Flatiron CEO Carolyn Starrett: We’re reimagining the infrastructure of cancer care, and we’re going global

As real-world evidence becomes ever more essential, a cancer health technology company that played a key role in modernizing 21st-century health data is capitalizing on its accomplishments in the U.S.—and moving into international markets to meet the growing demand for actionable data.

Founded in 2012, Flatiron Health, now in its 10th year, is entering the third phase of its evolution as a pioneer of real-world data and machine learning applications in cancer informatics. 

Carolyn Starrett, a long-time business operations and strategic development executive at the company, was named CEO in April 2021, after Flatiron co-founders Nat Turner and Zach Weinberg stepped down from management.

“We had our startup phase and days early on. We then had an acquisition and a period of learning what it meant to exist post-acquisition,” Starrett said to The Cancer Letter. “Now, we’re internally talking about Flatiron 3.0. And I think 3.0 is an opportunity to really think about how we further advance and realize the mission that we set out to achieve 10 years ago.”

A decade ago, Flatiron was conceived when its founders went on a road trip to visit community oncology clinics and cancer centers. 

In 2018, Roche acquired the company for $2.1 billion, signaling a turning point in cancer Big Data and sparking a heady race among health IT companies to expand their offerings. 

The company remains autonomous as an independent affiliate of the Roche Group (The Cancer Letter, March 2, 2018).

Now headquartered in a swanky open-plan office building between Hudson Square and SoHo in New York City, Flatiron has subsidiaries in Japan, Germany, and the U.K., works with the top 20 pharmaceutical companies, and serves a growing portfolio of academic cancer centers and community practices. 

Today, the company is composed of more than 1,000 full-time employees and 1,500 flex-time employees around the world.

“We’ve learned that it’s going to be important to understand not just U.S. populations, but also a broader global population,” Starrett said. “We’re in the process of designing custom approaches for the context that exists within each one of those markets. 

“It’s now been, I think, a three-year investment in understanding the legal regulatory and compliance requirements and forging our first early partnerships. We announced our first partnership in Japan last year to study patients with gastro-intestinal cancers and you can expect more coming in the U.K. and Germany.”

Starrett’s vision for Flatiron focuses on providing answers to the big questions in health informatics, cancer care, and research of the next decade: 

• Speeding up translational science and the approval of new medicines, 
• Steering the development of integrated evidence by building an ecosystem that links disparate sources of data, 
• Creating access to high-quality real-world evidence across emerging markets and global populations, and
• Exposing inequities in cancer care and improving outcomes for patients with cancer.

“If we look at the landscape today, there are just so many core problems that still persist,” Starrett said. “It still takes over 10 years to bring new oncology treatments and therapies to market. Clinical trials are expensive, inefficient, and slow. They don’t really represent the patients who might ultimately receive and benefit from those medicines.”

As definitions of scientific and clinical significance broaden, and as researchers and regulators move towards more holistic interpretations of health data, Flatiron is viewing real-world evidence—and its many facets—as one component in a much larger ecosystem in which previously siloed streams of information are becoming more interdependent.

“What we’ve learned over the last 10 years as we got started with real-world evidence is that a lot of the questions we want to answer require careful synthesis and analysis of healthcare data that comes from multiple different data streams and sources and methods,” Starrett said. “And EHR data alone doesn’t solve the problems that we want to tackle.”

Flatiron is a founding member of the Real-World Evidence Alliance, which now includes 10 health organizations that collaborate and advocate for novel applications of real-world data.

“The goal here is to think about how to provide guidance, so that the best practices for planning and conducting and reporting on real-world evidence studies are clear to everyone involved,” Starrett said. “The hope is that will then help to improve the quality and the transparency of the evidence and the acceptance of these data sources.”

Flatiron and other alliance members have been working together with Friends of Cancer Research to develop real-world endpoints, with the goal of convincing FDA that these endpoints can be validated and used to inform drug approvals and regulatory decisions (The Cancer Letter, Sept. 25, 2020; Nov. 22, 2019).

“It’s been an important area of collaboration, because the endpoints in clinical trials are operationally defined quite differently than real-world evidence endpoints,” Starrett said. “They are captured in very different ways. So, the scientific bar for use of any type of data, including real-world data, is really high and needs to be reliable and relevant to the scientific question.”

Flatiron’s renewed commitment to health equity is led by Cleo A. Ryals, who drives the company’s priorities and strategy on health equity research. Because of its ability to rapidly identify gaps in care delivery, real-world evidence is able provide timely snapshots of unmet needs in communities, at scale.

“We found out, for instance, that socioeconomic status information was not captured well in the EHR,” Starrett said. 

“And so, we had to take a step back and think about how do we pull the right external markers and validation with full respect to patient privacy and practice privacy, and all of the different considerations that are important to define new variables, that we can then add into our integrated evidence to really start to understand where the inequities exist.”

Starrett said the company is directing its resources into generating actionable integrated evidence and technology to reduce the time it takes to bring new therapies to market.

“The real potential is still very much ahead of us,” Starrett said. “We are, at Flatiron, deeply focused on reimagining the infrastructure of cancer care, so that we can come together across the ecosystem and accelerate this learning and do it more efficiently and more sustainably.”

Starrett spoke with Matthew Ong, associate editor of The Cancer Letter.

Carolyn Starrett: Like many at Flatiron, I have a deeply personal connection to cancer. I found my first early-stage melanoma when I was 28, and all of my grandparents have experienced cancer. 

What’s been cool over the last six years is seeing how Flatiron has evolved. We’ve started to talk about the three chapters of the company, if you will. We had our startup phase and days early on. We then had an acquisition and a period of learning what it meant to exist post-acquisition. 

Now, we’re internally talking about Flatiron 3.0. And I think 3.0 is an opportunity to really think about how we further advance and realize the mission that we set out to achieve 10 years ago.

Flatiron was founded to answer a pretty simple question: Can we take the data that’s generated via routine care and make it useful? How can we improve lives by learning from the experience of every person with cancer?

We’ve really lived that mission over the last 10 years. We were pioneers in the evolution of real-world evidence in oncology, we see patients with cancer who have new treatment alternatives and options as a result of that real-world evidence, and we’re proud to have contributed to some of these meaningful advances. 

Moving forward, the things that get me excited are how we take this vision even further. We cultivate a virtuous learning cycle where we can learn from real-world experience, and then embed that back into research and development and access decisions around the world, and then, importantly, back into the treatment decisions that are made at the point of care.

That, in and of itself, is really the opportunity. It’s how do we truly reimagine the infrastructure of cancer care, so that we are creating a more modern and connected oncology ecosystem together with clinicians and researchers and regulators, and how do we leverage technology and AI and data that’s already captured to transform and accelerate clinical trials, so that we can better understand the right treatments for very targeted patient populations.

So, that’s really our goal moving forward: Closing this loop and starting to bring new options and insights back to the point of care, and continuing to partner with our biopharma customers, independent cancer centers, and academic medical centers across the ecosystem to drive systemic change.

I’ve seen that happen over the years; I’m looking forward to seeing more. What’s your vision for Flatiron, as CEO, and what should we be paying attention to as folks in oncology?

CS: I love that you asked that. My vision as CEO is to empower teams at Flatiron to build a world where technology and science can help close the gap between care and research.

If we look at the landscape today, there are just so many core problems that still persist. It still takes over 10 years to bring new oncology treatments and therapies to market. Clinical trials are expensive, inefficient, and slow. They don’t really represent the patients who might ultimately receive and benefit from those medicines.

And then, with the rise of new modalities and personalized medicine, understanding the best treatment for every patient is increasingly complex. And I don’t think we’ve fully maximized the potential of all the new therapies and learning how to best use them.

To us, reimagining the infrastructure of cancer care means generating better and more actionable integrated evidence and technology to continue to reduce the time it takes to bring new therapies to market. 

It means understanding exactly who can benefit from these therapies, and then making them available to the patients who need them around the world. 

It means integrating clinical research into  routine care and making it easy for every patient with cancer to participate in a clinical trial to close the gap in access and ensure that we’re learning from patients who look more like the real world.

Lastly, there is an important need to use data for good—to surface and understand healthcare disparities that continue to exist in cancer and to partner with our network of cancer centers to drive towards improving the outcomes for their patients.

CS: I think of integrated evidence as an emerging discipline. 

And what we’ve learned over the last 10 years, as we got started with real-world evidence, is that a lot of the questions we want to answer require careful synthesis and analysis of healthcare data that comes from multiple different data streams and sources and methods. 

And EHR data alone doesn’t solve the problems that we want to tackle.

Taking the EHR data that we cultivated and linking it to claims and genomics and imaging to map in critical contextual information—e.g., mortality, socioeconomic status, smoking status. There are a lot of values that just aren’t captured consistently or accurately, and you can’t learn from data that isn’t captured.

So, real-world evidence is certainly one component of the broader discipline of integrated evidence, but what we’re working to do is think about how we bring together evidence from across the ecosystem to learn more quickly and efficiently whether it’s captured in routine care, intentionally collected in a trial, or prospective study reported by patients.

And we’re seeing a real need to start to link those disparate sources of data.

CS: We have active partnerships underway now with the U.S. FDA, Friends of Cancer Research, the National Comprehensive Cancer Network, NICE, among many others. 

And those are our research partnerships, where we pick research questions and actually explore these topics together, so that we can collectively learn from one another.

We are also a founding member of the Real-World Evidence Alliance, which now includes 10 organizations working in the real-world evidence field. And we’re working to come together to champion collaboration around important novel applications, and the policies that will enable them.

It’s been exciting to see the momentum in guidance from the FDA, and the same thing is now happening in Europe around HTA applications. And so, we are looking closely at those, and think we can play a meaningful role in providing guidance, because we have firsthand experience looking at these data over many years.

Some of the big wins we’ve seen to date include—and thank you to The Cancer Letter for covering many of these over the years—men with breast cancer now have an approved therapy option in Ibrance. And this is a population which was historically too small and challenging to study in a randomized clinical trial [The Cancer Letter, April 19, 2019].

Similarly, last year, we saw a new dosing regimen approved for patients with EGFR mutations for some cancers, which enabled patients to go in for chemotherapy once every two weeks instead of once a week, which means half as much time is spent in the infusion chair and days back in each of these patients’ lives.

And maybe one last one—I was really excited to see this. We have a partnership with Foundation Medicine (FMI) where we build a clinical genomics database, and their Foundation One CDX genomic test was recently approved by the FDA as a companion diagnostic to identify patients for a couple different indications of Rozlytrek (entrectinib). 

What was interesting here is, first and foremost, this is going to enable broader access to genomic testing and potentially breakthrough therapies for patients. But also as a condition of this approval, the FDA requested that FMI conduct a post-approval study powered by our joint clinical genomics database, which spans over 100,000 patients. 

So, now there is a new companion diagnostic that was approved last month. And we already have the built-in data set that’s going to enable FMI to address those FDA requirements via retrospective real world evidence.

That’s exciting to hear. How has the Roche acquisition informed Flatiron’s work and priorities, and what do Flatiron’s partnerships, research collaborations, and business transactions with other pharma companies look like now, post-acquisition?

CS: We’re an independent affiliate of the Roche Group and we remain a separate legal entity, and that autonomy has always underpinned our work with many life sciences companies, with cancer centers, with academic medical centers, with the regulatory groups. We continue to work independently and our priorities are informed through input from all of our customers.

We have strict firewalls in place and confidentiality obligations that ensure that we can do this work in a connected way across the ecosystem. And this is critically important for us to maintain. We work with all of the top 20 global pharmaceutical organizations that develop oncology therapeutics now. And part of what we learned from them are about the new possibilities that are emerging and the joint use cases that we think are most exciting.

What we’re seeing is that demand for real-world data and real-world evidence is continuing to grow, and we’re seeing across the landscape a continued increase in investment in R&D in oncology. 

At the same time, the drugs themselves—and I think COVID has contributed to a lot of this—are being approved sooner with more limited data.

There’s an increasing need to think about this concept, as I described, of integrated evidence to understand what happens after that first approval. If the first approval determination is made looking at results across 30 patients, that doesn’t tell you how the drug is going to work when we then bring it to a much broader patient population around the world.

And so, we’re seeing that our partners are excited to further explore integrated evidence approaches for post-marketing commitments and to make the case for reimbursement around the world more quickly. These market access use cases are an important area of investment. 

And there’s also a very broad recognition that we need to do better in terms of diversity and representativeness in research and in trials. We’re working with our biopharma customers on both fronts, to both broaden and expand the solution set in how we run clinical trials, and then to think about how we build more diversity and representativeness to all of the research that we do and the decisions that are made.

CS: We’ve learned that it’s going to be important to understand not just U.S. populations, but also a broader global population. 

First, in the context of learning from real-world evidence, and, second, in the context of broadening our approaches for clinical trials around the world.

Today, we have international subsidiaries in Japan, Germany, and the U.K. And we’re in the process of designing custom approaches for the context that exists within each one of those markets.

It’s now been, I think, a three-year investment in understanding the legal regulatory and compliance requirements and forging our first early partnerships. We announced our first partnership in Japan last year to study patients with gastro-intestinal cancers and you can expect more coming in the U.K. and Germany.

Why is this important? There are two million people diagnosed with cancer and 750,000 of them lose their battle with cancer every year across these three countries. 

So, first and foremost, in service of our mission to improve and extend lives for every person with cancer, this is an important focus area.

But then, as you said, when we talk to our biopharma clients, it’s important to them to tackle the challenge of access to high-quality real-world evidence from global patient populations. We are following their lead, following their priorities. And we’re seeing really, really great momentum here.

We also have a three-year partnership with NICE that started in 2020, and we were able to work together most recently to help provide input and feedback on the real-world evidence framework that they just launched.

The goal here is to think about how to provide guidance, so that the best practices for planning and conducting and reporting on real-world evidence studies are clear to everyone involved. The hope is that will then help to improve the quality and the transparency of the evidence and the acceptance of these data sources.

How far along are the real-world evidence frameworks that are being developed in other countries? Is it still primarily driven by the U.S. market and the intellectual advancements here? What is the landscape looking like elsewhere?

CS: There’s a ton of activity in this space. It’s been encouraging to see the momentum in the U.S. and with NICE and ISPOR in particular and across Europe, we’re seeing similar activities underway. I think that the theme and trend is consistent around the world, from what I see. 

We recently had the chance to participate and help shape the NICE Real-World Evidence framework earlier this year. The research coming out of our ongoing NICE Research Collaboration not only helped inform the development of this framework, but has been critical in making greater use of real-world data to resolve gaps in knowledge and driving forward access to innovations. 

We hope to continue partnering with countries to advance impact and use of RWE. 

How has your suite of services grown in recent years? And what should we know about Flatiron’s latest work in AI and machine learning? For instance, is clinical decision support—versus resource, depending on clinical utility—a reality now for users in your ecosystem?

CS: I’ll start with AI and machine learning. ML and AI have been the foundation of the way we’ve designed our products from the very beginning.

In the early days, we pioneered the use of machine learning technology, paired with natural language processing to start to extract clinically relevant information from EHRs and make that more available for research. 

What we started to see very quickly was that that wasn’t necessarily enough to ensure that these clinical data points were captured with the level of transparency and quality that we know are necessary. 

So, we built a clinically-trained abstraction workforce who look directly at the chart and make clinical determinations around, for example, line of therapy or progression of disease, actually looking at the core source data. 

Now, machine learning is getting much better and much more accurate. So, we’re in the process of running a lot of tests to understand what’s the best technique for any given data variable, and it looks different depending on the data field that we’re working on.

The right approach for line of therapy, for instance, might look different than smoking status, or extracting biomarkers or genomic information. And so, we have an opportunity to get much more customized and create the frameworks to properly evaluate and transparently communicate the quality of the data that’s being produced with machine learning.

In addition to using ML for building our cohorts, I would say the other thing I’m excited about is using ML to start to tackle some of these really important needle-in-a-haystack problems that would just be impossible to manually extract from the chart at scale. 

When we have a very rare patient population in question, we can use machine learning to get an early read and quickly answer a specific research question. We have also built ML models to predict patient data fields that aren’t captured well in the EHR in real time at the point of care, so that those data can inform treatment decisions and get to the clinician much more quickly. These data live in the EHR, but they might be buried in a document and are liable to be missed.

What we continue to see is that the AI is really only as good as the humans who are using it and the judgment they apply. There’s the art and the science and we continue to believe that it’s going to be really important to blend both of those.

I’ll highlight one interesting product in particular, called Flatiron Assist. It’s a clinical decision support tool that works in our EHR, OncoEMR, but is also platform agnostic and integrates with Epic, and over time, other large EHRs. 

It pulls all of the important information that you might use to make a clinical treatment decision and then showcases and integrates all of the guideline-concordant therapeutic options and available clinical trials to consider. 

And it also highlights preferred practice protocols and regimens—and unique payer considerations—to ensure clinicians are aware of all of these options and can make the best possible treatment decisions at that critical moment.

In this very complex world, I think that’s a really important starting point. How do we actually make use of all of the data that does already exist?

It also is an exciting opportunity coming back to the vision to start to close the gap between research and care. As an example, the DESTINY-4 trial that was reported out of ASCO this year, which is going to change the standard of care for women with HER2-low breast cancer.

We actually had implemented that early finding and that HER2-low status into the OncoEMR several months before that study was released. 

We think we can play an important role in compressing that cycle and getting new guidelines out to clinicians as soon as they’re available, and then building the right mechanisms to make use of all of the data that may exist, if not in the perfect form that we might like to see it.

CS: Very few patients take part in clinical trials, and it’s well recognized that those who do are not broadly representative.  

Today, clinical trials data is collected in one stream and routine clinical data is collected in another, and they don’t really talk unless humans move them back and forth. It’s crazy when you actually think about how much of the clinical trials workflow still exists on paper logs on walls in the clinic. 

We think there’s an opportunity to use technology to tackle that challenge and integrate these two different research approaches.

You may have seen we acquired a company called Protocol First last year, which has built software that automates the direct transfer of data from EHRs into EDCs, reducing the need for human transcription. The Protocol First suite of solutions is rolled out across many academic medical institutions, and trials with sponsors, and CROs around the world.

We are excited to bring that solution, which is so deeply complimentary to the work we’ve already done in trials management and in the EHR embedded software, together to think end-to-end, how do we map out the entire value chain of what does it take to design a trial, start a trial, find the patients, get them enrolled and make that process digital instead of deeply manual in the way that it is today.

You asked about clinical trials matching. I think many folks are indeed trying to tackle this one, and that’s fantastic. We’re excited to be part of the solution.

The place and the role that I think we can play most uniquely is in thinking about how do we, as I mentioned earlier, use machine learning to understand how to automate that pre-screening process, so that we can not only figure out which patients might be eligible for a trial, and do that in the background, rather than having a human manually go through all of the different inclusion and exclusion criteria.

But then, also build the workflows that put that information at the hands of clinicians at the right time when they’re making a treatment decision. That’s what we hear from our practices has been really challenging.

They might find a patient who might be eligible, but then, by the time the patient is seen and they think about talking to them about a clinical trial, they actually already went on standard-of-care therapy. 

So, they missed that window, and didn’t get to the doctor at the right time of the visit—leading to a big missed opportunity.

We integrate these pre-screening and research team decisions and alerting directly into the clinician workflow, and we think that’s going to really help improve enrollment.

CS: Yes. I think this is a perfect application of that integrated evidence concept we talked about, and we’re starting to get smarter about determining the right approach for each particular question at hand.

With rare patient populations, it may be that a real-world evidence approach and a retrospective study makes a lot of sense, but we also need to think about that in a broader context—that’s a unique decision depending on the specific question at hand.

Rare diseases are one of the important applications that we see for both retrospective real-world evidence and better clinical trials that are faster and more efficient and enroll more simply. But the trial technology we’re building is actually platform agnostic.

We are in trials today, and the solutions, the technology, supports all phases of trials across all settings. 

And so, the same benefit we can offer in rare patient populations we can also offer holistically in lots of different disease areas.

CS: We’ve had a partnership in place with the FDA over the last five years, and we also have a decent amount now of direct experience engaging with regulators on behalf of sponsors, in the context of specific regulatory submissions using Flatiron real-world data.

These are important sources of learning for us, because we can then help bring guidance back to all of the other customers that we work with and better inform the way we are designing our evidence solutions.

One of the research collaboration projects we’re excited about is focused on understanding similarities and differences between metastatic breast cancer patients treated in the real world and patients treated in clinical trials.

We’re effectively producing trial-like populations using Flatiron real-world data to understand how replicable those results are in real-world data. And we’re excited that projects like this can start to advance our shared knowledge in oncology for research and regulatory purposes, and further help to define what the right quality bar and methods are for evaluating fit for use in real-world data.

On endpoints in particular, it’s been an important area of collaboration, because the endpoints in clinical trials are operationally defined quite differently than real-world evidence endpoints. They are captured in very different ways. 

So, the scientific bar for use of any type of data, including real-world data, is really high and needs to be reliable and relevant to the scientific question.

What we’ve seen is that it’s helpful to sit together, to look at the methods, to look at and understand the feedback, and come together transparently around the best methodology to tackle what is inherently really complex and messy data. 

And we’re continuing to pursue this partnership.

CS: One of the great things about real-world evidence is that it is an important mechanism to expose inequities and expose the gaps that exist.

You actually wrote a cover feature story about a hackathon we did back in 2019 that ended up on the ASCO plenary stage. 

At the hackathon, we looked at the difference and disparities in time to treatment in states that had rolled out the ACA Medicaid Act. [The Cancer Letter, June 21, 2019] 

And that’s the type of investment that we’ve continued.

We now have a head of health equity and disparities research who drives our research strategy and cross company priorities in this respect. We’ve had a team most recently focused on building out the data inputs necessary to do the type of research we think is missing. And this comes back to the infrastructure of cancer care.

We found out, for instance, that socioeconomic status information was not captured well in the EHR. And so, we had to take a step back and think about how do we pull the right external markers and validation with full respect to patient privacy and practice privacy, and all of the different considerations that are important to define new variables, that we can then add into our integrated evidence to really start to understand where the inequities exist.

We also have an ongoing partnership with the American Cancer Society, where we are actually funding grants to support research in the study of quality and equity in cancer care and outcomes, called the Real-World Data Impact Awards. 

It’s been cool to see where those grants have gone and the research they have enabled. 

To give a current example related specifically to COVID, one of our researchers wrote a paper that was presented at ASCO in June looking at the increase of telemedicine use among patients with cancer during the COVID-19 pandemic. 

On the one hand, we saw a very large increase in the use of telemedicine in cancer care, which historically has been a place where that was very, very low.

On the other hand, we saw that there were pretty substantial inequities in where that telemedicine was used. And so, Black, uninsured, non-urban, lower socioeconomic status patients were less likely to be able to use telemedicine services.

We certainly don’t have all of the answers, but taking a step back, I hope that we’ll continue to see coverage of telemedicine services be permanent instead of tied to the emergency health declaration that came out via COVID, and continue to increase the reimbursement rates for these services from the commercial payers and the private insurers.

Thinking about everything we’ve talked about over the last 45 minutes, what are we looking forward to next?

CS: Thank you so much for having me, Matt. This has been a ton of fun. 

Reflecting on the last 10 years of Flatiron, I’m so proud of the progress we made, the milestones we talked about, the patients who have access to new therapies as a result of all of the work that we’ve done across the ecosystem to understand their stories.

Looking forward, I think we’re really at an inflection point, and the real potential is still very much ahead of us. We are, at Flatiron, deeply focused on reimagining the infrastructure of cancer care, so that we can come together across the ecosystem and accelerate this learning and do it more efficiently and more sustainably.

I think if we can do this, we will be on a path to accelerate the approval of new medicines, accelerate the understanding of who should receive those medicines, accelerate ensuring that they can be reimbursed around the world and then ultimately ensuring better patient outcomes for all of the people experiencing cancer today.

That is a problem and a vision that gets me really, really fired up. And I’m just really excited to be part of this journey.

Thank you for taking the time to speak with me.

CS: Thank you.