present, claims data continues to be the predominant form of data used by third-party payers. However, we are already seeing a gradual evolution in their utilization of real-world evidence, just as we have seen at the Food and Drug Administration and in the bio-pharmaceutical community.
Is there anything we’ve missed? Anything you’d like to add?
CH: I hope that the world sees in this deal that a professional society has taken ambitious steps to build a previously nonexistent resource.
We have expended tremendous resources to get to this point, and we have now identified a novel way to transparently partner with for-profit entities to bring expertise to allow us to deliver faster on our mission.
That really is why we’re doing this. We are of course sensitive to possible misperceptions about this deal. It is not a sale; there is nothing being sold. It is a licensing arrangement, it has a term, we will see how it goes.
But our number-one aim is to enable our members to deliver higher-quality care more broadly and faster than before.
