Moving Evidence From Research to Practice: Q&A With Ken Cohen, MD

Am J Manag Care. 2025;31(Spec. No. 6):SP370-SP373. https://doi.org/10.37765/ajmc.2025.89758

In 2025, each issue of Population Health, Equity & Outcomes (PHEO) will feature a profile of a health system leader transforming care in their area of expertise. This issue spotlights a conversation with Ken Cohen, MD, executive director of translational research at Optum Health. This interview has been edited for length and clarity.

PHEO: What is a typical day in your work as the executive director of translational research at Optum Health?

COHEN: You can think about it as being divided into 2 components. Translational research is really the rapid movement of high-quality evidence from bench into clinical practice. It isn’t as much what you think of in terms of publications—I’ll get back to that—but it really is a model of clinical care. I’ve spent my entire career building a care model that is evidence based, where it uses level 1 and level 2 evidence and puts it in the hands of the clinician at the point of care to make the most effective care decisions. The ways that I do that are relevant to the conversation today. One is that I work closely with our chief medical officers and market medical directors and go specialty by specialty to attempt to define best evidence-based care for major areas of care: coronary disease, spinal disease, etc. I also work a lot with primary care physicians [PCPs], and I’ve recorded a 10-lecture series that is the foundational series for the Optimal Care Model, one for each organ system, really focused on high-level evidence that drives clinical care within a given specialty, so you can think of that as a cardiology lecture, a pulmonary lecture, etc. And then, because this literature changes so quickly, we also write a forum every 2 months that takes the past 2 months’ worth of research and brings it rapidly into clinical care. For example, if there is a breakthrough in the management of COPD [chronic obstructive pulmonary disease], rather than waiting 5 to 15 years, which is the typical time it takes for translation from science to practice, with the next issue of the forum, we would put that literature in the hands of the PCP with recommendations about how treatment should be changed based on that science. So that’s the operational part of it.

And then there’s the research and publication part, and that’s a different animal entirely. There, you’re really talking about, how do you best take this approach to care and study the impact of it on our health care system? And to do that, you need very large data sets where you can look at specific disease entities and study them in terms of how different payment arrangements within our health care system impact those disease entities. As examples of that, the last 3 papers that we published this year, 1 in JAMA Network Open1 and 2 in The American Journal of Managed Care,2,3 all used a very large data set. We asked 17 large physician groups that are all part of America’s Physician Groups to contribute their data. And collectively, those data [were from] 15,400 primary care physicians, more than 5 million patient-years of data, coming from 35 different health plans. So we had large physician groups, large geographic reach, multipayer, and with that large data set, we then studied how different payment arrangements impact care. And we specifically studied full-risk Medicare Advantage, fee-for-service Medicare Advantage, and traditional Medicare in a variety of scenarios. We published, as I mentioned, 3 different papers, all showing how those payment arrangements have a fundamental, large impact on clinical outcomes. What those papers showed is that when physicians are in a fully accountable, 2-sided risk model within Medicare Advantage, patient outcomes are better. So again, 2 components of this translational research approach, and that really covers both of those.

The other important point is that there’s a lot of collaboration involved. As I mentioned, we had to collaborate with 17 physician groups to ask them to contribute their data. We also collaborate with academia, so we have ongoing academic relationships with Harvard University, University of Minnesota, and UCSF [University of California, San Francisco], among others, and there we work with academics within their fields to help them look at evidence-based medicine at a different level. For example, we have a workstream around low-value care with UCSF where we have published studies on the lack of benefit of spinal cord stimulators, the lack of benefit of implantable loop recorders compared with external cardiac monitors for patients with recent stroke, looking for diagnosis of occult atrial fibrillation. We’re now looking at peripheral arterial disease and whether intervention in patients with claudication helps or hurts patients with peripheral arterial disease. That academic collaboration is also important. And then lastly, we work with the Duke-Margolis Institute for Health Policy, helping them redesign the next generation of what the HCC RAF [Hierarchical Condition Category risk adjustment factor] might look like, and so we have important areas of collaboration there as well.

PHEO: That sounds like a lot of variety, so it must keep it interesting. What is the biggest barrier to improved cardio-renal-metabolic outcomes for patients?

COHEN: I think there are 3 main barriers, and I think they’re all probably equally impactful. First is at the level of the patient. We have not found a way, in most circumstances, to successfully activate our patients. When we launch initiatives to try to activate patients, what we often find is that the folks who are already eating a great diet and exercising 5 hours a week are the ones who are most interested in being activated, but frankly, the ones who are sedentary with poor diets and metabolic syndrome are much more difficult to activate. We still haven’t figured out the secret sauce to patient activation, and patients can be resistant to changes. For example, if you look at an obese population with comorbidities, the single most cost-effective thing that you can do for them is bariatric surgery, and yet, patients are remarkably resistant to bariatric surgery, and as a result, it’s significantly underutilized.

Next is a component of clinician inertia, which comes about in part because medical regimens have become very complex. If you look at heart failure, for example, we’re talking about a 5-drug regimen for every patient with heart failure with reduced ejection fraction, and that 5-drug regimen may be on top of a regimen for diabetes and a regimen for hypertension, and so often you get into very complex pharmacotherapies. As a result, clinician inertia sometimes keeps physicians from continuing to push for best care, and it can also be a function of adverse effects. Sometimes, for example, if you’re initiating a GLP-1 [glucagon-like peptide 1 receptor agonist], as remarkably beneficial as that class of medications is, they also have a lot of adverse effects, particularly in the first month of treatment. And as such, it requires a lot of hand-holding by the primary care physician, and when they are overtaxed and overbusy, sometimes that’s just not in their bandwidth to do.

[The third barrier] is, frankly, just the cost of these therapies. If you look at that foundational regimen I mentioned for heart failure, if that therapy uses 2 of the brand-name agents, which it is intended to do, you are already looking at a regimen that is around $15,000 a year. Add a GLP-1 into there, and you’ve just added another $10,000 to $15,000, so you can see that very quickly, these regimens can be so cost-prohibitive that the patients can’t afford them.

PHEO: How can payers and clinicians better collaborate to deliver value-based care?

COHEN: This is a really important question, and I think it hinges upon the sophistication of the medical group and the design of the contract. What I mean by that is, if you have small physician groups, generally those physician groups are not in fully accountable care models, and they generally don’t have a sophisticated infrastructure for data and analytics, and that’s where the health plan can really help them. The payer, for example, can share claims data, and the clinicians can then see exactly what they’re paying for and how they’re paying for it. The payer can also generate pursuit lists, so the case management team may be able to identify the subpopulation of patients with uncontrolled diabetes who are not on guideline-directed therapy. They can present that list to the primary care physician so that they can then act on it. The ability to help with case management and share claims is really important for smaller groups.

For larger groups, I think it comes down to changing the contract so that it’s a fully accountable contract, because, as I just mentioned in the 3 papers that I described, when physicians are fully accountable for care in a 2-sided risk model, they are given the revenue from that contract that allows them to build the infrastructure that they need—so a case management infrastructure; they can then have, for example, embedded behavioral health and embedded pharmacy. They can hire social workers. They can engage with hospitalists to manage inpatients. All of those things require revenue that’s not available unless the physician group is at full risk and is getting a significant portion of that premium that they then can use to deploy.

PHEO: What are some examples of how artificial intelligence (AI) is helping clinicians detect and then manage chronic disease?

COHEN: The responsible use of AI technology is essential to the future of health care. If we think of AI as computer technology simulating human intelligence, then we think of responsible AI as the practice of ensuring that AI is designed and used fairly, ethically, and safely. We have several guiding principles that provide the foundation for all AI work conducted across the enterprise, which we share on our website.4

There are 3 examples that come to mind, specific to AI helping clinicians detect and then manage chronic disease. One is ambient listening. Ambient listening can have a dramatic impact on the day-to-day life of a primary care physician—and specialist physicians as well. What it involves is having an AI-embedded program on a smartphone, so that the physician can walk into the exam room, put that smartphone on the desk, engage the patient in conversation, speak to the smartphone as they’re going through their physical exam, and then discuss with the patient what the assessment and plan are. In the background, the AI technology is building the entire patient encounter into a note, so that when the physician leaves the exam room, that note is essentially complete, and they just need to review it. That’s a dramatic quality of life and efficiency advance for physicians seeing patients in the office.

Next, we use AI for clinical decision support. Now, this is still in the testing phase, but we’re getting ready to roll this out. We do this by offering the next best step in prescribing, and it ties back to this clinician inertia. For example, if a physician is seeing a patient with type 2 diabetes and their [hemoglobin] A1c is uncontrolled, AI can ingest a lot of information from the chart. It’ll look at their age, their BMI [body mass index], their creatinine, and whether they have a positive urine creatinine. It’ll look at the drugs that they’re on. It’ll look at their cardiovascular risk factors. As it ingests all that information, it is making AI calculations about what would be the next best drug for a patient in terms of both improving their diabetes control but also improving comorbidity control. For example, if a patient is on a sulfonylurea and a DPP-4 [dipeptidyl-peptidase-4 inhibitor] and has underlying renal and cardiovascular disease and a BMI greater than 35, it might recommend that they stop the DPP-4 and the sulfonylurea and substitute a GLP-1. It does it in such a way that it’ll bounce it against the patient’s formulary and only present the formulary-listed drug to the physician, so that if the physician agrees with the recommendation, it can be literally a single keystroke for the physician. It will discontinue the 2 drugs and [suggest] the prescription for the third, and the physician is off and running with the best evidence-based therapy.

The other major area where we use it is around large language models for predictive analytics. Optum Health now cares for about 23 million patients, and about 5 million of those are in fully accountable care models. For those fully accountable care models, we have built more than a dozen models that will show us, for example, risk of readmission for those discharged from the hospital, frailty risk, fall risk, and polypharmacy risk. And then we have been able to stratify populations into 1 of 5 bands based on severity for diabetes, heart failure, chronic kidney disease, and COPD, and then those highest-risk patients can be directly given to the case managers to work with primary care and the specialty physicians to improve the outcomes for those patients. That’s a really important way to deploy AI in day-to-day practice.

PHEO: Finally, please tell me about any ongoing or upcoming initiatives that your team is working on that you’re passionate about.

COHEN: Yes, I’m passionate about most of our initiatives, but if I were to choose the one that rises to the top, it would be redesigning specialty care. This is a desperate need for our health care system. CMMI [the Center for Medicare and Medicaid Innovation] is well aware of this and took steps during the last administration to try to address this, although unfortunately, they don’t have a lot of levers to pull to improve it. The reason that this is so important is that approximately 85% of total cost flows from our specialists, either direct cost of the specialists’ care, pharmacotherapeutics that are prescribed by the specialist, or procedures that involve both a procedural fee. Then, if the procedures are done in an ASC [ambulatory surgery center] or hospital, facility fees, and you add all that together, it’s about 85% of the health care dollar. The problem is that a fee-for-service model pays equally for high-value care and low-value care, and it even pays for harmful care. How can you redesign specialty care so that the specialists are now engaged in a value-based care model? Because up until now, they have largely been disenfranchised.

What we have done is we’ve acquired and also developed highly granular metrics of quality and utilization that are very specific to a specialty practice, even to a subspecialist. If you look, for example, at cardiology, we have a set of metrics for general cardiology, a set of metrics for interventional cardiology, and a set of metrics for electrophysiology. We then can apply those metrics to every one of our markets, and we can study physicians in those markets, so that now we have generated reports where if you click on a market, it’ll show you performance across all of these metrics for every cardiology group in your market. If you then double-click on the group, it’ll show you these measures at the level of the individual physician. And we are now using these data on quality and efficiency to redirect our referrals to the highest-performing physicians, and we have done this for orthopedics, cardiology, gastroenterology, urology, ophthalmology—about 7 specialties in total. As a result, we can feel really good that the physician referrals are now based on quality and efficiency, as opposed to proximity to the office, personal relationships, all the other things that physicians based the referrals upon, and, unfortunately, they did that because they didn’t have granular metrics of quality and efficiency available to them. So, [it’s a] groundbreaking transformation. We share these data transparently. We have our specialists give us input on the metrics so that they’re on board with the metrics that we’ve chosen. And we’re developing compensation models where we’re having incentive payments given to the specialists based on how they perform against these quality and utilization measures, so a complete redesign of specialty care. We’ve completed year 1 and are just entering into year 2. This will rapidly evolve, and I think it’ll be probably one of the most significant initiatives that we’ve developed.

REFERENCES

1. Cohen KR, Vabson B, Podulka J, et al. Medicare risk arrangement and use and outcomes among physician groups. JAMA Netw Open. 2025;8(1):e2456074. doi:10.1001/jamanetworkopen.2024.56074
2. Vabson B, Cohen K, Ameli O, et al. Potential spillover effects on traditional Medicare when physicians bear Medicare Advantage risk. Am J Manag Care. Published online February 26, 2025. doi:10.37765/ajmc.2025.89686
3. Cohen K, Vabson B, Podulka J, et al. Health outcomes under full-risk Medicare Advantage vs traditional Medicare. Am J Manag Care. Published online May 9, 2025. doi:10.37765/ajmc.2025.89740
4. Ensuring responsible use of AI in health care. UnitedHealth Group. Accessed June 9, 2025. https://www.unitedhealthgroup.com/ns/making-health-care-work-smarter/ai-responsible-use.html