New algorithm better predictor of readmission following discharge, says study

por John R. Fischer, Senior Reporter | June 12, 2019
Artificial Intelligence
Baltimore B evaluated 382 variables
to predict the risk of readmission
for patients at UMMS
A new machine learning tool may offer a greater number of variables for better predicting a patient’s risk of readmission following discharge, according to a new study at the University of Maryland School of Medicine (UMSOM).

Developed at the University of Maryland Medical System (UMMS), Baltimore score (B score) is designed to help hospital better predict and manage risk of readmission, which is often due to patient harm and carries expenses.

"If hospitals can better target time and money in planning for discharge to home, then patients may not have to come back to the hospital, with the harm sometimes associated with hospitals, including risks for infection, falls, delirium and other adverse events," said lead researcher Dr. Daniel Morgan, associate professor of epidemiology and public health at UMSOM, in a statement.

The rate of unplanned readmissions within 30 days following discharge is a benchmark used to grade a hospital’s performance and quality of care. Readmissions occur among almost 20 percent of patients in the U.S. Though frequently preventable, many clinicians are not adequately equipped to identify which patients will be readmitted, with existing readmission risk assessment tools limited in the variables they assess for each patient.

Co-author William Bame, a senior data scientist at UMMS, provided the foundation of the algorithm by designing a neural network that can mine thousands of health data variables in real time and calculate a score to predict the chance of return following discharge.

The experimental B score algorithm was then used to evaluate more than 8,000 possible data variables from Sept. 2014 to Aug. 2016. It was then individualized for each of three UMMS hospitals in different settings, where it analyzed data on more than 14,000 patients to determine the likelihood of readmission. The final model drew from 382 variables, including demographics; lab test results; whether the patient required breathing assistance; body mass index; affiliation with a specific church; marital status; employment; medication usage and substance abuse.

Comparing their findings to actual readmissions at the three hospitals and predictions scored by programs such as the LACE index, the HOSPITAL score, and the Maxim/RightCare score, the researchers found B score was better able to identify those at risk, regardless of different settings. It was also most accurate among patients at highest risk, with those in the top 10 percent of B score risk at discharge holding a 37.5 percent chance of 30-day unplanned readmission, and those in the top five percent B score risk at discharge carrying a 43.1 percent chance.

The researchers and university credit the ability to use machine learning to predict this and other healthcare-related risks to the increased adoption of electronic health records.

"The widespread use of electronic health records has enhanced information flow from all clinicians involved in a patient's treatment," said Dean E. Albert Reece, university executive vice president for medical affairs and the John Z. and Akiko K. Bowers distinguished professor at UMSOM, in a statement. "This study underscores how patient data may also help solve the readmission puzzle and, ultimately, improve the quality of patient care."

The findings were published in the journal, JAMA Network Open.

You Must Be Logged In To Post A Comment