Geisinger researchers are developing a way to predict readmission risk — here's how

Researchers from Danville, Pa.-based Geisinger and University Park, Pa.-based Penn State are developing a way to predict a patient's risk for requiring more medical care three days after their discharge from the hospital. 

The researchers created a model called REDD, which stands for readmission, emergency department or death, using clinical, administrative and socioeconomic data from patients admitted to Geisinger between 2012 and 2014.

"REDD is a machine learning model designed to predict which patients will be at a high risk of adverse events after they are discharged," said Deepak Agrawal, one of the Penn State researchers and a PhD student in industrial engineering and operations research during the time of the study. "Using the REDD model, we were able to leverage large amounts of data to identify these high-risk patients at the point of discharge, which helps physicians target interventions to effectively reduce adverse events."

Readmissions closer to when a patient is discharged are more likely to be related to factors that are present but not identified at the time of discharge, the researchers said.

Researcher Cheng-Bang Chen, a PhD candidate in industrial engineering and operations research at Penn State, said the more time that passes after discharge, the longer it may take a physician to evaluate often-lengthy patient charts to determine what led the patient to be readmitted.   

"Our model focuses on just three days after discharge, which gives physicians a better chance to improve their processes when treating the patient in the hospital setting," Mr. Chen said.

"During the six months the program was piloted, Geisinger tracked patients that were identified as high risk and implemented additional services to try to minimize the chances of a patient being readmitted, visiting the ED or dying," said Eric Reich, manager of healthcare re-engineering at Geisinger. 

The intervention physicians applied to patients with a high risk of hospital readmission included: scheduling a return appointment with the patient's primary care provider; providing education on patients' prescriptions and care plans after discharge; ensuring inpatient pharmacists reviewed the discharge medication list after reviewing provider recommendations; filling the patient's prescriptions before discharge; and completing a follow-up check on patients discharged to a skilled nursing facility the day after leaving the hospital.

The REDD model was tested and piloted at a Geisinger facility, but a lack of nursing resources at the hospital at the time meant there were no reliable estimates on how it reduced readmissions.

But Mr. Reich said he considers the project a success that can be introduced in hospitals across the U.S.

"If the REDD model was fully implemented and aligned with clinical workflows, it has the potential to dramatically reduce hospital readmissions," he said. "Through these data-driven predictive models, hospitals will be able to provide better quality care while efficiently allocating their scarce resources at the same time."

More articles on clinical leadership and infection control: 
Why involving family in ICU rounds may boost care quality
6 steps to improve nutrition support for ICU patients
How executives are rethinking quality initiatives to improve proper antibiotic use: 4 insights

© Copyright ASC COMMUNICATIONS 2018. Interested in LINKING to or REPRINTING this content? View our policies by clicking here.

 


IC Database-3

Top 40 Articles from the Past 6 Months