Boston Children’s Hospital is shortening what could be a weekslong process of identifying similar patients by using AI to locate “clinical doppelgangers.”
The pediatric health system recently built a digital tool that allows cardiac ICU providers to surface past Boston Children’s patients with analogous conditions and desired outcomes to construct their care plans.
“I have a patient here. I’m trying to make a treatment decision. Can I find a cohort of patients who are similar and see how they responded to different sets of treatments, historically?” Boston Children’s clinical AI engineer Dinesh Rai, MD, told Becker’s. “A manual effort like this would take days to weeks when your time to decide is minutes to hours.”
This would typically involve the provider going back and forth with a data analyst to locate other patients from the EHR. Or relying on clinical studies based on a broad cohort of patients. Or “gut hunches” or “ephemeral, institutional type of knowledge,” as Dr. Rai put it.
The organization aims to speed up the process by a measure of five, reducing chart review by 80%. EHR integration is another long-term objective.
“Doppelgangers” are the latest example of health systems trying to speed up clinical decision support with AI and other technologies. Palo Alto, Calif.-based Stanford Health Care years ago launched the “green button” EHR project that it has since sped up further with ambient AI and a data startup. Some are also generating “digital twins,” or simulated replicas of patients.
But Dr. Rai said nothing existed commercially to this “fidelity” — current solutions search structured data in the EHR, but the majority of medical records are free text. Patient registries are narrowly defined to certain specialties, diseases and interventions.
So Boston Children’s developed its own solution, using technology from Amazon Web Services and OpenAI, converting its patient data in a way the large language model-powered AI agents could understand.
Boston Children’s started with the cardiac ICU because clinician champions there wanted to test the idea. And the cases are usually pretty complex, making this technology more translational to other specialties if it works there.
Dr. Rai foresees “many, many use cases” and said a tool like this is “one of the holy grails of research” for its potential to pinpoint patients for clinical studies.
While the goal is for the data to appear instantaneously, or in a matter of seconds, Boston Children’s isn’t there yet. It recently took a provider three days using the tool — as the clinician was doing other things — to uncover enough past patients to make an informed treatment decision on a current one.
Another aim is to get results in “one shot,” Dr. Rai said. Now, a clinician might make a query then have to keep refining things before getting an optimal answer or enough data.
At the Becker's 11th Annual IT + Revenue Cycle Conference: The Future of AI & Digital Health, taking place September 14–17 in Chicago, healthcare executives and digital leaders from across the country will come together to explore how AI, interoperability, cybersecurity, and revenue cycle innovation are transforming care delivery, strengthening financial performance, and driving the next era of digital health. Apply for complimentary registration now.
