Emerging role: How a CHIO harnesses data for population health
John Showalter, MD, is the chief health information officer at the University of Mississippi. He evolved into that role after studying biomedical engineering at Columbia, followed by combined residency and fellowship in clinical informatics at Penn State for medical school.
His deep connection to an Epic implementation there led him the University of Mississippi Medical School, where he took on the role of CMIO, leading the clinical work in adapting to the changes that incurred.
That role in turn led him and the institution to consider the wider role of community health and the adoption of his new title. Healthcare IT News had the chance to speak with Showalter at length, and here is what he said.
The chief health information officer has an independent academic role.
I'm currently parallel with the CIO, in a different leg of the leadership structure, I report to our associate vice chancellor of research. The CMIO is an M.D. reporting to the CIO.
The chief medical information officer role is really about physician engagement, whether it's the physician adoption of the EHR and the new workflows. It's about making healthcare function on an operational level and easing the technology burden on physicians and nurses as they deliver care. The CHIO role is much more about the data--the data integration, the data analytics, and looking at how we can use insights to improve care, and improve health care delivery.
Part of the reason why we pulled the CHIO out is so that he [would] be a truly independent assessor without having obligations to a department that might cause us to want to make someone look better than they actually are.
Healthcare analytics is the next great frontier for medical innovation.
I perceive of analytics as the antibiotics of our time. I know penicillin didn't cure everything, but it doesn't mean that penicillin wasn't a wonder drug. That's where we are with analytics. It is not a silver bullet. Analytics is greatly proven in other industries, and we have really good proven healthcare evidence that it works. It is dependent upon having to do something once you know what you need to do.
University of Mississippi uses dirty data to plan interventions.
If you're looking for something where you're going to pick which chemotherapy [a patient] should have, you need really tight, precise analytics, but for a lot of the care management activity we need to do, we just need to know who is at highest risk--not their exact risk. We have mathematical approaches that will handle how dirty the data is, especially in the deep machine learning scenarios.
We just implemented a pressure ulcer bedsore prediction algorithm, where we are dumping raw feeds of data into a deep machine learning, combining it with social economic risk factors, and are able to identify a group of patients that have a 40 percent chance of developing a pressure ulcer, when the background rate is less than 1 percent.
Even with how horrible and dirty that data is, we were able to identify a base population that is at 20 to 40 times the risk of normal. Since we're using dirty data, that might be off; maybe they're 15 percent higher risk, or maybe they are at 30 percent higher risk. But they're clearly in a group that needs to be addressed.
When you're able to identify that kind of risk, and the interventions are something simple like turning the patient, using the more expensive mattress, using the more expensive pads, and there's really no risk, only benefit, to the patient, it's clear that that data is good enough.
The patient's ability to engage is a core criterion for population management cohort inclusion.
We're looking at engagement: people that actually attend their appointments, fill their prescriptions, are engaged and do the paperwork. The things that suggest that they are going to ... communicate effectively with their care teams and value their interaction with the institution. If you've canceled five of your last six appointments, you're probably not the most likely to engage in a telemedicine program where you're supposed to be doing coaching once a week.
There's definitely a concern, not so much about punitive, but about eliminating for unknowns. So if they're canceling their appointments because they don't have transportation, the solution's not to exclude them, it's to get them transportation. I think most of us are in agreement that we need to do something to move the needle. We know that we're going to have some inequities, and we're going to have to be conscious and study and look for those inequities as they come up.
Technology partners have an important role to play.
In my group, we've done really powerful analytics out of Excel. We're also having really good success with the advanced machine learning piece.
We're partnered with Jvion for our advanced analytics. They are a layered deep machine learning vendor. They allow us to get social economic risk data. As part of their platform, they offer neighborhood-level data. They get median income, housing, environment, purchasing history, access to care. Not at the individual level. They know that 75 percent of that neighborhood is uninsured or that there aren't pharmacies nearby, and they pull all those variables back into the algorithms. So we not only have our clinical data, they also have environmental data, and they have a number of knowledge-based information they pull into their algorithms.
The other vendor that we're working with is M*Modal, for natural language processing. We're feeding all of our unstructured data into their engine. They're turning it into structured, clinical content.