AI early-warning system brings preventive care to critical paediatric patients

Rohit Rao, medical director of the cardiothoracic intensive care unit (CTICU) at Rady Children’s Hospital-San Diego, knows the value of quick reactions in a crisis. But better still, and the preferable model in any health-care setting, is to anticipate and resolve a crisis before a patient comes to harm.

“In critical care, we show up at the bedside, save a life, stabilize the patient, and then help with recovery,” the paediatric cardiologist and cardiac intensivist says. “That’s great, but I would rather be able to correct the problem so early that the child never reaches a crisis in the first place.”

Rao likens the intensivist’s current role to a firefighter, someone skilled at reacting to emergencies, but unable to prevent them. He’d rather be like a smoke alarm, he says, using an early-warning system to change a dangerous clinical trajectory at the earliest opportunity.

Rao and his colleagues are making progress towards this vision of preventive care with the CTICU’s addition of Sickbay, an FDA-cleared, artificial intelligence (AI)-enabled real-time remote patient monitoring platform. In 2022, thanks to a US$1 million grant from the Conrad Prebys Foundation, Rady became the first hospital in California to use the technology, which enables centralized haemodynamic monitoring of any CTICU bed.

AI in action

“The platform shows us vital parameters in a meaningful, actionable format, from 12-hour trends all the way down to one-minute trends, with all data stored in the cloud,” Rao says. “After an event, we can go back and look at the vitals to see what triggered the patient to decompensate. That’s important because many of these events do not trigger the monitor alarms.”

That was the case with a two-day-old infant in the CTICU who had been diagnosed with Ebstein’s anomaly, a rare congenital heart defect. A nurse practitioner looked at the AI platform and saw the heart rate was variable at 120 and suddenly increased to 160 without variability.

“That’s still within normal range for an infant, so no alarms were triggered,” Rao says. But when Rao arrived at the bedside, the newborn patient was in atrial flutter and had low cardiac output.

If the new technology had not detected the sudden change in heart rate trends before clinical symptoms developed, the patient would have become critically ill, Rao says. “I have no doubt we would have made the correct diagnosis even without the platform, because the child would have quickly shown symptoms,” he explains. “But even five minutes is a lot of time in our world, in terms of outcomes.”

Following cardioversion, the patient reverted to sinus rhythm. “What made the difference in this case is that a human being told me to look at a monitor that had high fidelity, data recovery and actionable trend lines on it,” Rao says.

Dana Mueller, a paediatric cardiac intensivist at Rady Children’s, is charged with helping to fully integrate the AI platform into the unit’s workflow. “The platform is up and running in all 30 beds in our CTICU,” she says. “We’re also rolling out the platform in all our cardiac catheterization labs and cardiac operating rooms, so we'll be able to fully monitor all of our patients, even when they leave the ICU.”

Research opportunities

The AI platform is also proving to be a valuable tool in Rady Children’s research programme. For example, Mueller and her colleagues are working on a proof-of-concept study using retrospective ventilator extubation readiness trial data.

Weaning a patient from a ventilator is a delicate balance, Mueller says, and one that may lend itself well to AI support. “Shorter ventilation times are associated with decreased infection risk and shorter hospitalizations,” she says. “But if we extubate too early, our patients run the risk of haemodynamic instability or cardiac compromise.”

To work out whether a patient can tolerate extubation from a mechanical ventilator, a respiratory therapist typically conducts a standardized test to determine whether a patient is ready to breathe on their own. An intensivist then considers the test results before deciding whether to extubate.

In the year it has been in use in Rady Children’s CTICU, the Sickbay platform has recorded hundreds of patient stays that included extubation readiness trials. To assess how well they did with hitting that extubation sweet spot, the investigators retrospectively reviewed these data while blinded to the outcomes of the human test reviews.

“It turns out, looking at our preliminary data, that some patients we deemed as not ready to extubate actually were, which is an important step in limiting complications,” Mueller says. “Incorporating Sickbay data into our decision-making might help us liberate kids faster from mechanical ventilation in future. We hope to incorporate Sickbay review into our workflow when assessing extubation readiness.”

To this end, the Rady Children’s intensivists are also hoping to design a prospective trial that will use the platform’s predictive analytics and algorithm development capabilities to examine AI’s role in helping to assess extubation readiness. To do so effectively, the trial’s scope will have to expand beyond the CTICU, and the walls of the hospital. “Unlike adult ICUs, paediatric ICUs have relatively small patient volumes,” Rao says. “That means we’re going to have to collaborate with other institutions for this trial to get a large enough dataset to generate meaningful results, so the impact will extend far beyond Rady Children’s.”

The CTICU may be the first ICU at Rady Children’s to use this kind of technology, but it almost certainly won’t be the last. “This combined approach to AI, predictive analytics and machine learning will eventually become the norm throughout neonatology and critical care,” Rao says. “I appreciate the opportunity to be an early adopter and show others how we can use AI to hone our clinical skills for the benefit of our patients and their families. My dream of becoming a smoke alarm is within reach.”