Roche’s acquisition of Flatiron Health signals the pharmaceutical industry’s interest in using real-world data to measure value, said Brigham Hyde, CEO and founder of Precision Health AI, a company that uses artificial intelligence to define cancer datasets for precision oncology.
“If I were a pharmaceutical company, having more detailed information about my customers, be it patients or physicians, is always a strong place to be and I think they will probably focus on the measurement of value,” Hyde said. “I think outcomes quality data is the key to all of this.
“If I was pharma, I would look in the mirror and say, ‘We really need to know our customers.’ I think that involves data and we need to be able to digitally engage them. And for Roche and Flatiron, they get that. They’ve got an EMR facing the patient-physician interchange, and they get really unique data to track and profile their patients. It makes sense.”
PH.AI is not involved in the Roche-Flatiron deal.
Hyde spoke with Matthew Ong, a reporter with The Cancer Letter.
Matthew Ong: Is the Roche acquisition a signal that oncology bioinformatics is moving away from the startup space into billiion-dollar investment sfrom pharmaceutical companies?
Brigham Hyde: I think the price tag is obviously exciting for those of us in this space. I think what is probably more interesting is that pharma would not have been considered a buyer, classically, for this type of business, and they’re now entering the space. That’s really interesting.
I also think this signals the importance in oncology in particular. That market is incredibly competitive and deep EMR clinical data is a real important differentiator for the brands in that space.
Has anyone else bought an oncology bioinformatics startup?
BH: Not to date, that I am aware of. I do think there are some companies who have made investments. Actually, AstraZeneca was announced as an early partner on CancerLinQ and continues to be a partner going forward. I think there are others who look around that have, I think, small investments in different places. Novartis has made a number of digital and RWD based investments over the last several years.
Although, it’s important to realize Roche is unique. They’ve always had a diagnostics business, which puts them a little bit in the data space already and also in the clinical decision support space. They have run that successfully, and somewhat separately for years. Not many other pharmaceutical companies have that sort of combo.
They’re certainly a unique player, but our perception is, this is at least related—maybe not directly, but certainly indirectly, to the noise around Amazon as well as a lot of public consolidations between players like Aetna-CVS, or Amerisource Bergen and Walgreens, because data is power when it comes to negotiating price and value of therapeutics. So, a very exciting time.
Do you think all of this will speed up evolution of clinical decision support infrastructure that can also generate regulatory-grade data in precision medicine?
BH: I think that’s the big question. Will providers accept clinical decision support from pharmaceutical companies, or will they be viewed as somewhat biased? I’m in the camp that providers and pharma are actually natural partners, because they both face the insurance companies in terms of negotiating value for their care, so I actually think there are examples in other parts of the space where pharmaceutical companies do a lot now with ACO’s and IDN’s to help support care coordination, and patient communication and outreach and things like that.
So, yes, I think that’s the directions it’s heading. I think the potentially interesting question here is: is Roche trying to be a precision medicine company and make individual decisions for patients alongside doctors? I think that’s an exciting question.
And with Roche now at the steering wheel, do you think that we might be seeing more of a push to set standards for health records?
BH: You know, that’s an interesting question. I’ve been in the business a long time, so I’m a little cynical about making that change over night for standards. I’ve seen a lot of those come and go.
On the other hand, I think you have a very progressive and forward-thinking FDA at the moment, who is really interested in having active discussions about how to regulate the use of AI and RWD in clinical decision support. I expect active discussions to continue between the FDA and a number of industry players.
The 510(k) pathway for approval of the decision support algorithms is out there. I do think there’s going to be some new and updating, at least guidances, if not as the more defined pathways yet to emerge. I for one am excited for that, being in the artificial intelligence space, I think there’s a lot to do, a lot of low-hanging fruit are available, given the right data to empower doctors to make increasingly more precision decisions for individual patients.
Do you expect other pharma companies to start snapping up more health teach companies?
BH: I do. I certainly think it’s possible. One of the things I would say: if I was a pharmaceutical company and I was trying to figure out what Amazon was going to do, I wouldn’t look at Amazon’s ability to manufacture drugs and run clinical trials and probably not be particularly concerned. On the other hand, their ability to distribute profitability and efficiency, as well as the general approach around the data they gather about their customers to better serve them goods and services, I think is a potential concern.
So, if I were a pharmaceutical company, having more detailed information about my customers, be it patients or physicians, is always a strong place to be and I think they will probably focus on the measurement of value. I think outcomes quality data is the key to all of this.
And speaking of Amazon, do you know if they’re doing anything in the informatics and precision medicine space?
BH: I know as much and as little as anybody else does, but I think it’s natural to assume that they serve the patients that pharmaceutical companies serve. Today they’re delivering them paper towels and bottled water today, or they could be delivering them other types of things tomorrow that are close to the pharmaceutical setting. I think that would make sense.
I know that they’ve gone into the consumer health product space and there’s some discussion about them going into the medical technology or device setting, at least as the distributor. I think that makes sense for their business.
I think what’s equally relevant as Amazon are the Googles and Facebooks of the world. I think these are the people who are capturing tons of information about how patients are experiencing care, and the behaviors they have that are leading to potential health outcomes. So, I think all of these companies will be heard from in one way or another.
If I was pharma, I would look in the mirror and say, “We really need to know our customers.” I think that involves data and we need to be able to digitally engage them. And for Roche and Flatiron, they get that. They’ve got an EMR facing the patient-physician interchange, and they get really unique data to track and profile their patients. It makes sense.
You’re saying that Roche’s is responding in a preemptive fashion to take the lead in shaping the health tech space?
BH: I think that’s very clear, absolutely. Pharma has dabbled in this for many years and they’ve tried different forms of digital engagement. It’s important to recognize that they’re regulated in what they can say about the benefits of their drugs, etc., most of their investments are centered around that. If you look at different apps that are out there, for instance, they’re mostly around decision support, or information for a given patient population.
This is a bigger step in that direction. It’s a question mark to wait and see how providers react to the idea of the EMR they use being owned by a pharma company. Even though that would be highly regulated, and I do take Roche at their word, that they will operate Flatiron separately. But it is an open question.
If you think about outpatient oncology treatment and you think about the drugs today that are sold as buy and bill—situations where the provider is a partner with the pharmaceutical company already. They purchase drugs ahead of time and they are responsible for selling them to their patients. There’s already a lot of alignment between providers and pharma. So this may be the natural next step.
I’m actually in the camp of, in the right setting, when they work together, they can actually provide a lot of value to patients, and coordination can drive that value. I’m sure that’s the message that Roche will be taking out around this topic.
Is Precision Health AI funded by any pharmaceutical companies at this point in time?
BH: We’re privately funded and we are not funded by any particular industry partner. That is somewhat deliberate, we like to be able to be Switzerland—sort of, a trusted independent, if you will, for a couple of reasons.
I think, number one, it allows us to gather data from multiple sources. So, I’m not an EMR. Flatiron is somewhat restricted by the amount of data they can capture, based on whether or not their EMR is installed. So, I don’t have that problem. I can buy data from everybody, both on the EMR side and on the genomic side. I’m not a genomic testing lab, so I can partner with those folks.
From that perspective, that’s important to me because when I’m trying to serve my pharma clients or my provider clients, it’s important that I have the biggest, deepest, richest dataset I can get my hands on. We are an AI company, AI only works when you have a lot of data to train it on. I need a good training set, so it’s important for us to stay independent at this moment.
I certainly work with pharmaceutical clients all the time and that’s the discussion, which has amplified a bit over the last couple of weeks for sure, but for now, I think it’s important to them to have a neutral party.
One of the other question marks is, how will other pharmaceutical companies react to the fact that Roche may own a data partner that they have worked with? And I think that Nat [Turner, co-founder and CEO of Flatiron] in his blog post made it very clear that they are going to operate independently and Roche has a history of being able to do that in their diagnostics business. So we will see how it plays out.
But that is definitely going to be a question mark, particularly where there are competitive assets. Be it Pfizer or BMS competing with Roche Genentech, how will they feel about this? So, that’s yet to be answered.
So, in the meantime, for PH.AI, we are trying to stay independent, and grow our data assets and be as useful as I can to pharma and these other types of partnerships.
Could you briefly describe PH.AI’s business model? Also, what does artificial intelligence in oncology mean for physicians and hospital administrators?
BH: We work with pharma and providers, and a little bit with payers, but pharma is our primary client. We sell three things. We sell data that can be used by those companies for research. We sell applications, maybe it’s tools for designing clinical trials or doing outcome-based research, so, software tools. And then we have our AI platform, which is called the Eureka Health Oncology AI Platform.
What AI means to me, and I’ll try to be as simple as I can about this: in AI 1.0, you have Watson, and you have basically a big tool box. You could load your data up into it, you could pick an algorithm, you could run it, do what you need it to do. You usually have to have people who know what they’re doing to do that, and then you have to know what to do with the outcome.
You’re talking about IBM Watson, right?
BH: Yes, that’s an example. They’re sort of the ground breaker. They laid the groundwork for a lot of what’s been done in AI in health care. And we know they’ve also had some struggles, places like MD Anderson and others.
So, what we’re doing with Eureka, is that we basically pre-trained AI on the data that we have to do specific things. And we have about 60 or 70 pre-trained AI models where somebody has gone in, come up with an important question, and with the data, they have tested out a whole bunch of algorithms.
Sometimes you have deep learning, sometimes it’s something more basic, just depends on what works best. We have built these productized modules to sit on top of our data, or any other data to make predictions and serve specific functions.
And I’ll give you an example. We have a number of algorithms focused on adverse events that can predict which patients are likely to have adverse events to things like chemotherapeutics, so neutropenia is big problem.
When giving chemotherapy, knowing who to give it to and who not to give it to are very important. And our models are essentially trained to predict that, offer suggestions to a physician, and they can take action from there and look at the evidence.
Right now, most of the time we work with pharma, these things are meant to better help profile their populations. So, help predict who’s going to take their drugs, who’s going to do well, things like that.
We have not rolled these out into practice in a big way, so we’re still in discussions with regulatory bodies and going through the peer review process and things like that, to make sure these things are really validated by the community.
But I imagine a world in which inside your EMR, if you’re a doctor, you could run one of our algorithms—think of them almost like an app—and I could run the adverse event app, and it would give me a prediction on that individual patient. And you could make a decision on how to use that data.
That’s what we’re doing. I think that’s why we’re gathering the data we’re gathering, is how do we build these prediction tools that are pre-trained for specific things that are oncology specific that can do work and help physicians.
Is clinical decision support an avenue you plan on exploring for PH.AI?
BH: It’s really too early to say anything specific, but it is certainly core to our roadmap. But all I can say is, stay tuned. We’ll have more to say about that in a couple of months.
How are things coming along in terms of your work with Tempus on ASCO’s CancerLinQ?
BH: It’s going great. It’s been fun diving in and getting deep into their data, which we’ve had access to for a while, but now it’s flowing on a regular basis and we’re prioritizing it, we’re in the market with it. In short, I think the data is of extremely high quality from the perspective of it’s representation of cancer patients globally.
There’s a great mix of communities and academic centers, so it’s got a lot of bulk and it has it’s own set of challenges like every dataset does. It represents over a dozen different types of EMR, so it’s very diverse. We were expecting that. We’ve worked through the data cleansing and all of these things, and are really getting some value out of it. I think you’ll see a number of exciting publications from us and our partners over the next couple of months. We’re very excited about it.
I’ll point out that while ASCO CancerLinQ is a great source for us and a good starting place, we have also added additional data to that. We continue to expand both our size, scale, and detail and all of that. Other EMRs and other sites, whether it’s other genomics data or data that helps round that out, we are full speed ahead on all of those fronts.
What can we expect to hear from you over the next few months?
BH: I think the next thing you’ll see from us will be around [the Healthcare Information and Management Systems Society 2018 conference]—a couple of key partnerships and product announcements around our Eureka AI platform, which we’ve been somewhat quiet about, but we’ll soon be very loud about.
And then a number of exciting things about the types of critical decision predictions where there will be new key clients working with us going into ASCO. So, it’ll be a busy three to four months.
