MultiCare CMO on Deploying Clinical Financial Decision Support ‘Nudges’

Arun T. Mathews, M.D., is a regional chief medical officer for Tacoma, Wash.-based MultiCare Health System, a nonprofit organization with 13 hospitals. He recently spoke with Healthcare Innovation about a new type of clinical decision support effort there that involves using CDS Hooks, an HL7 specification that allows for enhanced integration of clinical software applications within an EHR system.

Healthcare Innovation: I understand that MultiCare has deployed a clinical financial decision support “nudge" application from Illumicare. How did that come about? Was there a recognized problem that the health system was seeking to address with medication or lab ordering?

Mathews: Coming out of the pandemic, part of our strategy around financial recovery involved partnering with our physicians, because we recognize that a significant amount of healthcare costs comes at the point of ordering — whether it be laboratory work, medications, or imaging.

We wanted to understand if there existed opportunities to quantify what we have termed healthcare “waste.” So we partnered with Illumicare to do a retrospective analysis based on best-practice principles from a series of databases that they build at the core of their technology stack. One of those databases is called the American Board of Internal Medicine Choosing Wisely program. Essentially it looks to root out a combination of care that is duplicative and unnecessarily. If there's a cost equivalence between two types of medications, appropriate care leans toward the most appropriate medication from a cost containment standpoint. 

Illumicare ran an analysis based on their database and a series of other analyses that they've done nationally, and identified a series of opportunities for us, primarily in the medication and lab ordering space. We engaged them to use an advanced clinical decision support tool called the illumicare Ribbon that at the point of decision-making provides best practice or operational practice alerts. That’s in the Epic parlance. But they're effectively cognitive nudges that bring an individual to an awareness that there may be another opportunity here. This medication might be similarly efficacious.

HCI: Clinical decision support can be unpopular with a lot of clinicians because they find too many alerts are not meaningful to them at the point of care. How do you assess whether the clinicians find these particular alerts valuable or not? Did you do pilot studies or survey the clinicians to see whether they're actually going to use then or just turn them off?

Mathews: We started by just turning it on with very little fanfare and seeing if there would be any sort of organic use. And there indeed was some organic use, and it was a classic sort of early adopter/fast follower mindset there. But when we felt that we had a compelling story in terms of why this was important, we began a series of lunch-and learns and outreach with our clinicians to explain that if we're not good stewards of the care that patients receive, a lot of that expense can actually roll up to the patient. After this outreach, we saw an increase in the utilization of the tool.

I think we reached an inflection point after doing a series of surveys with our physicians. We uncovered that we had a well-intentioned assumption that this is just a quick little notification, and a physician can just push forward. But we actually did a cognitive workflow analysis of how a physician thinks when they have to deal with this new advanced clinical decision support notification, and it's actually a significant amount of thinking that needs to occur. 

You have to look at the recommendation, assess the evidence driving the recommendation, make a determination if that evidence is appropriate for the N-of-1 trial that is the human being that you are caring for, and if it is appropriate, then you have to go into another ordering pane and either stop or start a medication. That is a significant amount of cognitive burden, and that's just one alert. 

So that was a big aha moment for us, not for just this tool, but for any sort of clinical decision support. There are a lot of wheels turning for our doctors, and then for a busy rounding hospitalist seeing from 16 to 22 patients a day, every little moment counts. 

That's when we decided to partner with our contracted hospitalist group, Sound Physicians, and co-develop a clinical stewardship quality bonus metric. We felt strongly that if they were taking on this extra work of reconciling these recommendations based on best practice ABIM Choosing Wisely, they should get some sort of credit for that cognitive work. 

We did a literature review to understand if shared savings mechanisms had been used for this type of work, because we knew that we would need to navigate through legal and compliance to make sure that our organization  was comfortable with this type of approach. We were pleased to see that this has been used myriad times in the past, in the world of oncology, population health, orthopedics and even primary care. These types of mechanisms giving physicians credit for this extra work of essentially being thoughtful around healthcare costs and reducing healthcare waste had been done before, so we presented that literature review to our legal and compliance teams and received the green light to proceed — with the caveat that we also be thoughtful about tracking countermeasures such as patient experience and readmission rates. 

Once we did that, and we had these great conversations with our physicians about acknowledging that this would be rolled out in concert with a clinical stewardship metric that acknowledged, down to each individual decision, the work that they were doing here, it was like a hockey stick-shaped curve in terms of engagement. And so that was the big learning for us. 

HCI: When did all that happen? 

Mathews: It was in the third quarter of 2022, and then we got to see that expanded level of engagement through 2023 and 2024.

HCI: Are you able to track the impact on spending? Can you actually see the number of times they accept the recommendation and then downstream from that, the kind of cost savings you're seeing?

Mathews: Yes, down to the individual physician or APP, down to the individual decision, and what the downstream effect has been. In that first year, we set our targets as quite nominal. I believe it was less than a 5% cost savings. We wanted to be very conservative, and we ended up seeing our physicians rally behind this, and our savings tracked up over 15%, I believe, in that first year, amounting to about $2.5 million worth of healthcare waste reduction.

HCI: I read that the technology uses CDS Hooks. I read a lot about CDS Hooks when it was first introduced, but I haven't heard of too many examples where it's actually being deployed in health systems. Can you talk about that?

Mathews: I think this is a really fascinating follow-on story. When we first rolled this out, we we did not have the CDS Hooks capability, which means that after a physician made the decision to follow the recommendation or not, they would then have to exit out of the alert, go into the ordering pane, make the decision or stop the plan for ordering a new drug or a different drug, and that that required multiple additional steps.

The concept of utilizing CDS Hooks actually came from our design partnership and feedback sessions midstream. We said the docs are using this, but we've received feedback from them saying it involves a bunch of extra steps. Wouldn't it be neat if buried within the design of the alert is — if I agree to do this thing, it manages it all in the background for me. 

Basically, the company listened to the feedback from our physicians, applied CDs Hooks, and then built that into the design of the next-generation alert, and is now rolling it out at MultiCare. I think the feedback has been overwhelmingly positive. It has reduced additional steps in terms of clicks and the task of ordering. So I think it's a really nice human factors engineering responsiveness story. 

HCI: Do you think there is potential for other CDS Hooks-type apps to tap into the clinical decision support in your instance of Epic?

Mathews: 100%. I think this really demonstrated what I call medicine at the speed of thought. As we get closer to evidence-supported decision-making, then the number of hurdles for me to execute that decision should be decreased dramatically. And that, to me, is the promise of CDS Hooks. 

HCI: We are hearing so much about AI in healthcare Is it happening in the clinical decision support space as well?

Mathews: I think we're in a bit of a Wild West frontier in the AI space, and all of the initial excitement appears to be in the use case of ambient scribes, and rightly so. I think the next generation of that work is, of course, decision support. It's bubbling up the most appropriate information tailored to that individual’s care, with the most appropriate management strategies built into that. I think we can have clinical stewardship paradigms so that it doesn't feel like a separate need for a series of alerts. It's just built into that general conversation that the physician would have with the agentic AI, working kind of like a super resident, so that the quality of medical decision-making goes up a level as well. 

But there are so many different additional use cases for agentic AI: problem lists, medicine reconciliation, chart summarization, and pre-visit and post-visit patient education.