Skip to main content
Skip to footer
Hospital entrance sign

Q&A: How can predictive algorithms reduce hospital readmissions?

Published on 10/28/2019

Estimated read time: 6 minutes

Hospital readmissions ─ when patients who had been discharged from a hospital are admitted again within a specific timeframe ─ are bad news for everyone. Readmissions put stress on patients and cost Medicare about $26 billion annually, with $17 billion of that considered avoidable.

In this Q&A, we talk to Dayle Unger, clinical ITG advisor at Encompass Health, an integrated health care service provider specializing in inpatient rehabilitation, home health and hospice services. Dayle shares how Encompass Health is using a predictive analytics algorithm and other artificial intelligence tools to keep patients healthy and out of the hospital. 

Q: What is the impact of readmissions on patients, providers and health care organizations?

A: Hospital readmission is costly and impacts quality-of-care standards as well as patient satisfaction metrics. It’s challenging for any patient, but especially for those who come to an inpatient rehabilitation hospital. It can be extra frustrating after they’ve already completed a lengthy hospital stay to treat the injury or condition that led them to rehab in the first place. Readmission can easily derail a patient’s functional recovery even more.

Q: What are some common reasons patients are readmitted to the hospital? Why do these issues exist?

A: From the data we’ve studied, cardiac, pulmonary or infectious issues that can’t be managed within the inpatient rehabilitation setting are the top reasons a patient readmits to the acute hospital. Although a patient may enter an inpatient rehabilitation facility due to a new injury or illness, the chronic condition(s) the patient has managed for years may be exacerbated after experiencing a new injury or illness and can trigger complications, which may lead to an acute hospital readmission. 

Q: Are there certain populations that are more at risk for readmissions? If so, who are they, and what can be done to reduce their risks?

A: More time is needed to fully understand if specific populations of patients are more at risk than others. At Encompass Health, we now have a massive clinical database that we can use to continue to study distinct patient populations and their particular risks for readmissions. Traumatic brain injury, spinal cord injury, major multiple trauma (one or more serious injuries to areas such as the head, face, chest, spine, abdomen etc.), stroke and other neurological conditions are some examples of the rehabilitation impairment categories. Admission to an inpatient rehab facility (IRF) requires specific criteria be met. Any patient admitted to an IRF is medically compromised from a complex condition like a stroke or a traumatic brain or spinal cord injury. As mentioned before, many of these patients who are admitted to a post-acute setting, specifically an IRF, have a new onset of a condition and, oftentimes, multiple chronic conditions. Reviewing some early data has shown patients with neuro conditions or injuries may be at higher risk for readmission.

Q: What’s the Encompass Health ReACT model, and how is it helping your organization be more proactive in keeping patients healthy at home and out of the hospital?

A: The ReACT predictive model was developed to identify patients early who might be at risk of being transferred to an acute care facility. The goal is to give clinicians an earlier opportunity to intervene before a medical situation advances to the point where a transfer back to the hospital is the only option.

While predictive modeling with big data has been used by leading acute care health systems for some time, use in the post-acute space has been limited. Leveraging Encompass Health’s inpatient rehabilitation expertise and Cerner’s background with readmission prevention work in acute care, we created a team to develop a model that can be deployed at the point of care and uses original research, machine learning and statistical analysis. The team studied two years of inpatient rehabilitation data and more than 80,000 clinical records, and identified 30 predictors of risk, including appetite, vital signs, missed therapy, medications and more. The model was piloted and is now deployed to all 133 Encompass Health inpatient rehabilitation hospitals.

The ReACT model runs within Encompass Health’s ACE IT system ─ a proprietary version of Cerner Millennium® ─ throughout each patient’s stay. The system categorizes the patient’s level of risk of an acute care transfer (ACT) – from low to high or very high – for an unplanned transfer back to acute care. The model runs in near real-time, and a patient’s ReACT risk level will change as new information is entered into the system. If a patient is categorized as very high risk at any time, an alert is triggered for the patient to receive an additional clinical assessment, which may lead to enhanced monitoring and treatment.

The use of ReACT, along with our ACT prevention strategies, has empowered Encompass Health to achieve positive trends in both its ACT and discharge to community rates. In June 2019, we had our lowest ACT rate in more than a decade and our second highest discharge to community rate, despite serving more complex patients during this period.

Q: Cerner’s Sepsis Management solution is often used in the acute setting, but Encompass Health is using it to improve post-acute care. Why is Encompass Health using this tool, and how is it helping to reduce hospital readmissions rates?

A: One of our key strategies is to leverage all the data now available to us through our electronic health record and use it in various ways to improve our level of patient care and the overall outcomes of our patients. A condition like sepsis, if not identified early, would mean a return to the acute setting. Since Cerner already had a sepsis management solution available, it only made sense that we take advantage of it in the post-acute setting. Patients are coming to us earlier and earlier and with increasingly complex medical conditions. At Encompass Health, we average 5,000 sepsis alerts a month. Those alerts lead to detailed assessments of the findings and early intervention or treatments. Of the alerts received and subsequent intervention, we estimate that we‘ve been able to prevent sepsis more than 75% of the time.

Q: Are there other ways Encompass Health is using AI to enhance patient care?

A: Absolutely! Following the success of ReACT, Encompass Health and Cerner furthered their partnership in August 2017 with an agreement to form the Post-Acute Innovation Center. The purpose of the Innovation Center is to work together to improve all aspects of post-acute care. This will be done through developing evidenced-based technology and employing data sets from multiple care settings. Currently, we’re developing an advanced machine learning and comparative statistical analysis tool to identify patients at-risk for readmission across post-acute settings.

Q: What can the health care industry do to reduce readmission rates?

A: For patients at an inpatient rehabilitation hospital or any post-acute setting, one of the most negative emotional and clinical impacts occurs when a patient must transfer back to the acute care setting. Transfers are frustrating for patients who want to return home, as well as costly for all parties involved – patients, payers and the rehabilitation hospital — given capitated value-based payments. We must better understand why patients are re-hospitalized and develop care pathways and interventional strategies that can divert the medical condition before it ever happens. The health care industry has a responsibility to work together to study and share clinical data and expertise across the care continuum. Doing so will reduce cost, accelerate recovery and boost patient satisfaction.

Cerner’s performance improvement and clinical intelligence solutions use analytics, algorithms and models to empower users with data that enables them to impact care when it matters mostLearn more here

More like this: