Exciting things are happening these days at St. Francis Hospital in Columbus, Georgia. Charles E. “Chuck” Christian, the hospital’s vice president and CIO, has been helping to lead a data initiative that could put the organization on the leading edge in terms of leveraging data and analytics to improve readmissions reduction work. At the core of the initiative at the 375-bed hospital is an ongoing project, in collaboration with the St. Simons, Ga.-based Life2.
As part of that initiative, consultants with Life2 have been developing a database and associated set of analytics tools to help the St. Francis clinicians and administrators analyze and determine the level of risk of current inpatients being readmitted after discharge; what’s more, the goal of Christian and his colleagues is to determine which patients are at risk for longer stays upon readmission, given the penalties for such events, both under the Affordable Care Act’s avoidable readmissions reduction program, and in the context of the ACA’s value-based purchasing program.
Christian began laying the groundwork for this initiative shortly after joining St. Francis in January 2013 (before coming to St. Francis, he had served as CIO at Good Samaritan Hospital in Vincennes, Ind., for 23 years). Christian, who in addition to his role as CIO at St. Francis Hospital, is also the current chairman of the board at the College of Healthcare Information Management Executives (CHIME), spoke recently with HCI Editor-in-Chief Mark Hagland regarding the current initiative. Below are excerpts from that interview.
What were the origins of this initiative at St. Francis?
The folks in case management were looking at a better way to manage readmissions. And it’s like everything else in healthcare—if you’re looking at data from the past, you’re only able to guess the future, and you find yourselves in a continuous retrospective review process. You start looking for patterns and trends and so on. So we were looking for a real-time predictor of trends in readmission. It doesn’t take a rocket scientist to look at some of the basic co-morbidities that feed risk; however, not every patient is going to be at risk for readmission.
So we decided to look at predictors while patients are still in the hospital, and to look at the appropriate post-acute care setting for those patients to support readmissions reduction. And we wanted to look at patterns of where they’re admitted from—such as the particular post-acute care settings they’re coming back from—and under what circumstances under which they’re we’re coming back into inpatient readmission.
We started looking at this when I was [at Good Samaritan Hospital] in Indiana. But, looking at what led to a readmission while the patient is in the hospital, and to the extent possible, even looking at readmissions while patients are while they’re in the post-acute setting—if they get readmitted, we’ll take a hit on our value-based purchasing measures, and will have to care for them for free under the readmissions reduction program in the ACA. And furthermore, for a particular stay, Medicare will pay for, say, four days, and beyond that, we’re eating the cost of their care, in essence. So we’re looking at patients who are at risk for extended lengths of stay—four, five, six days, for example.
So with Life2, we’ve identified a particular person as being at risk, and then, what next? Based on historical interventions, what are some guidelines and evidence-based data that the case managers should look at and discuss with attending physicians and others? Those are the kinds of questions that have driven our work on this. Then, we’re building a communications platform between the acute-care setting, St. Francis, and various post-acute care facilities—we’re piloting that right now. And that’s one of the things we’re going to have to spend more time on.
If you look at the meaningful use program, the only two areas meaningful use covers is the physician practice and the hospital. But now you’ve got non-traditional care settings, like minute clinics and retail settings, for primary care, for example. And once a patient is ready to go home—and patients don’t get to stay in the hospital as long as they used to—they’re transitioned either to home care in the home, or to a rehab care facility, because they need a higher level of care than the home, but lower than in the hospital. And we need to be able to communicate with the post-acute care folks and provide them information to optimize them and to arm the next level of care with the appropriate level of information they need on a patient rather than their being surprised when they get there at what they get. There is an evaluation done before discharge, but if the patient’s condition changes from the time they’re assessed to when they arrive at the post-acute care facility, that facility needs to know if a change has occurred in the patient’s situation.
And maybe there’s something we can do before those patients come back to the hospital. I think this is a bit different from the usual approach, in that we’re using predictive analytics across the continuum of care. And in Indiana, one of the challenges we had was finding appropriate post-discharge primary care. So we set up a medical home staffed by a primary care physician and nurse practitioners, because often, patients were just ending up using the ED for follow-up care. And at St. Francis, the two physician practices we partner with—one has 27, and one has 14—do have medical homes—they’re both certified.
What are the key data points you’ve been setting up?
There’s a bunch of them. There were 10,000 data points that they’ve boiled down to 100 or so, using really good mathematical models. And they include elements like co-morbidities, past medical histories, previous frequencies of admissions, their ages, if they have chronic conditions, what those are. And they look at how many medications they’re taking; they look at vital signs and physician progress notes.
When did you begin developing the database, and then populating it with data?
The database development started a year ago, and we’ve been putting data from about six years’ worth of data into it, over the past six months. You’ve got to have enough breadth to get useful information over time. Ultimately, over 100,000 patients will be in the database over time.
When will you be going live?
Later this month, we’ll be meeting to put together the workflows and work processes to operationalize this. How will this information be used by the case manners to affect care? I’d like to have it up and running by the middle or end of the second quarter.
What have been the biggest challenges, of any kind, to date?
Normalizing the data, making sure we get the data in there that we need—that’s been our biggest challenge so far. We’ve got to establish a routineness in creating the data feeds. Every day, the case managers will get a scored report—here are the patients in-house who are at the highest level of risk, based on these factors. And of course, the case managers won’t be automatons; they’ll be using their clinical knowledge and judgment as they use the data that their dashboards provide them from the database.
And the data can’t be static; we’ve got to continuously fine-tune and tweak it; that’s what we’re aiming for. Our data people created models using the reference database. Once they had developed predictive measures, they applied historical data to the process—in a way somewhat similar to how one might set up a blind trial. They worked with several years’ worth of historical data, used it to create data models, and then ran it across a previous set of data not used in the models—a control set of data. What they found was that the models they had developed were very good at predicting what would happen to patients. And that’s how we knew we were prepared to move forward.
So within a couple of months, you’ll be able to actively predict the patients at greatest risk of readmission?
Exactly. And in fact, we’ll be able to predict the patients at the greatest risk for extended length of stay, upon readmission. And from a revenue standpoint, that will have a far greater impact than readmission rates alone. We won’t be discharging any patients prematurely at all; instead, we need to optimize the care and interventions the patients get while they’re inpatients here.
I know you’re still early in your process overall, but what have the major lessons that have been learned so far?
The lesson learned for me is that you really need to go out and bring in people who do this for a living, rather than doing this by yourself. The people at Life2 have a deep history in the insurance and transportation industries, in using analytics to do this kind of analysis. So, often, it will mean taking tools not born in healthcare, and then applying them to healthcare. That’s how we can learn: it’s why I like to talk to the CIO of FedEx or UPS or Coca-Cola, because you’ll see how they’re using information technology to streamline operations.
Do you have any advice for other CIOs and senior healthcare IT leaders around all of this?
Keep an open mind, and look for ideas out there that will enhance operations. We have to find new, creative ways to innovate. And because of meaningful use and other things, we now have these treasure troves of data. And if we’ve been diligent about making sure that the data has integrity and is valid, we should be able to mine that data appropriately. It’s a little bit like being in a cave and shining a light on the darkness: you may end up shining a light on something you didn’t’ want to, but at least you’re down there exploring.