Augusta Health is a 255-bed community hospital in Fishersville, Virginia, located in the Blue Ridge Mountains about 35 miles west of Charlottesville, in the western north-central part of that state. The hospital’s service area encompasses a population of about 200,000, spread out across six predominantly rural counties. Clinical and administrative leaders at Augusta Health, anticipating changes in Medicare reimbursement based on the mandatory readmissions reduction program instituted under the Affordable Care Act (ACA), began a proactive effort in 2011 to quantify readmission risks for patients and to use extracted data to intervene clinically.
At Augusta Health, one key element in improving its readmissions profile, in anticipation of the coming reimbursement changes, has been to develop a tool to stratify the risk of individual patients for readmission, with the goal of most efficiently allocating hospital resources by capturing existing data from electronic documentation without manual review, and using that data to clinicians in a position to intervene on behalf of high-risk patients.
Among those leading the effort have been Roger Gildersleeve, M.D., a hospitalist and the organization’s chief medical information officer, and Penny Cooper, who is the integration services/decision support manager for Augusta Health. Gildersleeve and Cooper spoke recently with HCI Editor-in-Chief Mark Hagland regarding their current work in this area. Below are excerpts from that interview.
Can you describe the origins of your work in this area?
Roger Gildersleeve, M.D.: Like a lot of hospitals, there was a lot of interest in understanding readmissions, particularly because of interest on the part of CMS (the federal Centers for Medicare and Medicaid Reimbursement) in modifying reimbursement for readmissions. Fred Castello, MD, our Chief Medical Officer was early to catch on to the importance of this, and he moved us along to action. A 2009 New England Journal of Medicine article helped shape our understanding and structure our thinking.
What led you to apply IT to this?
Over the last 15 years, people have realized the pitfalls of making decisions based on gut impressions, or making ad hoc decisions. There’s been increasing interest in using data to drive decisions, and Penny has been the person whom everyone has turned to here, with regard to the data repository.
How long have you had a data repository?
Penny Cooper: We have had a data repository for about 10 years and so we’ve been using data to a greater degree than the typical community hospital. We have a strong interest in using the data to its fullest and have the talent here to extract and analyze it. For this project we completed a two-year study, one year derivation and one year validation, built the model based on the study’s results and implemented it. We started out by reviewing recently published studies on readmissions and found that everyone was doing something different. We extracted various patient demographic as well as clinical variables, compared them with our own data, and came up with the 12 variables to analyze, 3 of which after further analysis were not significant.
What are the nine variables you ended up with?
The variables that were significant for our patient population are:
> acute admission
> Charleston co-morbidity index, adjusted for age (a frequently used tool)
> ER visits within the past 365 days
> inpatient admissions within the past 365 days
> current length of stay for the current analysis, and total length of stay for study period
> male gender
> inpatient medication count two days prior to discharge
> outpatient medication count prior to admissions
> self-pay patient or not
Which were the very most significant of the nine?
Gildersleeve: An unplanned or Acute admission was the single most significant predictor.
Cooper: Depending on their frequency of occurrence though Inpatient admissions may carry more weight for a particular patient or the Charlson Comorbidity Index score may be higher for a very ill patient.
Gildersleeve: The final score depends on multiple weighted variables more than any one alone.
What kinds of actions will be triggered?
Gildersleeve: So far, we have not yet released this tool for viewing by clinicians. I’ve informally polled clinicians about how they’d use this tool, and they’ve said, well, I’d keep the patient in the hospital longer. And in fact there might be an undesirable effect in keeping the patient in the hospital longer, since we found that longer length of stay is actually associated with a higher risk of readmission. On a higher level, we have a healthcare reform committee focused on reducing inappropriate readmissions, which is overseeing a new complex patient clinic. Case managers will use this tool to detect currently admitted patients who are at a high risk of readmission. Their attention may mean medication education, home health visits, and so on, in addition to referral to this clinic for select patients.
When might you actually release this for viewing by active clinicians?
Gildersleeve: That’s hard to say. We’ll start discussions in different department meetings over the next few months. As you know, clinicians already have so many things to be looking at.
Would the first clinicians using this during patient care be hospitalists?
In the meantime, you’re going to continue to look at patterns and do statistical analysis?
Yes. We’ve completed our validation cohort, but next up will be prospectively validating this tool.
Cooper: Currently we are writing our daily stats to a file including what a patient’s score was as well as the variables that made up the score. Next we’ll review patients with similar scores and evaluate their readmission status. We’ll continue this for the foreseeable future to verify the predictive capability of the tool.
What have you most learned that might be applicable in other settings?
Gildersleeve: Two things come to mind. One surprise that we’re still discussing is that ambulatory medications pre-admission are a favorable risk factor,-- negatively correlated to readmission. I didn’t believe that at first, and we reran it to validate it. Our best explanation is that a patients’ reporting of medications is a sign of compliance and also the fact that a physician is actively managing their problems. So, this should emphasize that medication education and compliance are in fact as important as everyone thinks they are. The second thing is how enthusiastically people responded to this tool. People of all disciplines who have seen this are excited by having an automated, real-time surveillance tool, which will encourage us to do more of these projects across the system.
Cooper: Part of the excitement over this was that the process is totally automated (including the creation of the Charlson Comordity Index) and doesn’t add more work for anyone. The Charlson Comordity Index is made up of an age adjusted review of ICD9 codes over the past year.
Any advice for organizations that might try to replicate something like this?
When it comes down to it, it wasn’t really that difficult to accomplish. Most hospitals have a considerable archive of their patient data. We extracted our data and ran simple statistical analysis against it… So, basically I would encourage those positioned to do so to make the best use of what they have.
Gildersleeve: Exactly. Make use of what you already have. And don’t be afraid of statistics and statistical analysis.
Cooper: The statistical package we used is MedCalc, which is inexpensive to purchase and easy to use.
NOTE: Readmissions is one of the important topics that will be discussed at the Healthcare Informatics Executive Summit, to be held May 6-8, in Orlando, Florida. Don't miss out on Session E07, "Readmissions and the Medical Home: Re-Visioning Care Management."