MU Health’s Dr. Thomas Selva on AI Algorithm Development
Earlier this year, Thomas Selva, M.D., CMIO at MU Health in Columbia, Missouri, presented at HIMSS22 in Orlando, on the subject, “Saving Lives with the NEWS Algorithm: Using Early Detection and Rapid Response.” Dr. Selva and his colleagues had received the 2021 HIMSS Enterprise Davies Award for their achievement. Later this spring, Healthcare Innovation Editor-in-Chief Mark Hagland interviewed Dr. Selva about the work that he and his colleagues have been advancing at MU Health, and what they’ve been learning. Below are excerpts from their interview from earlier this year.
Tell me a bit about your role as CMIO at MU Health?
My role is dual. They asked me to be a full-time CMIO; I’m a pediatrician. I told them I needed to stay at least 30-percent clinically active; that’s three half-days a week. I always encourage CMIOs to maintain their street cred. In my view, there’s a reason you became a doctor; you should continue to love it. Also, as they say, as a CMIO, you’ve ‘got to eat your own dog food,’ meaning, it’s important to be able to understand the impact of everything that you’re doing as a CMIO, on practicing physicians.
Could you speak to your Davies Award in the context of past Davies Award winners?
The HIMSS Davies Award has evolved over time. A few years ago, you did have to show some clinical results, but most of it was focused on things like improving nursing staff workflow and lowering overall cost. The criteria for the award have evolved, and are focused more on clinical outcomes. In our case, we had three clinical success stories, one of which I was able to highlight at HIMSS: one was reducing the inappropriate utilization of blood products. Through the effective application of working with people and implementing IT, we were able to dramatically reduce the use of blood products, while avoiding negatively impacting length of stay. Second, we applied usability metrics to building the interface for our nursing staff, to make sure they’re screening patients for potential for falls and depression, etc. We did classic AB testing, testing two interfaces at a time. Our interface allows nurses when to screen and how to do so effectively. Fall risks, smoking cessation, etc., dramatic decrease in risks.
Meanwhile, with regard to sepsis, we worked with the National Early Warning Score, or NEWS. A lot of people are using NEWS. But it’s the way we looked at the score, captured the metrics, and put it into the nurse’s workflow in a way that he or she can see it, and working with our nursing and rapid response team staff, to trigger the right actions at the right time. When to call the rapid response team; and then have their own protocols. Emphasis was people, process, to make sure you’re going to do what’s most effective at the right time and then put the right alerts, the five rights of clinical decision support.
Meanwhile, with regard to sepsis, we know that one out of five patients who die in the hospital, die of sepsis. And the signs can creep up subtly. You want to catch and intervene before the clinical decline occurs. That’s where the National Early Warning Score helps: it’s recalculating the score to trigger intervention. You have protocols in place so the nurses have a set of orders they can execute on, and not have to wait to call the doctor. And if you do have to call the rapid response team, you can do that.
And what you would hope to happen when you implement this kind of implementation is that the activation of your rapid response teams would tend to increase, so the team would come to the bedside earlier, before patients go into significant clinical decline. Applying the NEWS score and algorithm should cause the number of code blue applications to go down. And that is exactly what we saw: we saw the number of rapid response team go up, and the number of code blues go down.
Over the time frame, we saw a decrease of code blue activations of over 50 percent, and a 50-percent decrease in patients transferring to the ICU; and the ultimate measure is if you saved lives. And we saved probably 15 lives that otherwise would have not made it. I can’t give you a percentage, because we didn’t have a denominator to work with; but the time from was from August 2020 through January 2021. We calculated the number of rapid response calls per 1,000 patient days. In with regard to rapid response calls, from February 2019 through September 2020, we saw a 25-percent increase over baseline of rapid response team activations, which is fantastic. Similarly, we saw a 25-percent reduction in code-blue activations over that same time period.
What made the algorithm so smart?
I could say there’s nothing new about the “NEWS” score. In fact, since our success with this, we’re doing the same thing with our children’s unit as well. The real secret was, getting together with the nursing staff and asking them, where in your workflow would this be most helpful to display? And then getting agreement among the nursing staff around what level of score should we trigger rapid response? Call them too early and we’re wasting resources; call them too late, and you have decline already. We began this on a single med-surg unit. This is classic quality improvement, right? Plan, Do, Study, Act. So then we asked the question, can you do it on a second unit, slightly different, a surgical unit? We did that, proved it was successful; and then our governance committee asked whether it was ready to scale. Once we felt it was ready to scale, you’re bringing in nursing education and IT staff people to educate people on what this means. The success was in narrowing the trigger to the precise level of score. And you can often build algorithms to run in silent mode.
And that’s how you can then calibrate the score. And one thing that we emphasized throughout all our presentations for the HIMSS Davies Award, is that we learned a long time ago that whenever you’re effecting change in healthcare, your quality improvement teams will inevitably ask for something—an alert, an order set—be built by IT. And often, teams would be frustrated because they couldn’t answer certain questions, or help IT. Now, we take our clinical informaticists and we embed the very early in the process. For Ben Wax, R.N., a nurse informaticist, this was his passion, and we were tireless in working with the teams, helping the to understand the options they had.
You and your colleagues developed a multidisciplinary group from the beginning, correct? And everyone agreed on what to measure?
Yes. My wife and I came here as pediatric residents, and we were impressed upon our recruitment that we were told by leaders here, “We’re not the best pediatric residency program in the country, but we’re focused on becoming the best. Wanna be part of that?” That really impressed us. And I was one of the guinea pigs sent off to Intermountain Healthcare to learn about quality and process improvement, and we started to try it out here, and four or five years ago, we felt we had reached the turning point where the staff understood things cold, and that we would pick 10-12 projects a year and engage in very rigorous project management and governance, constantly asking how the projects were doing. That’s sort of the way we did this. And it honestly is a wonderful thing to watch evolve forward. At the end of the day, health IT is not the answer, it’s all the steps before that, and then you catch up with your IT stack.
