Early detection of physiologic deterioration in order to reduce in-hospital mortality and prevent unplanned transfers to the intensive care unit is a patient safety goal. The use of non-automated early warning systems and rapid response teams to reduce in-hospital mortality often is limited because of inadequate and inconsistent alerting mechanisms.

The Parkland Center for Clinical Innovation developed, validated and implemented an automated, real-time artificial intelligence-powered early warning system model for predicting risk of unplanned intensive care unit transfers or cardiopulmonary arrests outside the ICU. The automated model outperforms non-automated models and unaided clinician observation, leading to improved care and patient safety.

Clinical transformation and improvement in care delivery is a team sport that requires empowering operational stakeholders with innovative analytical insights powered by AI, said Vibin Roy, MD, medical director at the Parkland Center for Clinical Innovation.

“It’s one thing to have a powerful engine, but you can’t move the car forward without a good set of tires,” Roy explained. “Similarly, it’s great to have a strong predictive model that can enhance and improve patient care, but it needs to be paired with a strong and capable team that can combine the machine knowledge with their own clinical intuition and experience to make the best possible decision for the patient.”

Avoiding black box predictions can assist the end user in making more informed decisions and also helps with clinician buy-in, he added.

“Our initial version of the early warning system, 1.0, provided alerts on patients who were at risk for clinical deterioration but did not provide contextual information as to the reason for the alert,” Roy said. “With our improved second version, EWS 2.0, we provided the top one to two reasons that contributed to the risk score, and this allowed the rapid assessment team members to be more informed and better anticipate potential issues with the patient.”

He said other healthcare organizations implementing such systems will gain an appreciation for the complexity and scale of in-hospital deterioration and why these events occur. On a busy hospital floor, there are numerous challenges for care teams to pick up on subtle or overt changes that take place prior to a patient requiring ICU level care or passing away, he explained.

“And AI has the ability to identify clinically deteriorating patients that could otherwise be missed,” he said. “Real-time monitoring and predictive models have the ability to provide a second set of eyes on the floor by monitoring EHR data in the background and then providing an alert when the constellation of information points to a potentially deleterious event occurring in the near future.”

Roy will be speaking in the HIMSS18 session, “AI-powered early warning system for clinical deterioration,” at 11:30 a.m. March 8 in the Venetian, Palazzo D.

Twitter: @SiwickiHealthIT
Email the writer: bill.siwicki@himssmedia.com