Predicting readmissions is hard. Back in 2011, a research article in JAMA bluntly stated that “most current readmission risk models perform poorly.” That sounded like a challenge to me; I’m founder of Diameter Health and a data guy who likes tough problems. While not a physician, I’ve worked in healthcare for over a decade at large hospital systems and health IT companies. Penalizing 30-day readmissions was part of the Affordable Care Act (“Obamacare”) so I knew hospitals were starting to look for ways to improve. There are effective interventions to reduce readmissions rates (See Project Red from Boston University or Project BOOST from the Society for Hospital Medicine), the trouble is figuring out which patients to focus on. As you might guess, when patients leave the hospital, a minority pose the highest readmission risk. Do you think you could pick them out?
The first thing I learned when approaching the problem is that this is not a task for humans. An article titled “Inability of providers to predict unplanned readmissions” published in the Journal of General Internal Medicine found that nurses and doctors weren’t good at making predictions. There are a few reasons why: First clinicians don’t have easy access to the data needed to effectively predict readmissions. This requires a multivariate combination of laboratory, medication, past admission, vital sign, demographic and diagnosis data to do it well. Second, people are good at holding a few variables in working memory, like a telephone number, not the dozens of factors identified in most readmission models. Finally, who has the time? Most hospital nurses and physicians are already busy and don’t have the spare time to start doing logistic regression every day for every patient discharge.
So I say, let the computers compute! Wasn’t that the government’s intent in paying $15+ billion over the past several years for hospitals and physicians to adopt electronic health records (EHRs)? They wanted providers to use digital information to improve care, and reducing readmissions is a worthwhile goal. One in five Medicare patients readmits within 30 days of discharge, but patients would rather stay home and we would like them to stay healthy. So, now the question is what computer model to use. Five years ago, there weren’t very many models to choose from. Today, there are several ones, both published and proprietary, with varying accuracy.
In evaluating readmission models, the key measure to look at is the c-statistic. For the data nerds, that’s the area under the receiver operator curve that evaluates model specificity and sensitivity across thresholds. C-statistics range from 0.50 (complete chance) to 1.0 (perfect prediction) and readmission models with good discriminative capacity should be above 0.70. Nurses and doctors are in the 0.55-0.60 range and claims based models generally attain 0.60. The best clinical models I’ve seen exceed 0.75 (See Figure). Predicting readmission risk at the point of discharge is computable and possible today, but only if you can access the rich clinical detail in the electronic health record.
That’s where Diameter Health comes in. I founded Diameter Health with the goal of using digital extracts from EHRs to improve care. We do this through interoperability standards that have been advanced in the past few years, such as the Continuity of Care Document, an XML-formatted summary with key structured and codified data. I’ve published about the potential for such analytics in the American Journal of Public Health. Diameter Health is a startup working with innovative clinicians to prioritize patients at highest risk of readmission or disease progression. We’ve got our first pilot customer in Boston, and here’s how it works:
Our software runs through common web browsers and can be adapted to either a public (we host) or private (you host) cloud model. Diameter Health believes using digital information to improve patient care is the future of healthcare.
Big data and analytics lighten the load for providers by identifying patients at highest risk, but this alone doesn’t deliver results. Clinicians need to implement the right workflows for improved care transitions and post-hospitalization coordination. In addition, further research will clarify which post-discharge interventions work best based on individualized patient risk. Even with those limitations, readmission prediction is a valuable tool when deployed cost-effectively. Using interoperability standards and structured data required by federal EHR incentive programs is a viable approach to deliver return-on-investment when facing Medicare payment penalties. The average 2013 hospital readmission penalty was $125,000 and it’s scheduled to increase in 2014. Big data is a powerful tool to improve patient outcomes and reduce exposure to these financial penalties.
Allaudeen N, Schnipper JL, Orav EJ, Wachter RM, Vidyarthi AR. Inability of providers to predict unplanned readmissions. J Gen Intern Med. 2011; 26(7):771-776.
Amarasingham R, Patel PC, Toto K, Nelson LL, Swanson TS, Moore BJ, Xie B, Zhang S, Alvarez KS, Ma Y, Drazner MH, Kollipara U, Halm EA. Allocating scarce resources in real-time to reduce heart failure readmissions: a prospective, controlled study. BMJ Qual Saf. 2013 Jul 31. [Epub].
D’Amore JD, Sittig DF, Ness RB. How the continuity of care document can advance medical research and public health. Am J Public Health. 2012 May;102(5):e1-4. doi: 10.2105/AJPH.2011.300640.
Jencks SF, Williams MV, Coleman EA. Rehospitalizations among patients in the Medicare fee-for-service program. NEJM. 2009 Apr 2; 360(14):1418-28.
Kansagara D, Englander H, Salanitro A, Kagen D, Theobald C, Freeman M, Kripalani S. Risk Prediction Models for Hospital Readmission: A Systematic Review. JAMA. 2011 Oct 19;306(15):1688-98.
QualityNet Measure Methodology Reports for Readmission Measures. See multiple reports at www.qualitynet.org
John D'Amore has over a decade of experience providing informatics and strategic insight to healthcare organizations. He most recently founded Diameter Health, a health IT start-up to improve clinical, operational and financial outcomes through the intelligent use of data and predictive analytics. Previously, John was Vice President at Eclipsys (now Allscripts) overseeing enterprise performance management solutions. Before then, John worked at the largest health system in Texas overseeing clinical informatics, decision support and business intelligence. During his tenure, Memorial Hermann won accolades for its financial performance as well as the National Quality Forum Award for exceptional clinical care. John has published on best practices in population health and presented at national forums on how information technology can improve medical outcomes. He holds a biochemistry degree from Harvard University and a graduate degree in clinical informatics from the University of Texas, School of Biomedical Informatics.
Send this to a friend