Big data has the power to significantly speed up the process from discovery to clinical adoption, according to a presentation at the Association of Community Cancer Centers 2015 Annual Meeting in Arlington, VA.
Amy Abernethy, MD, PhD, Chief Medical Officer and Senior Vice President of Oncology, Flatiron Health, New York, NY, and Director, Duke Cancer Care Research Program, Duke Cancer Institute, Durham, NC, first described an encounter she had in 2009 while working in a melanoma clinic with an emergency department nurse named Janet, who presented with Stage IIIB melanoma.
“We could predict her risk of death pretty well. I could say, ‘Well, you’ve got red hair and you look kind of Irish, but I don’t even know the number of patients in clinical trials who had backgrounds similar to yours or personal characteristics that are just like yours,’” recalled Dr Abernethy. “She wanted to help others. But her story—because she wasn’t in a clinical trial or an observational study—does not get entered into the data record forever for the future, nor can all of those stories of taking care of patients like Janet inform the care of other Janets in the future.”
In addition to this challenge, she noted other data challenges, including the fact that electronic health records don’t sync up with each other, and that even within electronic medical records (EMRs), much of the data is unstructured. As conversations continue across healthcare about how to optimize healthcare delivery, improve quality and effectiveness, and tap into personalized medicine, Dr Abernethy said, a common theme emerges.
“Fundamentally, it’s about creating a foundation of linked, patient-level data that can be analyzed and reanalyzed, used and reused, to solve all these problems simultaneously,” Dr Abernethy said. “We’ve had this conversation for quite a while. How can we build this data substrate, this foundation of consistent information that we can draw upon?”
To help solve this problem, Flatiron has partnered with community oncology practices to have all patient-related information stored in a common, cloud-based database called OncoEMR. In addition, the company’s OncoAnalytics platform can be used to gather and parse clinical practice data. Approximately 220 cancer centers use these tools, with 1750 clinicians and 725,000 patients with active cancer in the Flatiron Health Cancer Center Network, she reported.
The systems normalize and harmonize structured data into a common data model to assemble consistent patient information for each of the many patients in the network. They also uniformly cull and categorize the key pieces of unstructured information for each patient.
“This process then allows us to pull out key data points that, as cancer care providers, we know we need to have; biomarkers, staging information, performance status, smoking status,” noted Dr Abernethy. “We call that, at Flatiron, technology-enabled abstraction.”
Users could then “build [complete] individual patient journeys that reflect what happened to the individual across time,” she said. For example, in a set of patients with melanoma, adding unstructured data abstraction to information gleaned from the EMR resulted in 100% completeness of the records for whether the patients had a metastatic diagnosis, their testing status for BRAF mutation, and the date of that testing, along with results from those who were tested.
The software suites also can enable higher accuracy in, for example, finding patients with a particular diagnosis in a patient cohort. A search can first be performed for patients with relevant codes from the International Classification of Diseases, Ninth Revision; chart data gathered via abstraction can be added to find other patients with the same specific condition. This approach yields lower rates of false-positives and false-negatives, according to Dr Abernethy.
“Getting to a right group of patients and having data to help discern the right cohort of patients is critically important for all of the questions in front of us, not just academic questions,” she added.
One example is her team’s effort to bring together experts from many disciplines to map patterns of cancer treatments, including which medications are used most commonly as first-line, second-line, and third-line therapy, she said. They can also refine the analyses by adding data, for example, on which patients receiving each type of treatment have mutated or wild-type BRAF.
Another facet of the technologies’ potential is to create dashboards and reports for practice management. These can highlight such information as how many patients have not yet had their tumors staged, or how many patients in each practice location are on treatments and what the associated charges and revenues are, said Dr Abernethy. Practice administrators could also sort the patient population by individual diseases and the treatments being used for them. She added that “unstructured data could be overlaid on top of that” structured data.
The data can also be used to facilitate quality monitoring (eg, examining which patients are receiving the appropriate treatments based on their individual characteristics, including risk factors for adverse events and contraindications). Another use of the data is to support “discovery sciences” such as bioassessments and clinical annotations, and also to help practices identify which patients might be eligible for particular clinical trials.
“My point today is that Janet gave me a very, very strict task in 2009; she said, ‘Don’t let my story go away.’ I think that one of the things that we all see in clinical practice is the stories of the patients that we’re caring for shouldn’t go away, nor the work that we do, nor the assessment of the work,” said Dr Abernethy. “By partnering with you, and with more and more insights about what tools will make us smarter in the future, we’ll be able to do this more and more.”