A new study in the American Journal of Managed Care explores the use of a machine-learning-driven solution called FeelBetter, using data from Brigham and Women’s Hospital in Boston to provide population-level analysis of medically complex senior patients’ risk of hospitalization and emergency department visits. Principal investigator Lisa Rotenstein, M.D., M.B.A., recently spoke with Healthcare Innovation about the solution’s potential use and the study’s findings.
Headquartered in Boston and Tel Aviv, FeelBetter says its technology pinpoints patients at high risk of deterioration and preventable hospitalization due to suboptimal medication management and proactively suggests interventions to reduce these risks.
FeelBetter CEO Liat Primor said that the company was established in order to solve the problem of polypharmacy patient management. Polypharmacy patients are patient who have multiple medications. “Those are people who have chronic conditions and live with a lot of medications, but not necessarily an optimized set of medications,” she explained.
“FeelBetter was established to help not only the specific patient, but to look at this as a resource optimization problem,” she said. “One of the problems is that health organizations, especially ones that are under value-based agreements that have incentives to keep patients in the community, have limited resource to address those patients.” She said FeelBetter analyzes the data to identify who are the patients that will be hospitalized in the coming 30 to 90 days due to polypharmacy, to help organization allocate resources to them.
Among the study’s findings was that FeelBetter’s platform delivered correct medication warnings in 89.2% of cases, and that in 97.3% of cases, the warnings presented would be helpful to clinicians making decisions to optimize medication therapy.
An assistant professor at UCSF, Rotenstein also serves as the inaugural director of the Center to Advance Digital Physician Practice Transformation (ADAPT) at UCSF and Director of the Center for Physician Experience and Practice Excellence.
Healthcare Innovation: Dr. Rotenstein, could you talk about the challenges a health system like Brigham faces when treating polypharmacy patients and why something like FeelBetter’s solution might be of interest?
Rotenstein: Sure. I will talk from the perspective of a health system, but also from my perspective as a clinician. Patients increasingly take many, many medications and some of those medications keep them healthy. Some of those medications, though, have risks for them or may interact with each other and may contribute to unwanted utilization.
As a primary care physician, you're looking at the patient in front of you, but on a population level, it's important for health systems to identify who is most in need of tailored attention to their medication regimens. That’s one of the first things this technology does in a way that is additive, beyond the types of prediction models we would have had previously, is identify who are the patients among those that sit in front of us every day — and even among those who don't make it in to see us every day — who are at greatest risk for polypharmacy-related unwanted utilization.
The second thing that the technology is really important for is pinpointing, from among the many medications that have different indications: which are the ones where we should focus our efforts on? What are the high-priority interventions that may lead to an unwanted outcome for a patient, and therefore, where should either we as healthcare providers, or should we as systems focus our efforts in terms of altering regimens, or at least considering alterations in regimens to ensure the outcomes we want for our patients?
So again, the two main benefits are helping us pinpoint among our many, many patients who is at highest risk due to polypharmacy, and among those, where should we focus within the particular medication regimens that are associated with highest risk.
HCI: Is there usually an integrated pharmacist on a care team, so that when it is highlighted that this person is at higher risk, you call that person in for a consultation on the medications?
Rotenstein: It varies by health system, I would say. In some health systems, there is a pharmacist. In some health systems, it's a primary care physician who manages this. In some cases, there's a high-risk care manager. But all of these folks benefit from the information, so I think that exactly who manages the information will vary system by system.
HCI: Is it also important to do this kind of work during transitions of care? Is that often a time when clinicians discover how many medications a patient actually has?
Rotenstein: Yes, transitions of care, whether from the hospital to home, the hospital to a rehab facility, rehab facility to home, etc., are high-risk times in terms of making sure that people are on the regimens we intend for them to be, that they understand the regimens, and that we as healthcare teams have thought holistically about the regimens. If we change a cardiac medication because they had a cardiac-related hospitalization, we have to ensure that we have thought about that type of medication in the context of the broader set of medications that they may be on.
HCI: Can you describe the hypothesis for this study?
Rotenstein: I mentioned the two main needs that we see this system fulfilling, and the study had two main questions. The first is the extent to which the FeelBetter system could effectively identify patients who are at high risk of ED visits or hospitalizations. And importantly, could the FeelBetter system identify them in an additive way beyond what is typically predicted by traditional variables that we use in our risk prediction system?
The second question is, among those patients who are high to medium risk, were the recommendations made by the system accurate? Were they appropriate? Were they valid? And that latter question was answered via pharmacist review of medication recommendations made by the system. I will note that all of the this was a retrospective study. We are working with FeelBetter right now on a prospective study, but this was using historical data from Brigham and Women's Hospital.
HCI: A press release said it delivered correct medication warnings in 89.2% of cases and in 97.3%, the warnings would be helpful to clinicians for making decisions. So what happens in the other 11% of cases?
Rotenstein: With any type of machine learning technology, we are going to have to have humans in the loop. And our patients expect humans to be in the loop. So that's one thing. These recommendations will always go through a human and then these systems are always learning. So we actually expect this system, along with the many others that are prevalent in our healthcare system, to improve over time in their performance.
HCI: What are some next steps?
Rotenstein: We are doing a prospective study where we are using the FeelBetter system to identify prospectively the patients who we think are highest risk of polypharmacy-related utilization. Our pharmacists are reviewing recommendations made for these patients, integrating them within the patient's plans of care, in partnership with the patient's primary care physician and healthcare team. And then we are evaluating the prospective impact on healthcare utilization.
HCI: On a different topic, I just saw an announcement from UCSF that said that you were being named to lead a center there called ADAPT (the Center to Advance Digital Physician Practice Transformation). Is that right? Do you have appointments at both UCSF and at Brigham?
Rotenstein: That's right. I started working with FeelBetter probably three years ago. In the interim, I have moved to UCSF, but maintain an investigator appointment at the Brigham. But yes, I am physically in the city of San Francisco.
HCI: Could you talk a little bit about what ADAPT will focus on initially?
Rotenstein: Of course. ADAPT is really focused on how physicians interact with technology, and then what that ends up meaning for healthcare systems, for patients, and obviously for physicians themselves. Part of the motivation is that we know that physician practice has really changed in the last decade. Physicians are spending more time interacting with technology in ways that sometimes take them away from their patients. There are obviously changes in the way that medicine is practiced. Fewer physicians are working for themselves; more physicians working for large organizations. And the confluence of these changes has resulted in high rates of burnout and intent to leave the workforce among physicians. So first of all, there's an impact on physicians themselves, and then there's an impact on our patients too, who want care from us.
At ADAPT, we are looking to build an evidence base around physician practice and physician experiences of care, and then particularly how interactions with health information technology impact the physician experience and the way we deliver care.
HCI: Will you be working with researchers from multiple institutions?
Rotenstein: That's exactly right. Work is already underway. Actually some of the founding funding for ADAPT was a grant that started when I was at the Brigham, and is now joint between the Brigham and UCSF from the Physicians Foundation, where we are conducting a variety of multi-site studies looking at the impact of technology on physician experience. So for example, we have an ongoing study about how AI scribes impact burnout.
Separately, we have another grant that is looking at the impact of AI scribes — who responds to them and how they impact physician time expenditure across nine healthcare systems across the US. So some of the work is in foundational research in terms of physician practice patterns, and some of it is really partnerships with health systems and innovators to understand the impact of technology.
