Progress on Clinical Decision Support for Chronic Pain Care
The Regenstrief Institute’s Chris Harle, Ph.D., has done years of research focused on clinical decision support tools to aid primary care clinicians in chronic pain care and safe opioid prescribing. He recently spoke with Healthcare Innovation about the potential impact of those tools.
Harle is a professor and chair of the Department of Health Policy and Management at the Fairbanks School of Public Health at Indiana University and a research scientist at the Regenstrief Institute. He will be inducted as one of 25 new fellows in the American College of Medical Informatics on Nov. 10 in San Francisco, during the American Medical Informatics Association (AMIA) 2024 Annual Symposium.
Healthcare Innovation: First, congratulations on your upcoming induction as a American College of Medical Informatics fellow. I read that you spent over a decade working on clinical decision support tools, especially with a focus on safe opioid prescribing. Could you first talk about some of the challenges that primary care physicians face in pain management and prescribing opioids that the decision support system can help overcome?
Harle: Thank you. Over the last 15 years or so, opioid prescribing and the opioid crisis in the country have gone through several different periods, I'd say. But fundamentally, when we started this work over a decade ago, and an issue that still remains, is that people who have chronic pain have long histories of suffering, of trying different medical treatments, sometimes opioids, sometimes other pharmaceuticals, sometimes other non-pharmaceuticals, and frankly, oftentimes are not relieved of their symptoms in a meaningful way. Pain is bio-psychosocial in nature. It's got biological, psychological, social components, and it makes treatment very challenging. It varies person by person. It can change over time.
So in the healthcare system, this leads to an accumulation of information — just a massive amount of data and information about their symptoms, about their care, about the treatments that they've tried. That becomes a challenge when they go to see a primary care clinician who may need to rapidly get up to speed on that history and think about what is the right course of treatment going forward. There are benefits and risks to every potential path forward, in terms of outcomes, unintended consequences of treatment, side effects and so on. What we try to do in our work is to synthesize that information, make it easy during a brief visit for the clinicians to get the information they need to make the best treatment decision going forward, one that maximizes that benefit/risk trade-off.
HCI: Does it work sort of like a decision tree? If the answer to question A is yes, try this?
Harle: Not yet. The first problem we tackled is just finding all of the information that is scattered in various places in the EHR and bringing it together in a single view for a clinician to look at. That’s been challenge number one that we've been able to solve with what's called OneSheet, and that's a great start. It brings a lot of information together: diagnoses, medications, lab results, prescription drug monitoring reports, which show opioid and other controlled substance prescriptions from various places. All of that is accessible from one place in the EHR via OneSheet. We've also built in some clinical actions. We make it one click to order a urine drug screen for a patient who might be on an opioid. If you are thinking about opioid alternatives, we try to make it a little faster to order non-pharmacologic treatments for pain. So there's some of that built in. It's not yet to that point of saying, based on all of your treatment history, consider x or y.
HCI: Is all that work done in Epic? And have you deployed that at a couple of health systems?
Harle: Yes, OneSheet has been deployed at two different academic health centers, one here in Indiana, Eskenazi Health, as well as Atrium Health at Wake Forest University in Winston Salem, North Carolina.
HCI: Did I read that you are working with Epic on a way to distribute it more broadly nationwide?
Harle: Yes, we’re hopeful that we're going to be able to do that in the near future. We're not there yet, but because the tools we've built use native Epic functionality, I think they became interested, and they obviously have a massive platform to potentially distribute it. So we're hoping to get there, and then other places could adopt OneSheet, and adapt it to their organization, and use it with their patients.
HCI: Would it be starting over from scratch to do it in Oracle Cerner or another EHR because you have to figure out where all this data is located in the EHR?
Harle: Going from one Epic site to another Epic site reduces that challenge. At the same time, I think other EHRs have a lot of the same native capabilities that make it feasible to build a similar interface and to bring these types of data together. To your point, there will be more effort involved in the accurate mapping so that the full scope of pain-related diagnoses show up in this panel of OneSheet.
HCI: Have you had a chance to either survey or get a sense of the provider reaction at Eskenazi or Atrium Health after their use of this for a while?
Harle: We have done a fair amount of evaluation, and these tools have been live there for several years. We conducted a trial at both of these sites where we looked at several different outcomes related to opioid prescribing guidelines, including use of OneSheet.
When it comes to use of OneSheet, it's really interesting. It's sort of a tale of two different kinds of users, if you will. I think we rolled this out to 69 different primary care providers, and each of them tried it. There are some clinicians, physicians and other primary care clinicians, who love OneSheet and use it all the time. It has become part of their native workflow, and they said, ‘Don't take it away from me.’ At the same time, some of them tried it a few times and then they moved on. They have not used it since, or they didn't use it through that year and a half or so that we ran that study. So we're also interested in understanding why they use it or don’t and how we can perhaps make others use OneSheet more as well.
HCI: What are some possible explanations for why some physicians are not incorporating it into their workflow?
Harle: In primary care, an individual patient comes in with many different conditions and not all of them are pain. Some people have pain, plus other things. Some people don't have chronic pain. So if you're a busy family physician, making OneSheet a consistent part of your workflow is difficult. Sometimes it's a matter of not remembering it because I don't see a patient with chronic pain every day. Our power users are those who tend to see a lot of chronic pain and/or have patients on long-term opioid therapy. They build a habit of using it. The other thing we've built in is an alert that tries to identify patients for whom OneSheet use might make sense.
HCI: Well, let me shift gears a little bit. I was reading on your bio that for several years you were the Chief Research Information Officer for University of Florida Health. Could you talk about some of the issues you dealt with there in terms of strategy and operations of research data across the UF academic health center?
Harle: This was a really fun role, and I continue to be excited about research IT infrastructure and research data infrastructure. I often say we live in this world now where it's really easy to collect lots of data, and of course, the question is, what are we going to do with it, and how do we use that to improve healthcare delivery?
As CRIO at the time, we were really focused on, how do we maximize the research value of the data that spins off of healthcare every day. Every time a patient has a visit, has a test, is prescribed something, that creates data, and of course, that's warehoused. It's used for treatment, payment, and operations, as part of the care delivery processes. My team was interested in how we can maximize the reuse of that data for research.That was about running an operation that had researchers as customers, whereby we had to deliver them data in a timely and high-quality manner. We also, of course, had significant regulatory rules that we needed to be following research ethics, IRB, HIPAA and so on, so that we were reusing the data appropriately.
HCI: Was genomics data a big part of that?
Harle: Not a big part of it. There were projects here and there. For example, there's a fantastic team at the University of Florida focused on pharmacogenomics, and they were working to deliver pharmacogenomic testing in day-to-day practice. Some of those results would be part of the EHR, and there was some study of the delivery of those services.
I think a bigger, more important point that's exciting for people in roles like I was in as a CRIO is not just bringing together the structured EHR data that we often think about, but bringing in other modes of healthcare, or health-related data: genomic data, all of the text data that lives in EHRs, imaging data. If we can bring all of those together, the scope of research we can do is much bigger.
HCI: What about participation in clinical research networks that involve multiple health systems? Was UF involved in any of those?
Harle: Yes, it was in several. One of the major ones was One Florida, which is now called One Florida Plus. It’s a research network to which many different health systems in Florida and now the Southeast contribute data. My team, on a routine basis, delivered data to that network. Then you start to reap the benefit of data that spans more patients, more regions. You get more diverse data that can be applied to different research questions. We often described what we had access to within the health system as very deep, because we had a really deep set of rich data for each of the patients. But then when you start to bring in data from many sites over a network, it's broad in the sense that you have more diverse patients. You may not have as deep a data set and as much detail about a given patient, but you have a much broader set of patients because you have many organizations feeding into these networks. So sometimes you answer the same research questions with different data and triangulate your results. Are we finding the same patterns? And sometimes certain research questions lend themselves to a single site of EHR data versus one of these multi-organizational consortiums.
HCI: We are all hearing about the impact of AI developments. Will you or other researchers at Regenstrief be studying the impact of AI on clinical decision support?
Harle: Yes, absolutely. I think folks that here in Indiana and Regenstrief have been thinking about artificial intelligence for a long time, and continue to think about applications in clinical decision support.
We talked a little bit already about OneSheet, and I think some of the potential of more advanced AI is to get us beyond what I shared with you about OneSheet now, which is as kind of an aggregator of information. We may bring it more towards something that provides a little more specific guidance around decision making. There are a couple of things we're working on with OneSheet, just as exemplars of ways to use AI. One is to try to find all of the related information that exists in clinical notes. Oftentimes, clinicians are not documenting data in discrete boxes and fields in the EHR. They're writing notes. When it comes to pain, things like how pain is affecting your daily life, your ability to work, your ability to do things with your family — lot of that gets put into unstructured clinical text data. So we're super excited about natural language processing methods, including some of the newer, large language model approaches to querying notes and seeing if we can accurately and consistently find pain symptoms and also accurately find pain treatment history, because we also know a patient may come in and say, ‘You know, the only thing works for me is smoking marijuana, and I actually have been doing that for several years, and I think it alleviates my pain symptoms.’ That may get documented in a note and may not get documented in a structured way, because it's not being prescribed in many cases. So AI has great potential to help us pull that out and then help a clinician make a more informed decision.
HCI: Is there some cross-pollination of interests with other researchers at Regenstrief that make it advantageous to work there, or fun to work there?
Harle: Regenstrief is a fantastic, amazing and unique place. It sits here on the Indiana University campus and is heavily tied into our Indiana University schools, but it also is sort of a convener and a home to researchers from lots of different disciplines, all of whom are focused on improving healthcare and health outcomes more generally. You can work with experts in AI. You can work with clinicians from lots of different backgrounds — nurses, physicians, physical therapists, dentists. You can work with people who are sociologists. People come from lots of different disciplinary backgrounds, and we're all working toward a general direction, but we can bring our own methods and our own backgrounds and our own knowledge to answer applied questions. The other big, exciting thing about a place like Regenstrief is that it is a steward of health data from around the state of Indiana, so it has a great asset in the data that it has access to, and it can share with all the appropriate approvals for research.
