Novel real-time technologies powered by predictive and text analytics can help organizations avoid readmissions—and wasted effort—by targeting the right interventions for patient populations.

At a Glance

The most effective readmission-prevention technologies include capabilities to:

  • Accurately predict risk and stratify patients
  • Synthesize compartmentalized data, both structured and unstructured, in real time and transform the data into actionable insights
  • Streamline workflow to shift time toward higher impact activities
  • Bridge care and communication across the continuum 

The Centers for Medicare & Medicaid Services (CMS) is on a campaign to eliminate waste, poor-quality care, and skyrocketing costs, imposing financial penalties on hospitals for readmissions. In 2013, two-thirds of hospitals faced such penalties, with the result that hospitals now have strongly aligned clinical and financial incentives to avoid readmissions and are actively seeking ways to do so.

Most hospitals and health systems have instituted some form of care management to reduce readmissions, but this approach often has proven surprisingly costly and ineffective. Many organizations also have implemented manual surveys for readmission risk, but this approach has proven to be labor intensive and often inaccurate.

For hospital executives, the efforts have been like playing darts blindfolded. As one hospital vice president of quality stated, “It has been hard to know what the right thing to do is and for which patients. We started a complex care management program, but it still seems like we are spinning our wheels with no traction.” 

Hospitals need both traction and targeting to truly make headway with readmission reduction. Success requires understanding that one-size interventions do not fit all patients and that targeting the right interventions to high-risk patients hinges on big data business intelligence. Progressive organizations have begun to leverage novel technologies enabled by big data, which identify high-risk patients in real time, segment disease-specific subpopulations for management, and provide targeted clinical decision support at the point of care. 

The Value of Big Data Technologies in Readmission Prevention

The application of big data to real-time intelligence technologies can become a sort of magnifying lens for hospitals. To date, readmission prevention technologies have shown limited success because they have lacked capabilities to target risk for readmission accurately or early enough to act, and to identify specific interventions that will work for individual patients. 

Meanwhile, hospitals have been allocating grant and capital money to set up “ideal” programs to reduce readmissions, including heart failure programs, follow-up phone-call programs, and skilled nursing facility outreach programs, only to find that these programs have limited impact and are difficult to sustain as funding runs out. With the development of more sophisticated big data analytics, progressive hospitals have made significant and sustainable strides in both improving care team productivity and reducing readmissions by using big data not only to identify patients at high risk for readmission early in the care process, but also to prompt interventions that could make a difference. 

Results of an intensive review by The Advisory Board Company indicate that the most effective readmission-prevention technologies include critical capabilities to:

  • Accurately predict risk and stratify patients
  • Synthesize compartmentalized data, both structured and unstructured, in real time and transform the data into actionable insights
  • Streamline workflow to shift time toward higher impact activities
  • Bridge care and communication across the continuum 

Exhibit 1


Accurately predicting risk and stratifying patients. Highly sensitive and specific predictive modeling capabilities are crucial. The problem is that most predictive models and tools currently on the market require significant manual input and overestimate the patient’s risk for readmission. Such overestimations can result in the system targeting the wrong patient, creating unnecessary tasks for the care team as crucial resources are brought to bear for a patient who does not need it. Meanwhile, a patient at-risk for readmission who is not identified in time will likely be readmitted. 

To be highly accurate, readmission-prediction technology not only requires access to a large volume of data, but also must be customized to the local patient population of each individual hospital. Some individual health systems are attempting to build their own large-scale data repositories, but they rarely can achieve the scale required for high specificity and sensitivity. The technology also must be a dynamic machine that can learn, adapt, and evolve over time with improved accuracy and insights.

Including clinical data feeds such as vitals, lab results, and medications can make predictive models even more accurate. The accuracy of predicting readmissions can be boosted from about 70 percent using admission, discharge, and transfer (ADT) and claims data alone to nearly 80 percent with additional real-time clinical data feeds. With the addition of text data, often drawn from case manager and nursing notes, sensitivity and specificity can reach nearly 90 percent. Boosting accuracy helps sub-specify the population and contain cost by deploying specific interventions to those patients who will most benefit from them. Ultimately, the aim of predictive modeling at the point of care is to provide flexible, real-time support and to know when, where, how, and for whom to intervene. 

Exhibit 2


Synthesizing compartmentalized data, both structured and unstructured, in real time and transforming the data into actionable insights. Readmissions technology should synthesize compartmentalized data in electronic health record (EHR), billing, and information systems and transform these disparate data into actionable insights. Some EHR vendors are attempting to create predictive models, but their data tend to be siloed, obscuring a complete picture of readmission risks and needs. 

Because it is especially difficult to access, unstructured data, such as nursing narratives and progress notes, frequently goes unused in predictive modeling. Ashish Jha, MD, MPH, comments in an editorial in the August 2011 issue of JAMA: “(U)ntil now, most of the benefits of EHRs have come from clinical decision support, such as offering advice on avoiding two drugs with serious drug-drug interactions. Yet, there is so much more that they could do. … Much of the value resides in unstructured clinician notes rich in detail about signs and symptoms, patient’s response to treatment, and other details key to understanding the patient’s condition” (“The Promise of Electronic Records: Around the Corner or Down the Road?”).

Increasingly, text analytics are being applied to mine these rich data sources and generate a full picture of readmission risk. Both academics and front-line clinicians agree that psychosocial variables add significant accuracy to a predictive model and also help identify more targeted interventions. However, these factors are almost always buried in unstructured information and must be carefully teased out and interpreted to yield accurate results. 

The extraction, interpretation, and transformation of psychosocial text data into insights pose significant challenges for health systems and technology developers alike. Extracting data requires technical expertise, interpreting data requires text analytics, and transforming data into relevant and actionable insights requires clarification of context-related ambiguities in the source data, statistical accounting of variables, and most important, clinical validation. 

The Advisory Board Company has been working with an interdisciplinary team of mathematicians, clinicians, and developers to create these capabilities within a real-time readmission application. Funded by the National Institutes of Health, the initiative’s review of several million patient encounters has demonstrated that psychosocial risk factors play a key role in readmissions as well as repeated emergency department (ED) visits after all-cause, heart failure, pneumonia, and myocardial infarction index discharges. When isolated as individual patient risk factors, psychosocial issues such as homelessness, forgetfulness, financial distress, depression, and malnutrition suggest very different clinical pictures. A homeless patient with methamphetamine-induced heart failure requires very different interventions than the typical elderly patient with heart failure, even though both patients present with heart failure.

Streamlining workflow to shift time toward higher impact activities. Readmissions technology should help to streamline and organize workflow by making key insights available at the earliest possible moment in the patient experience, thereby shifting the focus of the care team toward activities with the highest impact. Eliminating the need for manual chart review, presenting information to clinicians in a way that is quickly actionable, suggesting the right interventions for the right patients, and reducing documentation workload all contribute to transforming work time from transactional to productive. An effective model should enable high-risk patients to be identified right in the ED or on admission, hours before care managers could target such patients, if they could complete a chart review at all. The automation of this effort could reduce the 30 to 40 minutes required for thorough chart review to a few minutes.

Care management departments have long struggled with subpar and poorly interconnected systems and an increasing volume of patients that cause tasks to “fall through the cracks.” When provided systems that integrate information with workflow, these departments excel. One Midwest healthcare organization testing real-time technology saw a 60 percent reduction in chart review time and a 14 percent reduction in heart failure readmissions on its pilot unit. A progressive West Coast hospital system benefited from a 65 percent reduction in chart review time and a $10,000 average savings per case manager per year on its pilot unit. 

Major efforts such as Reengineered Hospital Discharge (Project RED), Project BOOST (Better Outcomes for Older Adults Through Safe Transitions), and Transitions in Care have consistently shown that interventions can reduce readmissions. However, we have yet to clearly understand which specific interventions work best—and for which patient risk factors. Ultimately, the end goal of technologies fueled by big data should be a smart feedback loop: not only enabling “smarter workflow” at the point of care, but also leading individual hospitals to understand which protocols are optimal for their local patient population. 

Bridging care and communication across the continuum. To be successful, technology must bridge and support the transition from inpatient to outpatient care environments. Many successful interventions are initiated on the inpatient side and then continued or completed on the outpatient side. One primary care physician interviewed for our study expressed his frustration at having no knowledge of his patient’s hospital stay, particularly the patient’s medication regimen, which had been significantly altered upon discharge home. Medication reconciliation, patient teaching failures, nonadherence risks, follow-up call content and results, and home care needs are all transitional issues for care teams and patients. Real-time business intelligence can facilitate success in bridging these care processes between care settings.

Making Big Gains with Big Data

It is time for hospitals to use big data to their strategic advantage. CMS will continue to escalate readmission penalties, and commercial payers are following suit as well. Real-time business intelligence technologies can assist hospitals in reducing their readmissions, yet thoughtful leaders understand that single-event technologies lack the staying power needed for true transformational strength. 

Readmissions are likely only the tip of the iceberg. There are many other applications for which real-time technologies can improve clinical outcomes (sepsis early warning), reduce healthcare costs (medical necessity), and improve quality indicators (treatment adherence), just to name a few. Novel real-time technologies powered by predictive analytics and text analytics are here to stay, and they will increasingly assist organizations in targeting the right interventions to optimal patient populations, minimizing low-impact work, and ultimately, improving quality as well as financial outcomes. 

Barbara S. Harvath, RN, BA is senior director, strategic planning, The Advisory Board Company, Washington, D.C..

Karoline Hilu, MD, MBA, is senior director, strategic planning, The Advisory Board Company, Washington, D.C..

Ravi Nemana, MBA, is managing director, strategic planning, The Advisory Board Company, Washington, D.C..

Ramesh Sairamesh, MPhil, PhD, is managing director, strategic planning, The Advisory Board Company, Washington, D.C..

Publication Date: Monday, December 02, 2013

Login Required

If you are an existing member, please log in below. Username and password are required.



Forgot User Name?
Forgot Password?

If you are not an HFMA member and would like to access portions of our content for 30 days, please fill out the following.

First Name:

Last Name:


   Become an HFMA member instead