In a talk I gave in 2009[1], I addressed the challenge of turning the enormous quantities of digital data that will be produced by EHR systems into “reliable, usable” information and emphasized the importance of creating new models of medical research and statistical techniques for dealing with these large universes of new outcomes data. I went out on a limb at the time –since my training in statistics was in business school, not medical school—but have since found many who support my position among the best and brightest in the medical community.
Currently we use the term “translational medicine” for the process of taking the results from research studies, typically Randomized Control Trials (RCTs), and transforming them into data that can be used in a clinical setting. The process includes writing articles for publication to describe the research findings in medical journals (after peer review), which then get further interpreted by clinical teams at medical societies/associations/public health authorities to produce guidelines or other clinical applications. At the same time, published research results get disseminated in the mass media by healthcare journalists. Many of the difficulties and inadequacies of this approach have come to light as the health IT infrastructure in the US struggles with integrating clinical evidence into electronic health records (EHRs). The two toughest issues are: 1) the difficulty of extracting data from journals (print or electronic format) because the journal format wasn’t designed for data extraction, integration or machine analysis and 2) the lack of collaboration between research institutions that leads to an enormous number of research results that don’t build on previous studies and are sometimes contradictory. As a result, even elaborate data mining engines—or a redesign of journal formats—won’t by themselves solve core problems with our current research system.
Beyond lack of collaboration, there is a more fundamental issue about the appropriateness of RCTs as a source of clinical guidance. Marya Zilberberg, MD, MPH writes in JAMA that we have an “avalanche of published research” yet the “data generated in an RCT are frequently irrelevant because of their limited generalizability” [2]. The cause of the limited generalizability is the “hetereogeneous treatment effect” and I won’t go into details of what that means here; for that, read Dr. Zilberberg’s blog series on Reviewing Medical Literature, which is now at part 5 and is highly recommended reading. But, as you can probably guess from the word “heterogeneous”, the problem relates to the fact that we don’t all respond to a specific treatment in the same way.
Another recent article published in New England Journal of Medicine (NEJM)[3] buttresses my position that we need new research models and new incentives for collaborating. This article presents some stark data on how little emphasis – and financial support – has been placed on outcomes research, best practices, new care models, quality, comparative effectiveness or service innovations relative to total biomedical research (0.1% v. 4.5% of total health expenditures). The 0.1% will increase to 0.3% in 2010 according to the authors, but that’s still a small percentage considering that we are facing a huge pool of new outcomes data that will be generated by EHRs and haven’t yet agreed upon the best statistical methods for organizing and analyzing these collections of data that hold so much promise for enhancing the clinical value of medical research.
There are models that have been developed for analyzing outcomes data, especially within the payer segment. Registries of patient data have existed for some time, but have been limited in scope or the reliability of the data sources (e.g., claims data). But, in order to achieve a “higher standard” of clinical value, more collaboration and “development of more robust analytical techniques for ascertaining clinical value”[4] are needed.
This emerging field offers opportunities for data publishers that understand how to apply master data management (MDM) principles and data analytics/predictive analytics companies in biomedical research that can adapt their models for clinical applications. Perhaps the biggest opportunity is for standards organizations and data publishers to build a collaborative infrastructure for better aligning biomedical research and clinical decision support systems.
[1] http://www.slideshare.net/janicemc/epatientconnections2009-health-content-for-epatients
[2] The Clinical Research Enterprise; Time to Change Course?, JAMA, February 9, 2001—Vol 305, No. 6, Marya D. Zilberberg, MD, MPH,: http://jama.ama-assn.org/content/305/6/604.short.
[3] Biomedical Research and Health Advances, NEJM, February 9, 2011, Hamilton Moses, III, MD, and Joseph B. Martin, MD, PhD: http://healthpolicyandreform.nejm.org/?p=13733.
[4] Ibid.