Artificial Intelligence Tools Could Reduce Hospital Readmissions

An artificial intelligence method could reduce 30-day hospital readmissions by an estimated 12 percent, leading to lower annual healthcare costs and improved patient care, according to a study published in PLOS One.

The US spends $3 trillion annually on healthcare, researchers stated, with hospital readmissions being an important and potentially preventable source of utilization. The healthcare industry has introduced several measures to ensure readmission rates stay low for particular medical conditions, but these standards could soon extend to general surgical procedures as well.

To help reduce the rate of hospital readmissions after general surgical procedures, researchers from Boston University (BU) and MIT developed an artificial intelligence approach to offer personalized treatment recommendations for patients during their hospital stay.

The team used data from 722,101 surgeries to build predictive models that analyzed how pre-operative factors (including underlying medical conditions) and post-operative factors (such as complications from surgery) could impact current hospital readmission rates.

Using both pre-operative and post-operative factors, the predictive models were able to forecast 30-day hospital readmissions by an average area under the curve (AUC) of 87 percent. Using only pre-operative variables, the predictive models achieved an AUC of 74 percent, outperforming earlier models.

The group then developed prescriptive models based on the pre-operative predictive models, ensuring more actionable interventions to reduce 30-day readmissions. Researchers found that pre-operative hematocrit levels, or the volume percentage of red blood cells in blood, significantly affected hospital readmission rates. The team developed a prescriptive model that suggested targeted pre-operative transfusions to increase hematocrit.

The results showed that using the prescriptive models, clinicians could administer blood transfusions that would reduce the likelihood of 30-day hospital readmissions by an average of 12 percent, potentially saving the healthcare system over $20 million annually.

The study’s approach is unique in that it goes beyond predicting readmissions to offer personalized treatment suggestions, further ensuring patients stay out of the hospital after undergoing surgery.

“The vast majority of AI in medicine develops models that learn how to predict some outcome,” said Ioannis Paschalidis, Professor of the College of Engineering (ECE, BME, SE, CDS) and Director of the Center for Information & Systems Engineering at BU.

“What is novel in this work is that we train a model that offers personalized recommendations/prescriptions aimed at improving patient outcomes. The approach has broad applicability and the potential to improve medical decision making.”

Researchers noted that the approach could extend to other areas of healthcare quality as well.

“A further potential use of our model is to decrease the length of hospital stay for patients with low-risk of readmission. We can choose a threshold for our models to have high specificity and thus able to accurately identify those that are at low risk of readmission,” researchers said.

The study was limited in that it measured only the short-term readmission outcome, and not the long-term oncologic outcome. Future research will need to focus on long-term impact in surgical patients.

Still, the team noted that their method further demonstrates the potential for artificial intelligence in healthcare.

“The use of AI in medicine is likely to increase, helping to improve patient outcomes and reduce costs,” said Paschalidis. “This paper is an example of how AI could be used to affect modifiable factors that improve outcomes through personalized medicine.”