Are you receiving the most up-to-date KM news? Subscribe to one or more of our newsletters to make sure you are!

Creating a knowledge infrastructure for the ‘learning health system’

Page 1 of 2 next >>


   Bookmark and Share

The idea that the healthcare industry can study the data being created in electronic health records (EHR) to foster ongoing improvement is not a new one, but it is gaining momentum. A “learning health system” is one that commits to the use of data as a byproduct of care for continuous learning. Clinicians and health system researchers want to tackle perhaps their industry’s most significant knowledge management challenge: how to capture the results of research into clinical best practices and more quickly feed it back to doctors and nurses at the point of care.

As Jonathan Perlin, M.D., chief medical officer of Hospital Corporation of America (HCA), has noted, 98 percent of hospitals and 95 percent of physician practices are computerized, and all that data being entered creates an opportunity.

Speaking to a National Academy of Medicine meeting last fall, Perlin said,“We are clicking, but we’re not yet learning. By virtue of all that clicking, a ‘data exhaust’ is created, and in the data are answers to numerous questions.” That data exhaust in the form of structured data could be fed back into the system to spur discovery, knowledge and better population health management.

One of the epicenters of the learning health system movement is the University of Michigan, which has established a Department of Learning Health Sciences on its campus. One of its projects is called the “Knowledge Grid.” I recently interviewed research analyst Allen Flynn, PharmD, who is leading the project as part of his PhD dissertation.

Flynn said that healthcare researchers are focused on analytic processes, machine learning and methods for building predictive models. “The question is how is the world going to organize knowledge that is in that form. We think of those analytic results as knowledge.” He said. The question is: What kinds of library and librarianship are going to be needed, and what skills and platforms are required to manage all of those outputs well?

The Knowledge Grid platform at the University of Michigan is still in its infancy. The goal is to make knowledge about health more findable, accessible, interoperable and reusable (FAIR) than it is today. It involves the creation of a digital repository and associated metadata.

“Think of a repository of computable health knowledge instead of PDFs and articles,” Flynn said. That is not a new concept. Other groups are doing knowledge management work focused on computable knowledge. “Our focus is on creating the infrastructure that would allow many others to stand up repositories like this,” he explained. “We are creating the platform that may help others manage this knowledge in their own context.”

One pilot project of the Knowledge Grid involves the oncology care model and patient-reported outcomes. Providers are trying to collect information on symptoms and patient experience between visits and put in place early warning interventions when a pain profile or mental health profile changes. The Knowledge Grid can place the analytic model and visual model into externalized, modularized objects in a library that can be deployed in one or more applications. “Those could be much more easily shared than they can be today,” Flynn said.

Longtime goals for clinical decision support

For years, health information technology researchers have been trying to improve clinical decision support in ways that make it more useful for providers, but also shareable across institutions so that each provider organization doesn’t have to reinvent the wheel in terms of creating its own guideline-based support in the electronic health record.

In 2013, I wrote a story for Healthcare Informatics about the Clinical Decision Support Consortium (CDSC), a completed five-year project funded by the federal Agency for Healthcare Research and Quality (AHRQ) to find ways to make clinical decision support (CDS) knowledge more easily shareable between healthcare organizations.

In pilot projects, for example, physicians in the ambulatory clinics of Wishard Health Services in Indianapolis got an alert or reminder in the computerized physician order entry (CPOE) module of their homegrown electronic health record (EHR) that was generated at Partners HealthCare System in Boston and sent to Indiana as a Web service. The project extracted a limited data set about a patient, including labs and allergies, in the form of a continuity of care document (CCD), which was sent to Partners. Its system adjudicated the rules against that data and sends back information regarding which reminders or alerts to show the clinician.

Developments today build on CDSC’s work as new interoperability standards are developed. One is called CDS Hooks, which grows out of the SMART Health IT effort developed at Harvard Medical School. SMART, which stands for substitutable medical applications, reusable technologies, takes advantage of a burgeoning interoperability standard, HL7’s Fast Healthcare Interoperability Resources (FHIR), to create an application program interface for substitutable health apps that run across multiple electronic health records.

The SMART project has had success in getting different types of applications integrated into several EHR systems. One challenge, though, with integrating into clinician-facing EHRs is that in order to launch an app, the user has to know which app is going to be useful at which point in their workflow and launch it. For instance, if a provider wants to run an app that is going to help them adjust the dosing of a drug based on a patient’s genotype, they might have to invoke that app while they are making a prescription.

That is where CDS Hooks comes in. Last fall, I spoke about CDS Hooks with Josh Mandel, M.D., a research scientist in biomedical informatics at Harvard University and lead architect for SMART Health IT. He called it an attempt to help clinicians know what apps to load by running checks automatically for them ahead of time and then providing information within context within the EHR.

Mandel described the basic approach of CDS Hooks: As a clinician is writing a prescription, the EHR might send off a notification to an external decision support service. That service learns that the physician is in the process of writing a prescription, and it has the opportunity to return some information in the form of a “card” that will be displayed inside the EHR. It could offer up a new proposal. “If I am writing a prescription for a brand name drug, the card that is returned by a CDS service might propose a drug that is going to be cheaper or more effective and present that in a way that I can just click a button and accept that proposal if I like it,” Mandel explained.

CDS Hooks could also provide a link to an external app. “If I am at a point in my workflow where it might make sense to run an antibiotic selection application, a service might return a link to that app,” Mandel said. “As the user, I can say, ‘yes, this does sound like it would be a useful tool for me right now.’ And even if I didn’t know about that app or wasn’t thinking about it, suddenly there is a card there reminding me that it exists and offering to run it for me if I want.”

Real-time public health clinical decision support

Another example of an effort to create CDS as a module that providers could plug into their EHR comes out of the Children’s Hospital of Philadelphia (CHOP), where two researchers have worked to apply the concept to real-time public health alerts. Mark Tobias, M.D., and Naveen Muthu, M.D., the two informatics fellows and practicing physicians at CHOP, posed the question: Could hospital physicians and ambulatory providers subscribe to real-time public health clinical decision support information from within their EHR workflow? They sought to answer that question by creating PHRASE (Population Health Risk Assessment Support Engine). Their prototype solution takes advantage of FHIR to match public health recommendations to relevant patient encounters in real time.

Last October Tobias and Muthu gave a presentation at CHOP about the problems they saw in getting relevant public health information at the point of care and how their solution might help. They gave an anecdotal example of the type of problem they are trying to solve: how a physician gets knowledge or support when seeing a pregnant patient who traveled to the Miami area and now has a fever. The clinician can go to the Centers for Disease Control (CDC) website or look for an e-mail from the local department of public health that summarizes what to do. The CDC has found that providers know they are supposed to do something, but often don’t order all the tests that CDC suggests for optimal care. CDS is critical because the knowledgebase is changing so fast. Muthu said, “Clinical decision support removes the need for providers to know what they don’t know and provides them with in-context information.”

Page 1 of 2 next >>

Search KMWorld

Connect