Optimizing Care Transitions: Where Do Predictive Analytics Tools Fit In?

One thing we’ve been working on is a real-time sepsis monitor. It’s a very simple list: here are the patients at risk. The challenge is identifying the appropriate mechanism for integrating into the clinical workflow. The easy part is building it technically, though that’s still challenging. But the big challenge is, once you have the answer, putting in place the processes for improvement. Because there’s a lot of, “Ooh, wouldn’t it be cool to do this?” But then, afterwards, did the people asking for this realize there would be a 40-50 percent false positive rate? And the same questions go into alerts and reminders. And this sepsis alert provides clinical decision support. It says, for these patients, you should consider ordering these extra lab tests, and extra screenings. But if you deliver high false-positive information to physicians, they just ignore that information. So you have to figure out how you implement the screening process, and which people need to act on it.

So, some of the lab tests in this area will come out as false positives?

No, the lab tests themselves are quite accurate. It is the conclusions that one draws from a collection of tests and observations that predict the probability of a disease that is more art than science. Alerts built on uncertainty naturally produce false positives. So if you deliver that alert: here are the 100 patients with abnormal vital signs and abnormal lab tests, and those patients may or may not be septic, the physician may say, “Well, OK, but some of those patients may have other reasons for having abnormal tests.” In fact, in organizations that do this well, this monitoring is often not done by the physicians, but by nursing or by a centralized sepsis monitoring in the hospital. One of the challenges is that we’re building our analytic capability while we’re implementing Epic, so we’re dealing with limited organizational attention span and resources.

And a lot of the stuff we’ve done around quality so far is very, very simple stuff, such as when people say to me, ‘Danny, we don’t know who in the hospital has congestive heart failure. We could find out via chart review. But can you tell me who has CHF based on those labs and medications?’ And sure, that’s easy. We have a process here called Transitional Care Management (TCM), which is about how to engage patients while in the hospital, and how to make sure that all the discharge and follow-up processes are engaged, to prevent readmissions and adverse events.

And we had that program in place, but the case managers just didn’t have the data. So they said, “Can you tell us who in the hospital has heart failure, diabetes, and COPD [chronic obstructive pulmonary disease]?” And I said, “Sure, that’s easy.” To date, a lot of the rate-limiting steps in some of these areas have been around lack of data. But now with EHRs—and I’ve trained on at least 15 different EHRs in the course of my practice—now, we’re transitioning into the next phase of, what are we going to do with the data? And this is where I see informatics making a shift away from building and designing systems, such as building CDS or interfaces or building a button here and there, to now, with EHR vendors picking up the pace, following implementation, the question will be, how do we effectively use the data for analytics and for population health?

What do you see as the biggest patient safety issues related to transitions of care?

I think that some people believe that patients just sort of just hang out in the hospital; but in reality, these days, if you’re in the hospital, you’re very sick. So everyone in hospitals is sick 24/7, and the biggest risks to transitions of care are adequate communications processes. Sometimes, clinicians are just following too many patients. And residents are often overwhelmed, and the many institutions rely on residents for the complete management of patients. In community hospitals, nurses take a lot of that burden off the physicians. In teaching hospitals, it’s often not having adequate processes of transitions of care. Maybe you’re working off an Excel spreadsheet, and the information isn’t being captured in a structured way.

But even with great EHRs, it’s challenging, because EHRs may not have the appropriate fields available, or they’re not structured appropriately, or the workflows and printouts are awkward. So it’s not that vendors can’t do it well, but that they haven’t chosen to do it. That’s why we provide a structured, standardized process for sign-outs with Ward Manager. It’s not rocket science; it’s a web-based dashboard that allows you to manage the patients that you’re following.

So, in the inpatient setting, the challenges of transitions of care are around structured process and structured data, and making things very, very easy. If you have a system that can generate a sign-out list, it may be 50 pages long. You may have 50 patients on your list, and having 50 pages in your pocket is not workable. Another big handoff is between inpatient and outpatient, around discharge. We don’t do that yet [at WardManager] but would like to. I’ve approached all of our clients with this functionality, and they all said it sounded good in theory, but had questions as to who should pay for the service as patients are being discharged to outpatient.

Might that change with the development of accountable care organizations?