Model Reliably Predicts Risk of Hospital Readmissions

Cleveland Clinic-developed forecasting tool could improve care, guide discharge planning

A “readmission risk score” instrument developed at Cleveland Clinic consistently predicts the risk of a patient readmission to the hospital, and has the potential to improve quality of care and lower health care costs, according to researchers.  

Advertising Policy

Cleveland Clinic is a non-profit academic medical center. Advertising on our site helps support our mission. We do not endorse non-Cleveland Clinic products or services Policy

The investigators, led by Anita Misra-Hebert, MD, MPH, Director of Cleveland Clinic’s Healthcare Delivery & Implementation Science Center and Associate Professor of Medicine at Cleveland Clinic Lerner College of Medicine, undertook a study to ascertain how well the model was performing and to determine the factors that might be affecting variation in its performance. They published their findings in the Journal of General Internal Medicine 

The readmission risk score model was created in 2017 from electronic medical record (EMR) data from admissions at all Cleveland Clinic hospitals. Based on peer-reviewed evidence and clinical input, it utilizes 18 variables available in the EMR, including patients’ previous healthcare utilization in the Cleveland Clinic system (particularly Emergency Department visits); type of admission; discharge disposition; comorbidities; medications; laboratory values; insurance status; and potential barriers to accessing care.

Although similar models have been developed at other facilities, the study’s authors note that readmission prediction may vary based on differing hospital and patient characteristics. Therefore, risk models need to be evaluated on a regular basis to assess their performance and validity.  

Advertising Policy

In this study, the authors measured the performance of the Cleveland Clinic readmission risk model in 11 Cleveland Clinic hospitals over a three-year period (2017-2020). The team looked at hospital admissions and readmission rates across medical and surgical, hospital site and diagnosis categories (including COVID-19), and by race and ethnicity. Risk scoring for hospital readmission ranged from 1 to 100, with a score of greater than 40 indicating the top 5% of patients at risk for readmission. 

Assessing the model’s performance

Over the three-year period, there were 600,872 discharges from 321,470 unique patients at the 11 hospitals. The 30-day readmission rates averaged 15.9% over the three-year period. (The COVID-19 patient readmission rate was 14% in 2020.) The authors found that, based on EMR data and statistical calculations, the Cleveland Clinic risk score instrument performed consistently well in evaluating the risk of readmission across all medical and surgical categories, including during the pandemic.  

The readmission rates varied by hospital and diagnosis. Patients who made up the largest number of readmissions had diseases of the circulatory, digestive and respiratory systems, as well as injury and poisoning. The categories in which the model underperformed in terms of accurate readmission prediction included COVID-19, infectious and parasitic diseases, benign neoplasms, and congenital anomalies. 

Advertising Policy

“Our overarching goal with this study was to assess the continued accuracy of readmission risk prediction in order to improve health care delivery—as this information can help focus the programs offered to patients at the time of discharge,” said Dr. Misra-Hebert. “We see that the clinical diagnosis— the one that might prompt a hospital admission — may ultimately affect the chances of readmission. It is important to increase physicians’ awareness of the risk of readmission at the time of discharge and consider which programs after discharge may benefit the patient. 

“At the same time, we must be diligent about evaluating these prediction models on a regular basis. Over time, as patient and hospital characteristics evolve, we have to be sure that our risk assessment models accurately reflect those changes, and make adjustments to the models as needed.”