Skip Navigation
Service Delivery Innovation Profile

Hospital Uses Data Analytics and Predictive Modeling To Identify and Allocate Scarce Resources to High-Risk Patients, Leading to Fewer Readmissions


Tab for The Profile
Comments
(0)
   

Snapshot

Summary

Parkland Health & Hospital System, a safety net provider in Dallas, TX, uses a system of data analytics and predictive modeling developed by PCCI (a nonprofit affiliate of Parkland) to analyze a hospital’s electronic medical record data and identify patients at high risk of readmission. On a near real-time basis, case managers and other frontline providers receive the names of these patients and in some cases risk scores that quantify the likelihood of readmission. They use this information to prioritize their work, allocating scarce resources to support those most in need, both in the hospital and after discharge. Originally developed for congestive heart failure, the system has been expanded to incorporate other diseases, including diabetes, acute myocardial infarction, and pneumonia. The program has reduced readmissions and costs for congestive heart failure patients; preliminary, unpublished data suggest similar trends among patients with the other targeted conditions.

Evidence Rating (What is this?)

Moderate: The evidence consists of pre- and post-implementation comparisons of readmission rates within 30 days of discharge among CHF patients served by the program, with additional comparisons to similarly high-risk pneumonia and AMI patients not served by the program. Additional evidence includes post-implementation analyses of the ability of several of the program’s predictive models to accurately identify at-risk patients and estimates of the cost savings associated with the decline in CHF readmissions that resulted from the program.
begin do

Developing Organizations

Parkland Health & Hospital System; PCCI
end do

Use By Other Organizations

Texas Health Resources, which operates more than 20 private hospitals throughout the Dallas metropolitan area, began using Pieces™ on July 1, 2013.

Date First Implemented

2009
Parkland began using the system with congestive heart failure patients in 2009.

Problem Addressed

Recently hospitalized patients are often discharged with complex health care needs and/or experience complications that lead to readmission. Many of these problems can be prevented if high-risk patients are identified and provided with support in the hospital and immediately after discharge. Yet many hospitals lack adequate information to identify those most at risk, and as a result scarce resources are allocated to patients on a first-come, first-served (rather than need) basis.
  • High readmission rates among at-risk patients: Approximately 1 in 5 Medicare beneficiaries discharged from the hospital are readmitted within 30 days, and more than one-third are rehospitalized within 90 days.1 Readmission rates are even higher for chronically ill seniors, particularly those with multiple comorbidities, functional and cognitive impairments, emotional problems, and poor health behaviors.2
  • Often avoidable with adequate support: A review of nearly 100 studies indicates that one-quarter to one-third of readmissions among older patients could be prevented.2 Readmissions can be avoided if patients and their family members have adequate support after admission, including help in understanding how to follow complicated medication regimens and when and how to obtain periodic followup care from different providers.3
  • Inadequate information to identify those most in need of support: Case managers and other frontline providers often lack the information needed to identify hospitalized patients who are at high risk of readmission and hence could most benefit from inhospital and postdischarge support designed to reduce that risk. Without such information, they typically perform their work on a “first-come, first-served” basis rather than specifically targeting those who most need their support.

What They Did

Back to Top

Description of the Innovative Activity

Parkland Health & Hospital System uses a system of data analytics and predictive modeling developed by PCCI to analyze a hospital’s electronic medical record (EMR) data to identify patients at high risk of readmission. On a near real-time basis, case managers and other frontline providers receive the names of these patients and in some cases risk scores that quantify the likelihood of readmission. They use this information to prioritize their work, allocating scarce resources to those most in need, both in the hospital and after discharge. Originally developed for congestive heart failure (CHF), the system has been expanded to incorporate other diseases, including diabetes, acute myocardial infarction (AMI), and pneumonia. Key program elements are outlined below:
  • Near real-time analysis of EMR data to identify patients: The system reviews EMR data to identify patients with specific illnesses or other circumstances that put them at high risk of readmission. At present, the system identifies those with any of four targeted conditions: CHF, diabetes, AMI, and pneumonia. Because valuable information often exists outside the structured fields of the medical record, the system uses natural language processing to evaluate information found only in free-text fields, such as provider notes.
  • Assigning risk score: For patients with diabetes and CHF, the system assigns a risk score indicating the likelihood of readmission within 30 days. To generate this score, the system relies on predictive models that integrate information not only on the patient’s medical condition, but also on other factors that might increase risk, such as having limited social support or an unstable living situation. To make this determination, the system searches all fields of the EMR (including free-text notes), incorporating information such as the number of emergency contacts and family relationships listed and the frequency of changes in home address.
  • Providing near real-time information to case managers: Case managers can see which of their assigned patients have one of the four targeted conditions on a near real-time basis. For those with CHF or diabetes, they also see the assigned risk score. Originally case managers received these lists each day, following a nightly sweep of the EMR data. The system later changed to incorporate ongoing review of Health Level 7 (more commonly known as HL7) messages, enabling the identification and risk stratification to be done on an ongoing basis.
  • Daily prioritization of patients: Case managers and others in the hospital use both the patient lists and, where available, the risk scores to prioritize their work. Each morning, they use a 20-minute rounding period to review newly identified patients within the four disease categories and prioritize those patients based on the information provided. This system replaced the old approach in which new patients went to the bottom of the queue and case managers and others organized their work on a first-come, first-served basis, working their way down the list.
  • Intensive support to those most at risk: Case managers and other frontline providers give intensive support to the highest-risk patients, both during their inpatient stay and for the first 30 days after discharge. (Other patients who are not at high risk receive usual care.) For example, the highest-risk CHF patients—typically about one-quarter of all hospitalized patients with the disease—are offered a set of inpatient and postdischarge support services from nurse practitioners, pharmacists, nutritionists, and case managers, with patients free to accept or reject any service. A summary of key services appears below:
    • Inhospital medication reconciliation and education: Hospital-based pharmacists perform medication reconciliation with high-risk patients, and educate them to ensure they understand their postdischarge medication regimen and the importance of adhering to it. They also explain potential side effects and adverse reactions, including what to do if they manifest.
    • Inhospital disease management education: As appropriate, case managers, nutritionists, or other staff engage in self-management education for CHF patients. These sessions emphasize the importance of appropriate diet (such as limiting sodium); physical activity; and other behaviors, including daily weight measurement.
    • Inhospital scheduling of followup care: Before discharge, the case manager proactively schedules followup specialty or primary care appointments, with the goal of ensuring that specialty visits occur within 1 week and primary care visits within 30 days of discharge. The hospital works closely with its affiliated specialty and primary care clinics to offer appointments within these time parameters.
    • Postdischarge case management support: Transitional care nurses and case managers proactively follow up with the patient after discharge. Nurses call within 48 hours of discharge to confirm that the patient has obtained his or her medications. During these calls, they also ensure that the patient knows about and has transportation to and from any scheduled followup appointments. On an ongoing basis for 30 days after discharge, case managers provide customized support to patients, proactively following up and otherwise supporting them based on their needs. In addition, patients can call their assigned case manager at any time.

Context of the Innovation

Parkland Health & Hospital System is one of the nation’s largest safety net hospitals and health systems; Parkland operates the only public hospital in Dallas County, TX. PCCI (formerly known as the Parkland Center for Clinical Innovation) is a nonprofit affiliate of Parkland that focuses on data analytics and predictive modeling.

The impetus for this program began in 2007, when Ruben Amarasingham, MD (who at the time served as an assistant medical director for medical services) began conducting reviews of the medical charts of CHF patients who had been readmitted to the hospital. He noticed that these patients often had fragile social networks or unstable living situations. This process also made him realize that medical records contain important clues that can help identify patients at highest risk of readmission. With several colleagues, Dr. Amarasingham began a more systematic search of the hospital’s EMR data to identify such clues, including patients with frequent changes in their home address or a small number of emergency contacts or family relationships.

Did It Work?

Back to Top

Results

The program quickly and accurately identifies those with conditions or other factors that put them at high risk of readmission, and has reduced readmissions and costs for CHF patients. Preliminary, unpublished data suggest similar trends among patients with the other targeted conditions.
  • Accurate, timely identification of those most at risk: An external review of various models found PCCI’s approach to be among the best for identifying CHF patients at high risk of readmission.4 A separate analysis concluded that Parkland’s model also accurately identifies patients with diabetes.5
  • Fewer readmissions, lower costs for CHF patients: Between December 1, 2008 and December 1, 2010, the proportion of Parkland’s CHF patients readmitted to any Dallas area hospital within 30 days of their initial discharge fell by 19 percent (from 26.2 to 21.2 percent),6 and the proportion of Medicare CHF patients readmitted fell by 31 percent (from 29.9 to 20.5 percent).4 The likelihood of readmission declined as the intensity of the support services increased, with those receiving the most services (especially after discharge) being the least likely to be readmitted.6 Over the same time period, admissions among similar high-risk pneumonia and AMI patients (who did not receive additional support during the study period) did not meaningfully change.6 The decline in readmissions among Medicare CHF patients has saved the hospital an estimated $500,000.4
  • Similar trends in other diseases and conditions: Preliminary, unpublished data suggest that similar, positive trends are occurring in the other diseases and conditions targeted by the program.

Evidence Rating (What is this?)

Moderate: The evidence consists of pre- and post-implementation comparisons of readmission rates within 30 days of discharge among CHF patients served by the program, with additional comparisons to similarly high-risk pneumonia and AMI patients not served by the program. Additional evidence includes post-implementation analyses of the ability of several of the program’s predictive models to accurately identify at-risk patients and estimates of the cost savings associated with the decline in CHF readmissions that resulted from the program.

How They Did It

Back to Top

Planning and Development Process

Key steps included the following:
  • Developing model to identify at-risk patients: In 2008, Dr. Amarasingham and his colleagues created a model for analyzing EMR data before discharge to identify CHF patients at high risk of readmission.7
  • Creating software application: Based on this work, Dr. Amarasingham and his colleagues developed a a software application to interface with Parkland’s EMR system. Launched by Parkland and known as Pieces™, this system runs the models to generate the real-time information about patients' risk of readmission.
  • Educating and collaborating with community-based clinics: Because the success of this program depends on at-risk CHF patients being seen quickly after discharge for followup appointments, Dr. Amarasingham met with leaders of Parkland-affiliated specialty and primary care clinics to inform them about the new prioritization strategies being used in the hospital and the corresponding need to see at-risk patients in a timely manner after discharge. Clinic leaders generally liked the approach and agreed to see these patients quickly, in some cases setting up block-scheduling programs for them.
  • Forming nonprofit center to develop additional models: Buoyed by the initial success with CHF, organizational leaders launched a nonprofit research and development corporation to focus on expanding the approach. Opened in 2012 and known as PCCI, this center replaced a smaller, internal department within Parkland that had been conducting the work on the CHF applications and related areas. With 57 staff members, PCCI actively develops additional models for identifying patients at high risk of readmission. In addition to the now completed and implemented models for diabetes, AMI, and pneumonia, PCCI staff have developed and are currently testing models to predict readmission risk for patients with cirrhosis, HIV, chronic kidney disease, and sepsis; an “all-cause” readmission risk model; and a model to identify patients at high risk of death due to cardiopulmonary arrest. In total, the center has approximately 30 models in development that cover a broad array of specialties and settings, including surgery, medicine, pediatrics, and outpatient conditions.
  • Making platform available to others at low cost: PCCI is working to spread use of Pieces™ and other applications to interested users at low cost. With the support of grant funding, PCCI will be making these software applications available as a service, running the platforms remotely from PCCI rather than requiring that they be installed locally at the implementing site.
  • Creating local health information exchange: As an extension of this initiative, program leaders are working with others to spearhead development of the Dallas Information Exchange Portal. The goal of this work is to create a seamless system for the flow of information across hospitals, clinics, and community-based organizations, thus facilitating the care of underserved, high-risk patients as they move between these settings.

Resources Used and Skills Needed

  • Staffing: The development of each disease- or condition-specific program, including the predictive model, related software, and associated workflow changes, requires the work of a large team over several months. Each team is made up of information technology staff, statisticians, and physician scientists who perform this work as part of their regular job responsibilities. On an ongoing basis, each disease- or condition-specific initiative has required the hiring of roughly 0.5 full-time equivalent case managers.
  • Costs: Data on program-related development and operating costs are not available. As noted earlier, the CHF program has generated significant cost savings for the hospital.
begin fsxml

Funding Sources

National Cancer Institute; National Institutes of Health; Commonwealth Fund; Gordon and Betty Moore Foundation; University of Texas; W.W. Caruth Jr. Foundation
The University of Texas and the Commonwealth Fund provided funding to support development and evaluation of the CHF program.

The Commonwealth Fund, the Gordon and Betty Moore Foundation, the W.W. Caruth, Jr. Foundation, the National Institutes of Health (including the National Cancer Institute), and other organizations have provided more than $9.3 million in grants and donations to PCCI and Parkland to support use of data analytics and health information exchange to improve care, including but not limited to developing models and algorithms to identify patients at high risk for readmissions and other adverse events.end fs

Tools and Other Resources

More information on PCCI and Pieces™ is available at: http://www.pccipieces.org/.

Organizations interested in learning more about using PCCI’s software applications should contact the innovator.

Adoption Considerations

Back to Top

Getting Started with This Innovation

  • "Sell" program to hospital leadership: The program’s success depends in large part on the willingness of institutional leaders to support it, both by allocating resources (staff and funding) and making it a priority in internal and external communications. To that end, share information with program leaders about the significant quality and cost problems associated with readmissions and the ability of this type of approach to address these problems effectively.
  • Target conditions in which hospital stands to benefit financially: Aware of plans by the Centers for Medicare & Medicaid Services (CMS) to institute payment penalties for hospitals with high readmission rates in certain conditions (including CHF), program leaders decided to focus their early efforts on these conditions, thus increasing the likelihood that the hospital would benefit financially from any successes achieved. Similar incentives may be available to public hospitals and other organizations that participate in CMS’s waiver program under Section 1115 of the Social Security Act, which allows disproportionate share hospitals to receive incentive payments by improving key metrics for Medicaid enrollees, such as readmission rates.
  • Enlist community support: As noted, program success depends on the ability of community-based organizations (e.g., specialty clinics, primary care practices) to serve at-risk patients in a timely manner after discharge. Consequently, institutional leaders should meet with representatives of these organizations before the program’s launch to gain their buy-in and commitment.
  • Start small and expand over time: As noted, Parkland and PCCI began with CHF and then expanded to other conditions and diseases once success in this area had been demonstrated.

Sustaining This Innovation

  • Regularly monitor and share data on program’s impact: To maintain momentum and support, monitor key process and outcomes metrics on an ongoing basis, and share data on the program’s impact with frontline clinicians, department heads, and institutional leaders.
  • Maintain regular communications with community-based providers: To ensure that patients get timely care after discharge, periodically meet with community-based providers to identify, discuss, and address any operational issues or other problems stemming from the program.
  • Leverage scalable processes: As noted, PCCI is creating systems to disseminate software applications to other sites (both inside and outside the organization) at relatively low cost, thus facilitating expansion of the approach.

Use By Other Organizations

Texas Health Resources, which operates more than 20 private hospitals throughout the Dallas metropolitan area, began using Pieces™ on July 1, 2013.

More Information

Back to Top

Contact the Innovator

Elizabeth Dwelle, PhD
Communications and Media Manager
PCCI
6300 Harry Hines Boulevard
Dallas, TX 75235
(214) 590-4339
E-mail: elizabeth.dwelle@phhs.org

Innovator Disclosures

Dr. Dwelle reported having no financial interests or business/professional affiliations relevant to the work described in the profile other than the funders listed in the Funding Sources section.

References/Related Articles

Amarasingham R. Applying data analytics and information exchange to improve care for patients. Health Aff. 2012;31(12):1-2. [PubMed]

Amarasingham R, Moore BJ, Tabak YP, et al. An automated model to identify heart failure patients at risk for 30-day readmission or death using electronic medical record data. Med Care. 2010;48(11):981-8. [PubMed]

Makam AN, Nguyen OK, Moore B, et al. Identifying patients with diabetes and the earliest date of diagnosis in real time: an electronic health record case-finding algorithm. BMC Med Inform Decis Mak. 2013;13:81-7. [PubMed]

Amarasingham R, Patel PC, Toto K, et al. Allocating scarce resources in real-time to reduce heart failure admissions: a prospective, controlled study. BMJ Quality Saf. 2013;22(12):1-8. [PubMed]

McAlister FA. Decreasing readmissions: it can be done but one size does not fit all. BMJ Qual Saf. 2013;22(13);998-1005. Available at: http://m.qualitysafety.bmj.com/content/early/2013/09/04/bmjqs-2013-002407.full.

Footnotes

1 Jencks SF, Williams MV, Coleman EA. Rehospitalizations among patients in the Medicare fee-for-service program. N Engl J Med. 2009;360:1418-28. [PubMed]
2 Naylor MD. Transitional care for older adults: a cost-effective model. LDI Issue Brief. 2004;9(6):1-4. [PubMed]
3 Naylor MD. Transitional care of older adults. Annu Rev Nurs Res. 2002;20:127-47. [PubMed]
4 Amarasingham R. Applying data analytics and information exchange to improve care for patients. Health Aff. 2012;31(12):1-2. [PubMed]
5 Makam AN, Nguyen OK, Moore B, et al. Identifying patients with diabetes and the earliest date of diagnosis in real time: an electronic health record case-finding algorithm. BMC Med Inform Decis Mak. 2013;13:81-7. [PubMed]
6 Amarasingham R, Patel PC, Toto K, et al. Allocating scarce resources in real-time to reduce heart failure admissions: a prospective, controlled study. BMJ Quality Saf. 2013;0:1-8. [PubMed]
7 Amarasingham R, Moore BJ, Tabak YP, et al. An automated model to identify heart failure patients at risk for 30-day readmission or death using electronic medical record data. Med Care. 2010;48(11):981-8. [PubMed]
Comment on this Innovation

Disclaimer: The inclusion of an innovation in the Innovations Exchange does not constitute or imply an endorsement by the U.S. Department of Health and Human Services, the Agency for Healthcare Research and Quality, or Westat of the innovation or of the submitter or developer of the innovation. Read more.

Original publication: January 29, 2014.
Original publication indicates the date the profile was first posted to the Innovations Exchange.

Last updated: January 29, 2014.
Last updated indicates the date the most recent changes to the profile were posted to the Innovations Exchange.