An integrated, interoperable system of electronic health records should not only facilitate access to medical records, test results, and imaging, but also has the potential to provide oncologists with new knowledge to make better treatment decisions. By constantly aggregating real-time patient data on treatment efficacy, toxicities, and drug interactions, what is now being called a Rapid-Learning System (RLS) for cancer care should help to drive scientific discovery as a natural outgrowth of patient care.
The National Cancer Policy Forum of the Institute of Medicine held a workshop last fall examine the elements of a such a system for cancer care, including registries and databases, information technology, clinical decision support, and policy issues and implications.
An Early Release article in the Journal of Clinical Oncology (Epub ahead of print: Jun 28 2010: doi:10.1200/JCO.2010.28.5478) reviews the findings of the IOM report and describes the rationale for an RLS system for cancer, how it will work, and its potential to improve patient care.
Lead author Amy Abernethy, MD, Associate Professor of Medicine at Duke University and Director of the Duke Cancer Care Research Program, noted in an interview that it may now take between seven and 15 years for new discoveries to reach clinical practice.
Furthermore, the JCO authors (which also included Lynn M. Etheredge, Patricia A. Ganz, Paul Wallace, Robert R. German, Chalapathy Neti, Peter B. Bach, and Sharon B. Murphy) wrote that there are pervasive problems in quality of care and large variations and disparities in care. “The promise of personalized medicine cannot be attained until we can match individual patient characteristics with detailed evidence available in real time at the point of care,” they said.
The Rapid Learning System has the potential “to help bridge that gap between the research engine and the clinical processes, so that research is directly fed into clinical care and clinical care informs research,” Dr. Abernethy said. “Fundamental to all of this is linking to data.”
As explained in the IOM workshop report, innovations that will be needed for rapid-learning health care to work and to improve care include developments in health information technology such as cloud computing, analytics, user interfaces, and interoperable and secure electronic data transmission and exchange. Substantial federal investment, of course, is already spurring development of electronic health records (EHR) for all Americans.
Some useful elements already exist, especially databases—for example, Medicare data sets and the Surveillance, Epidemiology and End Results (SEER) and National Program of Cancer Registries (NPCR) databases. Unfortunately, the SEER-Medicare dataset has a four-year lag time for linkage, making it presently inadequate for “rapid-learning.” Linking clinical datasets with administrative ones, such as all-payer claims data, will allow rapid quality-reporting. Central to the rapid-learning model is a fast and secure system of data transmission, storage, and access.
OT caught up with Dr. Abernethy while she was in an airport on her way to China for a cancer conference. She used the analogy of an automatic teller machine to illustrate rapid, trustworthy data interchange. She said she will insert her bank card in China, and the ATM will issue local currency (yuan renminbi), debiting her bank account at home. “That magic happens because of the capabilities of linked and reliable data and secure transmissions going back and forth across the world.”
A Rapid-Learning health system, too, will depend on linked and growing databases. But whereas the ATM transaction is a single event, an RLS will form an ongoing, cyclical process. “The care of the patient sitting in front of me is informed by all the people who come before her who look like her, and her care is reinvested into the system to then inform care of people who look like her in the future,” Dr. Abernethy explained.
Basically, this patient-centered system is an iterative process of data gathering, analysis, and application that constantly “learns” as it goes. According to Dr. Abernethy and coauthors, it will collect patient data in a “planned and strategic manner” from the EHR and analyze the captured data. It will generate evidence from these retrospective data, as well as from prospective studies, and then produce new insights into subsequent clinical care. Finally, by evaluating the outcomes of changes in clinical practice, it will generate new hypotheses for investigation.
Great Value, Great Challenges
ASCO President George Sledge, MD, Professor of Medicine at Indiana University Simon Cancer Center, said that he sees great value in an RLS. But this great promise also carries great challenges, he said.
An RLS will allow rapid aggregation of large amounts of data, analysis of practice patterns, the creation of hypotheses, and a movement toward better care. But, he warns, “sometimes large collections of anecdotes are still collections of anecdotes”—great at generating, but not testing hypotheses, which will require well-designed randomized clinical trials.
On the other hand, Dr. Sledge says that large parts of modern medicine will never have Level I evidence since there are too many niches in typical practice to test with RCTs.
“I think for many of those smaller areas, this is going to provide us some real information and real data that hopefully we'll be able to use wisely,” he said. “The infrastructure to do this rapid-learning is certainly likely to be there sometime in the next few years.” He predicted that everyone in the United States will have an EHR within five years.
Dr. Abernethy said the IOM sees rapid-learning health care as being the pinnacle of the evidence process. Today, information coming from rapid-learning approaches falls into an observational cohort study classification—“kind of mid-range on the evidence hierarchy,” she said. As information becomes more reliable and “as bias starts to get out of the system you'll start to see this move up that [evidence hierarchy] ladder.”
Already on a smaller scale, ASCO's Quality Oncology Practice Initiative (QOPI) has implemented bi-directional information flow, using ASCO's quality guidelines to inform practice, with clinical sites reporting back on their practices (but not on outcomes).
Similarly, Kaiser Permanente uses its health information technology to improve cancer care for the 40,000 patients diagnosed with cancer each year within its enrollee pool of more than eight million people. As the JCO paper notes, the Kaiser model incorporates many aspects of a rapid-learning system, using universally accessible and standardized patient treatment plans, pathology staging system, referral to clinical trials, quality reporting, assessment of practice patterns, and detection of adverse event patterns.
Overall, an RLS is expected to guide treatment of individual patients based on aggregated data from thousands of similar patients. Dr. Abernethy and her coauthors said that insights may be uncovered through such a model that are not available in current practice, where clinical data are not gathered and systematically analyzed.
Some examples of potential benefits are an understanding of:
- Clinical effectiveness for the advanced elderly.
- Effects of multiple comorbidities.
- Effects of polypharmacy (where such subjects are frequently excluded from clinical trials).
- Comparison of anticancer drug combinations and their timing, discovery of unexpected drug responses correlated to particular co-morbidities, correlations of outcomes with epidemiologic, ethnic, geographic, or performance parameters, and historical experiences of similar patients to guide treatments—e.g., for a pregnant patient with Stage IV breast cancer. Of course, no matter what an RLS produces and says, all treatment decisions will have to be made in the context of each individual patient and his or her preferences.
Part 1 of a 2-Part Article. Next time: the feasibility and acceptability of a Rapid-Learning System for cancer care.
IOM Report Online
The Institute of Medicine workshop report is available at http://books.nap.edu/openbook.php?record_id=12868)