Baptist Health is a three-hospital, nonprofit system serving Montgomery, Ala. and the surrounding region. It has 680 beds, 550 affiliated physicians and is the largest private employer in the area.
Like most healthcare facilities, Baptist Health has been working to reduce unnecessary admissions and readmissions by using massive data stores in electronic health record systems (EHRs) — in this case, Cerner EHR system.
Baptist Health had been using a LACE index tool, a widely used predictive analytics tool healthcare facilities often deploy within their existing EHR systems. LACE — it stands for Length of stay, Acuity of admission, Co-morbidities and Emergency room visits — ranks patients: the higher the scores, the higher the risk of returning to the hospital.
Five years ago, Baptist Health piloted an AI software tool from Jvion to bolster its data analytics results.
The Jvion Machine is a combination of Eigen-based mathematics, a dataset of more than 16 million patients, and software that can be applied to 50+ preventable harm vectors without the need to create new models or to have perfect data. More recently, Baptist Health added two additional vectors to its AI platform to determine a patient’s general risk of readmission and find ways to lower those risks.
In doing so, the hospital system was able to reduce readmissions and unnecessary admissions by 18% over two years — and save $13 million in related costs, according to Kelly Benson, the director of Community Care Management at the Baptist Health Center for Well Being.
In this Q&A, Benson discussed the AI deployment and its results.
Can you explain how AI is assisting healthcare in general?
"If I have to talk about the Jvion product to a new group of physicians coming through orientation, I always give them this spiel. Most of us carry around smartphones. And if you ever leave your locations on, and you drive to work three or four days in a row, the system is going to pick up the next day you get into your car at the same time, the system is going to flash up and say you’re headed to work and it’s going to take 35 minutes to get there today, for example.
"But, if you get into your car on a Saturday or Sunday, it may also flash up, telling you you're heading to work. But, you can look at that and say, no I’m not going to work today.
"AI can do the same thing for healthcare. It can tell you a pattern of things going on. Just like it can pick up you’re traveling into work this morning, it can do the same thing for your health. If you see a pattern of something over and over… then AI can pick that up earlier."
Dell EMCTell me about your facilities.
"Baptist Health Montgomery is a three-facility system and we have approximately 680 beds. We’re pretty much the market leader in our area. We cover about a 14-county service area. It’s the largest private employer in Montgomery. All in all, we have about 550 physicians affiliated with Baptist. We have numerous outpatient services, including specialties such as neuro, ortho, primary care physicians, convenient cares — just lots of different screening and treatment facilities throughout the community.
"Our largest facility is Baptist Medical Center – South. That’s where a lot of our specialty units are located. We’re joint commissioned certified for stroke center, Level 2 trauma center, Level 3 NICU, which is your neonatal intensive care. Then we have a pretty robust population of orthopedic joint center and cardiovascular services.
"One of the other hospitals in the system is Baptist Medical Center – East. It’s a broad range of services, but it’s known throughout the state as the labor and delivery center; they have a Level 3 neonatal intensive care unit as well."
What was the issue with your existing Cerner EHR system?
"Avoidable [admissions] and readmissions: You always, as a facility, try to strive for best patient outcomes…. You want your patients to do the best. So when you take a look at your data, there are patients understandably where you couldn’t have changed that projected path and as the disease worsened, they were going to come back into the hospital. But then, on the other side, you look at data on other patients and you wonder, 'What if we’d tried this a little sooner; would it have projected a change in that patient’s pathway?'
"We’re always trying to look at what we could do to try to do better. So, with Jvion being able to project those pathways a bit differently, it has really changed the way we think about healthcare.
"The vectors we work with, the different modules within Jvion, is all calls, readmissions and then we have one for our own Baptist employee population.... For the all calls [and] readmissions, one thing Jvion has been able to provide us is disease-specific interventions. In healthcare generally, for heart failure, we do this, this and this, and for COPD [chronic obstructive pulmonary disease], we do this, this and that. But Jvion is able to take that a step forward and they’re able to not only predict that you’ll need to do this, but you’ll have a greater impact on your patient’s outcome if you do this in this order.
"The first intervention they give you may be [to] review medications really well, make sure they’re able to afford those medications and that they’re able to pick them up regularly, whether through transportation or something else. It may say, according to our vector, this one is what you need to spend the majority of your time on because that’s going to give you the most impact with this particular patient. Generally, what they give you is the top five things to do."
What was the shortcoming of your existing Cerner EHR system that it couldn’t predict readmission rates?
"We were actually using what we called the modified LACE tool. So there’s different information that flows into that LACE tool and you get a score out of that. For my department, in particular, Community Case Management, we have limited resources, so a lot of patients were falling under that bucket. We needed to make sure we could see into the correct resources with the correct patient, as much as we possibly could. With limited resources, Jvion’s tool allows us to say, 'OK, these patients are at the highest risk for readmission,' and then I can send these resources within the community to those particular patients to be able to impact them effectively.
"The modified LACE was just a tool we built inside Cerner. [It] is considered one of the best practice tools within healthcare.... It’s just that the Jvion tool has been able to be much more specific. The tool we were using only gave us a score; it only told you this person is at risk. It didn’t tell you what to do with that person, now that you’ve identified them. Jvion’s tool tells you, 'This is what you can do for this patient to have the most impact.'"
When did you begin the AI project with Jvion and how long did it take to roll out?
"We were using Jvion for another module in the red spackle area of our system. In March 2015, we rolled out Jvion to community case managers. We had the structural part of things, but what we didn’t have was a very good system to be able to identify those patients and say, 'This is now what you need to do for them.' So the rollout was fairly easy. Jvion was already receiving data from us because of the other module we already had. We did have to pull more data from our clinical information, but rollout was very easy. Anything they weren’t already getting, we were able to put in place though our IT department.
"Once that happened, the Jvion team came alongside and was able to help us understand what the workflow would look like and we were able to make it our own product as far as operationalizing it."
What was involved in rolling out Jvion Machine and to what legacy systems did you apply it?
"What we did is our IT department set up a file where single DTAs (a single sale in our charting system) were sent to Jvion. When we first started, it was once a day, but I think it’s more than that now."
What kinds of challenges did you run into?
"As an organization, or with any healthcare system, already there were tools to offer risk stratification scores. But you still had to be able to sell the product to healthcare workers. We’re all about facts and science. And AI makes you think a little bit differently. You’re thinking a little more globally; it’s not the here and now.
"Also, you’re able to take information in from all these external layers, your socioeconomic layers. As healthcare workers, we’re most often thinking about the clinical data they’re presenting you with right now. You have to tell to them you’re still going to make a decision based on clinical data, but there’s so much more information out there you need to take into consideration when making decisions for a patient.
"For example, a food desert. If you have someone in a small, rural community who’s on a limited income, they may not have access to fresh fruits and vegetables like someone in a city who can go to a Whole Foods or go to a market. This person may be shopping out of a convenience store or a Dollar General that’s not going to carry fresh fruits and vegetables. And so the food they’re choosing may be processed foods that can sit on a shelf longer, your canned foods, your boxed foods, which in turn have a higher sodium content. That could be affecting their cardiovascular disease, but until you drill down into that socioeconomic piece of data, you may not have been considering that before."
How does the AI determine a patient’s risk, and the clinical and non-clinical factors driving that risk?
"A lot of times you’re looking at their past medical history. So what have they been diagnosed with in the past and how is that effecting their current medical condition? Then, also, you’re taking into consideration any lab values, diagnoses and current vital signs. And it’s not static. So you’re not looking at a snapshot of lab work or vital signs just today. They’re able to pick up what that patient’s baseline is — historically, what have you seen with that patient. Now that EHRs are starting to — I don’t want to say interface with each other — but you’re able to see more of a global picture of that patient through a baseline of data. What’s the lab work the last time they came in and is that getting better or getting worse?"
Have you been able to apply that AI to potential coronavirus patients?
"They don’t have a vector for that. We did actually say AI could spread so much further as the machine continues to learn and as viruses change and bacteria change, the machine will eventually be able to help predict more and more. I do believe AI will make a huge difference with health."
Once you know the patient’s risk, how can that risk trajectory be changed?
"Let’s take that food desert example. The patient may be taking in lot of sodium with their food choices. So it gives you a chance to do a deeper dive with that patient to say, 'Let’s talk about the foods you eat. Do you understand by choosing to shop in convenience stores closest to you, although convenient, do you understand that sodium content could be doing you harm?' It may be that they don’t understand that. So you provide that education first and that may solve your problem.
"But what if it’s not an education problem? What if the patient doesn’t have transportation and can’t get to a market for fresh foods and vegetables? Now, you can reach out to community resources to see what’s available. Are there public transportation options? Are there churches or community services that will help provide transportation? Is there a service like Meals on Wheels that can deliver to the home?"
How accurate was the AI at flagging employees who are at risk of a major health event and how did that affect the outcome?
"One of the things we did with the program was look at our needs and the data to see what our team members needed the most. And then we looked at our current resources within our system to say, 'we’re already helping patients in our community with this; now let’s use that same concept to help our own team members.' And we’ve actually seen really good results.
Two things we’re focusing on right now…, patients with diabetes and patients with asthma or COPD. Those are our two primary dragons we’re working on, and we’re able to let the employees know there are resources for you as a team member you can tap into. It shows them ... their health and wellbeing is just as important as the health and wellbeing of our patients."
How many readmissions or avoidable admissions were you able to cull through the use of AI?
"Jvion is able to give us that information back on a regular basis. I don’t know that I can share an exact number. I will say overall, we’ve seen a decrease in readmissions in our 'all calls' [general diagnosis] readmissions and congestive heart failure readmissions."
Some people would speculate that cutting readmissions and admissions would boost revenue. How did you save money?
"In general, readmissions aren’t going to save you money. Every insurance company has ways of paying, whether by denial for readmissions or penalties Medicare has. In the end, you always want to do what’s best for your patients, anyways."
What’s next in your journey toward more accurate identification of at-risk patients and staff?
"There’s going to be lots to come with AI. I think it’s going to be very broad. I know there are several different vectors Jvion is currently working on [that] we’re considering here. One of them is fall risk — patients at higher risk for falls and injuries. We’re also hearing there are new vectors for sepsis. That’s something we’re looking at implementing as well; that is a big deal. It’s going to be good, it’s very useful."