The increasing focus on clinical measurements, the need for large data sources, and population health management have raised new questions that are particularly pertinent to cancer care, such as how to organize the data in a meaningful way, and how to translate the “big data” to the individual patient. Amy Abernethy, MD, PhD, Chief Medical Officer and Senior Vice President, Oncology, Flatiron Health, NY, had spent years at Duke University’s Cancer Institute and has recently joined Flatiron Health with the goal of finding solutions to these questions. Oncology Practice Management (OPM) asked Dr Abernethy to discuss the implications of big data for oncologists and oncology practice administrators.
OPM: What does big data mean to you, and where do you see big data making the most profound impact in oncology?
Dr Abernethy: When I think of big data in healthcare, I mean rapid accumulation of increasingly more varied information. This is important in oncology, because when we think about the 360-degree view of the patient, it is not just the tumor, or just the cancer drugs. It is the healthcare system as a whole, the values of the individual, as well as molecular issues, such as genomics and proteomics. Why does big data matter in cancer? Because cancer is complex, and it spans the gamut from the very fundamental biological aspects of a patient’s disease all the way through the societal, environmental, and healthcare system delivery aspects. To be able to get a 360-degree view of the patient, we have to figure out how to pull the data together all around 1 particular person. One well-organized and well-curated data point can be used for multiple tasks simultaneously. It can be used to optimize healthcare and the delivery of cancer care. It can be used for clinical decision support to improve decisions between doctors and patients. It can be used for discovery and for new research, for learning about what works for whom, and when. When I think about the value of big data in cancer care, I see it as the combination of the accumulation of data that are organized and structured in a way that are readily analyzable. Then comes the question of what to do with big data—how to analyze it and make sense of it. The metric that I use is how many questions can be answered with 1 data point.OPM: You recently became the Chief Medical Officer for Flatiron Health in New York City. Can you briefly tell us about Flatiron Health’s mission?
Dr Abernethy: Flatiron Health is trying to organize the world’s cancer data to make the data useful. When we get to the point where 1 data point can solve multiple problems simultaneously, a fundamental gap in getting there is how to organize the data, and making sure that we have all the data points we need to tell the full story of cancer. Flatiron Health is trying to solve that particular problem. We do that by partnering with providers through technology solutions that allow them to do their work. For example, Flatiron’s OncoEMR is an electronic health record (EHR) system used by nearly 2000 cancer care providers in the United States. We also have a population health management tool, OncoAnalytics, which providers can use regardless of what EHR system they are using. Either tool brings data into a central repository, where the data are processed to create very rich, complete data sets that are prepared to answer the kinds of questions cancer care providers prioritize. The mission of our company is to fight cancer through data. For that we need to organize the data. I have been trying to do this for the past 10 to 15 years. I used to practice how can I take 1 data point and put it into as many analyses as possible. My metric used to be, “How many papers could I write with 1 data point?” I did that to find out what it takes to make that single data point useful. What happens if we now overlay the concept of big data? How do we make all the different data elements useful? One of the things I figured out is that this is a “tech problem” as much as it is an “analytic problem” or a “research problem.” That’s why I came to Flatiron Health. Here at Flatiron we are merging the technology, healthcare, and clinical research—with the patient and cancer care provider as our core focus.OPM: With your experience in oncology and palliative care, what prompted you to shift your focus from the university to the tech setting?
Dr Abernethy: Being an oncologist and a palliative care expert makes me respect the needs of patients and families first. I am also formally trained as a cancer informaticist, which leads me to ask how we can do clinical trials and outcomes research in the setting of big data. I have been fortunate to see the story from different directions, and I am convinced that it could not be solved inside of academia. Inside the university system, it was hard to prioritize people of all backgrounds and give them equal standing and pay. Unless the software engineer, professor, analyst, and oncologist can be treated as equals at the table, it is hard to get this work done. I was also involved with the story of big data in cancer from other angles. At Flatiron, we recognize that it takes partners coming together as equal colleagues at the same table to solve the problem. Inside the company we bring together people from many different industries to attack the question from all sides. Similarly, the American Society of Clinical Oncology’s CancerLinQ, the National Cancer Institute (NCI), IBM, and others all bring important components to the table, but none is sufficient by itself. Only by working together can we find solutions. At Flatiron, our core customer is the doctor and the patient sitting in front of her. We have to figure out, “How do we first solve problems for doctors and patients?” Then we back our way into the big data story.OPM: When did you realize that big data have an essential role in oncology?
Dr Abernethy: From 1999 to 2003, I ran a large pragmatic clinical trial in Australia evaluating coordination of care. This study was similar to the current Centers for Medicare & Medicaid Innovation demonstration projects. In fact, the study that I ran was similar to what we are now calling the Oncology Care Model. We were looking at 2 different care models for patients with advanced or metastatic cancer and other life-limiting illnesses. Palliative care and care coordination were key features. Patients were randomized to care model version A versus version B, and the study involved all eligible people from a huge geographic area in South Australia. That is important, because I had a very large geography, with a huge population. We had 1.6 million people in our state, and through the trial we touched approximately 350,000 people. I needed to deliver what looked like the Oncology Care Model, including care planning, and help for nurses, patients, families, and doctors. In palliative care more than in any other part of the cancer care spectrum, a huge part of our important data sets comes directly from patients and families. Symptom control can only happen if we ask people about their symptoms, which means that we need to collect data directly from patients, who may be sick and are at home. They cannot come to the healthcare location and answer questions in our clinic, so we have to go to them. At the same time I was managing patients. On Mondays, Wednesdays, and Fridays I was responsible for this huge clinical trial, and on Tuesdays and Thursdays I saw patients. I came to realize that if I could figure out how to use the research data I collected on Monday for my Tuesday patient visit, I could refocus my Tuesday visit on the patient exclusively. The clinical bar, unfortunately, is usually not as high as the research bar. That is why researchers do not want to use clinical data, but what if I could raise the clinical data bar to the level of the research bar at all times? And what if I could figure out in the clinical setting how to actively collect research data? I had to gather and organize the data from across the state of South Australia, from many different sources (patient, clinic, insurer, government). If I was going to use it for clinical care, it needed to be available in real time—data gathered in the moment needed to be ready for use immediately. If it was going to be appropriate for palliative care, it had to include information gathered directly from patients and their families at home. It became clear to me that I had to solve the data problem first, to take better care of patients and generate better evidence of care.OPM: How is the trend toward the collection, organization, and stratification of data impacting patients and physicians at the point of care?
Dr Abernethy: Let’s talk about near-term, midterm, and long-term. Near-term right now, by having an EHR system such as OncoEMR, we can get the information back to doctors and patients in real time to solve real clinical problems. For example, within OncoEMR we can give doctors information about clinical pathways programs or better decisions about biomarker tests that are informed by the type of data we collect. The second midterm answer is that as we develop and do research as a by-product of the data, we can bring the answers back to the EHR system and at least make sure that doctors are armed with that kind of information. In that midterm, we can also help doctors and patients access, for example, new research such as clinical trial findings or the results of our big data analyses—new evidence that helps to move patient care forward. In the long-term, all this ultimately can be compiled into smarter algorithms for clinical decision support at the point of care—computers that surface exactly the information that we need at the time doctors and patients are making decisions together.OPM: Does the risk of missing information keep you awake at night?
Dr Abernethy: One of the core things that Flatiron Health is working on solving is that much of the critical background information—the key nuggets that we need to take care of patients—is trapped in those case notes and patient files. These data points have made it to the EHR system in some kind of electronic picture, whether that’s a scan, a PDF, or whatever else. This means that the information is not readily digitized and available for analysis. We are using a process called “technology-enabled abstraction,” whereby our abstraction team can pull out those core nuggets from the data by hand, but also at scale. In essence, we are seeing that the type of data points that have historically been missing from digital stories can be found by using a combination of technology and hand review of clinical charts.OPM: Oncologists are asked every day to enter data into EHRs to comply with various programs or reimbursement requirements. Is this information being used effectively?
Dr Abernethy: At Flatiron, we do not believe in asking doctors to enter more and more information into the EHRs unless it will specifically help improve the care of the person sitting in front of you right now. We believe that it is better to let doctors take care of patients as they always have. We should not put the EHR as a barrier to patient care, but rather make the EHR be something that creates easier access to patients and patient care. Have other people do what other people can do, and only ask the doctor to do what only a doctor can do. For example, when it comes to information about pain, the best person to report on pain data is the patient. Don’t ask the doctor to report on it. By contrast, when it comes to the question of, “Why did you make that treatment decision?” the only person who can answer it is the doctor. Information that doctors don’t enter but is otherwise available in the medical record can be turned into digital data points using technology-enabled abstraction. For example, information about a brain metastasis—the doctor probably wrote in her notes that “the patient has brain mets,” and the information is in a radiology report too—the appropriate data point gets completed in the data set by an abstractor who reads the chart and fills in a form, and the doctor is not responsible for the data entry.OPM: There are countless EHR systems. What will it take to facilitate the exchange of information from one system to another?
Dr Abernethy: It is all about corporate policies. Interoperability fundamentally requires EHR vendors and the healthcare organizations that implement them to believe in talking to each other. Regulators can try to push it along, but fundamentally technologists and health systems have to allow it. It is true that some patients feel uncomfortable with the exchange of private health information, but, for the most part, patients want their information available to each doctor providing their care regardless of which health system the doctor works in. For example, imagine you own a big health system in central North Carolina that competes with another big health system 30 minutes down the road. You want to keep your patients in your health system, the 3 million patients and 3 million people who are in your geographic area; the way to do that is to make it so that your health system cannot talk to the health system next door. So there is a health system–related reason for the lack of interoperability. Meanwhile, if you own an EHR, building technology features that allow your EHR to talk to another requires engineering resources that must be diverted from other tasks. In addition, it is not necessarily clear how EHRs can talk to each other safely while maintaining all the right security and privacy standards. And aligning Mary Smith in 2 different electronic systems can be tough, because we don’t have a national system of unique identifiers. These are the kinds of problems that have to be solved. But they must be solved.OPM: Why should practice managers or oncologists be concerned with big data?
Dr Abernethy: First, it is important for them to have some general knowledge of the topic, because everybody talks about it, and it is so abstract in many ways. In addition, data mean power. Practice managers, reimbursement specialists, and oncologists—the smarter they are with their own data, the more power they have in making good decisions. That power comes in multiple ways. Sometimes it is the power of making a smarter decision between you and your patient. Another way is it gives you power to not make mistakes; for example, billing errors are some things we see all the time. At Flatiron, we are now starting to find billing errors and show them to practice managers. At the moment it is kind of funny, but it is one of the most celebrated things we sometimes do. Financial counselors are trying to help patients navigate the path to avoid bankruptcy. At the end of the day, if oncologists do not have data about critical information such as the efficacy of a drug, and whether the patient’s insurance company covers it, how can the financial counselors do their job well? The whole thing has been predicated on some guesswork, and the data add a level of transparency to interactions that we have not had before. Finally, as our system of cancer data gets organized, research and analyses about what works and what doesn’t will become more instantaneous. Ultimately, data help us make more confident decisions, and that is what we really want.