How to accelerate knowledge adoption in healthcare with data
Translating information into changes in medical practice traditionally has taken years, but the pieces are in place to speed up that critical process.
Massive medical research efforts yield insights daily for medical care. But historically, changes in medical practice don’t occur quickly. This is rooted in history over hundreds of years, but even recent discoveries in medical care take a long time to reach common usage.
Just as a recent example, the Food and Drug Administration declared Vioxx safe and effective in 1999, and it subsequently became a blockbuster drug for Merck. Then the larger medical community started to become aware of scattered reports of cardiovascular complications due to the drug. In 2004, the evidence of cardiovascular risks became overwhelming, and the drug was removed from the market. But in those five years, it is estimated that Vioxx caused tens of thousands of myocardial infarctions, hundreds of thousands of other types of cardiovascular disease and similar numbers of deaths. What if there had been a mechanism to discover this more quickly? How many lives could have been saved?
The situation is ripe for technology that can gather information across large numbers of patients and can enable clinicians and researchers to look for patterns that can speed up the process.
The knowledge adoption problem
Why is knowledge adoption such a problem in healthcare?
In 2003, the Institute of Medicine noted that it typically took 17 years after significant discovery to adoption into routine practice. This contention was sourced from an article two years earlier with this quote (Balas, 2001): "Relying on the passive diffusion of information to keep health professionals' knowledge up to date is doomed to failure."
In addition, actually changing care and practice patterns isn't based purely on hard data or well publicized publications – it requires the community to achieve the difficult concept of consensus. Handler, in 2004, noted that "consensus works when it takes something that ‘everybody knows’ and makes it something that everybody agrees to actually do."
And the highly recognized Larry Weed, MD, noted in 1997 that “Medicine lacks an information infrastructure to efficiently connect those who produce and archive medical knowledge to those who must apply that knowledge.”
These should raise questions within the community. How can we get past the tendency to allow knowledge to diffuse passively? How can we more rapidly get a critical mass of consensus? What can we do to create this connective infrastructure?
We can cite typical "change management" or "inertia" as the primary issues, but these three quotes identify an infrastructural issue. We have a fundamental problem with information and knowledge discovery and diffusion. If you think about a key medical "discovery" or evidence that becomes established, the provider has a certain set of patterns and habits with which he or she is comfortable that allows absorption of new findings into practice. However, the problem with feedback mechanisms related to these discoveries is that they are either infrequent or temporally removed from prescribing events.
One effective change mechanism is getting feedback from peers. But what if the colleague is not familiar with the paper or is similarly inertially impaired? The issue isn't purely knowledge generation – presentation and “consumability” are just as important to speed adoption.
While our new digital EMR landscape has some well-publicized warts, we can improve the way we exploit and expose those data and information. And while we have lots of data, not enough of it is truly accessible in a way that can create that information infrastructure to which Dr. Weed aspires. Perhaps new digital systems can be used to leverage our natural inclination to change with observation and simultaneously facilitate these “observations” with a visualization foundation that is based on data.
Promises and pitfalls
This is where initiatives such as Cosmos can play a key role as a connective infrastructure. However, anyone who has actually looked for answers to questions in Cosmos (or other similar databases) knows that the path to these answers is not particularly intuitive.
For example, much of the raw data in Cosmos is entered in as codes or other vocabularies that are not easily understandable or accessible. But I found some Cosmos filters that are highly accessible and meaningful.
For example, the COVID-19 filters are quite useful, and I found that their contextual and semantic meanings are very easy to understand and access. They were not raw diagnosis or claims data. They had been processed by the Epic teams by combining multiple pieces of data into more meaningful and intuitive information. Instead of having separate pieces of data housing hospitalizations, deaths and COVID diagnosis, they were combined to be more intuitive and meaningful.
We need more filters like these to navigate and make these massive databases easier to use. We need to move up the DIKW pyramid (data, information, knowledge, wisdom). Making the jump from information to knowledge is far easier than creating and abstracting information from raw data then using that information to create new knowledge. Meeting end users halfway and presenting information derived from raw data is the key step in democratizing tools like Cosmos.
A vision for the future
In the near future, what if a primary physician in the heartland without any academic experience notices a future Vioxx causing an odd predisposition to a bad side effect, then could run a Cosmos query to confirm (or remove) suspicion and potentially raise the alarm This physician could then socialize the findings and get a groundswell of support and changes in practice based on well-curated data.
The concept of Epic’s "patients like mine" is a close cousin of my fluoroquinolone queries – my curiosity was triggered by studies (“knowledge”) that was presented to me, and I was able to explore, confirm and, as a result, change my practice. The medical community needs such analysis tools to be far more democratized and accessible.
With Cosmos in particular, the medical community needs more filters like the COVID filters that are far more semantically meaningful and accessible for end users. More of these filters will enable front-line clinicians to ask meaningful questions and get meaningful answers. The key would be to facilitate a vast library of curated, abstracted data. Imagine if, in addition to these filters, others were available – reason for not vaccinating; contraindication for monoclonal antibody; rapid reduction in oxygen needs; and long COVID.
Won’t that be hard?
That level of abstraction requires lots of effort, but we are doing it already. We all have small armies of abstractors already pulling such data and information nuggets out of our systems. The exact details of the process vary from organization to organization, but the general themes are consistently similarly disjointed and inelegant.
Beyond the laborious workflow, the more important fact is that we are basically donating enormous labor and giving that labor away to accrediting organizations and specialty societies. Yes, these highly curated, highly valuable data points are being collected, but typically are lost to the organization that generates them because these data points are siloed in the databases of the accrediting and specialty society databases. They are not only siloed from the generating organizations; they are also siloed from each other.
Additionally, data are being dumped into outside databases that are siloed from other data. This is an enormous lost opportunity because the real power of these data sets isn't in the sets by themselves. We have lost the opportunity to intermingle these disparate datasets to rapidly generate new knowledge and insights.
A better way
A few years ago, I saw Epic’s chart abstraction process as the seed of a better mousetrap. Instead of the workflow gymnastics required by having abstractors jumping from patient lists to Excel spreadsheets to patient charts and finally into a web form, there is a single tool within which the abstractors would work.
Within the Epic abstraction process are Epic “SmartForms,” each attached to underlying “SmartData” embedded in registry-based reporting structures. I believe the original thought was that this process would help with the abstraction process.
However, I think the truly unexploited value is having those SmartData Elements floating around in provider-accessible data structures such as Epic’s Cosmos where they can intermingle with many other richly meaningful and intuitive data. You could see if there was a statistical correlation between data in an anesthesia database and bypass surgery patients; or we could see if a chemotherapy agent had a previously undiscovered positive effect on certain autoimmune diseases; or there are an infinite number of other permutations. The potential is mind boggling.
Finally, we could couple the query and knowledge discovery tools directly with a way to “publish” and facilitate direct interaction with the methods and data. We could invite others to post their thoughts and poke holes in findings. As we iterate, we will be able to find more meaningful information more rapidly than the status quo.
The place to start is with the databases that most of us feed. Our abstractors are already collecting these semantically and contextually rich data points. Can we collectively convince those groups and societies to which we belong to make these the data set elements available for incorporation into our digital systems? Epic and other EMR vendors then can create standard forms and abstraction workflows for each of these databases. As I noted, I think the real value of Cosmos and other data compilers isn't necessarily to come up with huge volumes of research publications. Traditional research publications are a piece of the puzzle, but they don't mean anything unless they affect patient care. I think we can more rapidly affect patient care if we can democratize these data stores. Ultimately, the ability to rapidly affect and change patient care has to be easily accessible to all.
John Lee, MD, has been a clinical informaticist since 2006 and has been Chief Medical Information officer at two health systems. In 2019, he was honored with the prestigious HIMSS Physician Executive of the Year award. He is a firm believer that the key to solving the myriad problems we experience in our healthcare system is efficient and transparent delivery of information.