I attended a fascinating Health Affairs (www.healthaffairs.org) briefing on “A Rapid-Learning Health System” this past Friday, January 26 here in Washington. The project was led by Lynn Etheridge and Health Affairs and sponsored by the Robert Wood Johnson Foundation (www.rwjf.org), Kaiser Permanente (www.kp.org), and the US Agency for Healthcare Research & Quality (www.ahrq.gov).

Your first question may very well be, “What the heck is ‘rapid learning’?” The vast real-world databases created by electronic health records (EHRs) maintained by integrated delivery systems such as Kaiser and the Veterans Health Administration (VHA) create a phenomenal research capacity. With literally tens of millions of longitudinal, clinical member/patient records, the combined power to understand the effect of all kinds of care practices is staggering.

As Kaiser’s Paul Wallace (also the IxCenter Board Chair) pointed out, the number of newly diagnosed cancer patients in Kaiser’s EHR each year (about 40,000) is roughly the same as the number of patients enrolled in US cancer clinical trials. As Geisinger’s Buzz Stewart wrote in the Health Affairs Web exclusive, there is a clinical trials also deal with “clean” populations (often excluding the “messy” patients with multiple co-morbidities). EHR databases can help to bridge this “inferential gap” to help us figure out what to do about those people with conditions for which the traditional scientific process doesn’t provide a good answer.

Perhaps even more important than the contribution that this database can make to helping to heal people with new cancers is the impact it could have on the woefully understudied issue of cancer survivorship (understanding the health impact of “cured” cancer on survivors years or decades later). When the clinical trial is over, researchers often stop collecting data on their “subjects,” but Kaiser has a quarter of a million longitudinal EHRs on cancer survivors. That could have a great impact on our ability to address unexplained health issues that arise from the intense therapies to which people with cancer are subjected.

But answering these kinds of questions are just the tip of the iceberg. As Archimedes (www.archimedesmodel.com) Co-Founder and Chief Medical Officer David Eddy (also one of the pioneers in evidence-based medicine) noted, while tremendously valuable, applying the “look up” method to EHR databases tells you a lot about the past and the present, but it can only tell you so much about the future if nothing new happens. This is where the miracle of modern mathematical modeling comes in.

Now I’m not going to embarrass myself by trying to explain how models like Archimedes’ work, but there are a few critical points to understand. First, there are powerful representational modeling techniques now used in every industry from entertainment to transportation to architecture, and there’s no reason why—with adequate investment and data sharing—we can’t do the same in health care. Second, these models employ techniques to integrate data from much more discrete components (such as the progression of disease on the physiological level). Third, because of that, these models have a tremendous capacity to assess virtually everything that can happen (depending on the data that we have). Just to give you an idea of the potential power, Eddy and colleagues have prospectively predicted the outcome of many clinical trials (not that he’s saying that we should just get rid of RCTs)—for an example, see Exhibits 1 and 2 in the Health Affairs January 2007 Web exclusive by David Eddy.

With EHRs, there suddenly is a vast expanse of new data that can be integrated into models like Archimedes. The combination of EHRs and sophisticated representational modeling techniques can, to paraphrase Eddy, “put rapid learning on turbo.”

As amazing as these models are at this point, they still may not help us address some of the major quality-of-care gaps in the US if they rely only on the clinical and physiological data that currently power them. We know that one of the critical reasons for poor performance on quality measures is our inability to inspire healthful behaviors. For example, we know that a substantial portion of mortality and morbidity in the US are due to three behaviors: smoking, poor diet, and lack of adequate exercise.

Luckily, the science of behavior change is evolving, and with it our ability to understand how to effect positive behavior change. What if we integrate the models developed by people like David Eddy with the behavior change science developed by people like Jim & Jan Prochaska? (In case this is new to you: The Prochaskas have not only done pioneering work on the transtheoretical model but have developed a series of science-based tools for effecting behavior change.)

At the briefing, I asked Eddy if this was possible. He remarked that, as long as we can measure it, we can integrate virtually anything into these mathematical models. Indeed, new measures of behavior change and patient activation have been developed by people like the Prochaskas and Judy Hibbard. Others, including those in government health programs, are beginning to explore that measurement arena as well.

The opportunity not only for rapid learning, but for truly transforming care is enormous. We need to push the envelope on the scope of our inputs to EHR databases. At places like Kaiser, Group Health Cooperative, Geisinger and the VHA, they already have the opportunity to tap into rich sources of patient-reported data (such as health risk assessments, secure messaging, and other online applications where consumers enter in personal health information—all of which, of course, needs to be protected as with any other human subjects research). Let’s find ways to move this agenda forward.

–Josh

This entry was posted on Sunday, January 28th, 2007 at 10:03 pm and is filed under Behavior Change, EHRs, PHRs, Research. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.