You’re playing catch in a casually maintained field. The ball lands in a patch of clover a few feet away. As you reach down to pick it up, you note a sprig with four leaves. You pluck it. Then, as you stand to announce your good fortune to your throwing partner, the four-leaf clover slips from your fingers and back into the mob of its three-leaf brethren. You start looking for it. Your friend asks: What’s the holdup? You pick up the ball, toss it back, and move on, leaving the lucky charm behind.
When doctors review X-rays or CT scans to look for a specific issue like a broken bone or pneumonia, they often don’t have time to look for unrelated medical “four-leaf clovers”: findings that might show up if diagnosticians hunted for surprises that are known as “Incidental findings.”
If experts happen to see unexpected results when they are looking for something else, they can save lives.
In the past, there haven’t been good ways to hunt down these incidental findings. Now, the Denver-based health-technology firm Eon is trying to solve the puzzle of incidental findings. With help and investment from UCHealth, they’ve built a system that integrates into UCHealth’s Epic electronic health record (EHR), the engine behind the My Health Connection patient portal and much more.
The system, Eon Patient Management, uses artificial intelligence (AI) to parse the standardized language of radiologists for incidental findings, categorize them based on risk, and then make sure they don’t go lost like so many sprigs of clover.
Catching lung cancer earlier through AI
The Eon-UCHealth team started with incidental findings of lung nodules. The impetus came from pulmonologist Dr. Diana Breyer, who is UCHealth’s chief medical officer for northern Colorado, and a team that called itself “the haystack group.” That’s because, she explained, “getting incidental findings into the appropriate track for many of our clinicians is like trying to find a needle in a haystack.”
Let’s say a patient arrives at a UCHealth Medical Center of the Rockies emergency room with heavy lungs. He gets a chest CT scan for possible blood clots. The radiologist finds no blood clots, but notes pneumonia and something else: lung nodules. She notes their prevalence, location, and size (they’re typically less than a centimeter in diameter), and then dictates these incidental findings into the patient’s electronic health record.
But then what? While some lung nodules do turn into cancer, more than 95% of them end up benign, so it’s less pressing than the pneumonia. Plus, the patient may be visiting from out of town and not on the Epic EHR. And even if he’s a UCHealth primary care patient, how does the radiologist alert the appropriate physician to check out the nodule-related findings in the EHR note? Emails and even faxes remain mainstays in what has continued to be an ad-hoc process.
And while fewer than 5% or so of lung nodules turn out to be cancerous, the sheer number of patients whose lung cancers might be caught early thanks to incidental findings could be large across UCHealth, which has more than 40 imaging centers doing nearly a million radiology exams a year. Doing incidental findings right could avoid difficult treatments for more advanced disease – and save lives.
“These are obviously huge impacts to patients,” Breyer said. “You miss a nodule that becomes a cancer and it’s not curable anymore.”
The question was how to automate the process. Breyer’s team reached out to the UCHealth CARE Innovation Center, which works with and invests in companies developing digital technologies that improve health care decision-making and delivery. They soon settled upon Eon – but with a caveat.
“They didn’t have a fully developed product in a way that would have enabled us to partner with them as a vendor,” said Dr. Jennifer Wiler, cofounder of the CARE Innovation Center’s and chief quality officer for UCHealth University of Colorado Hospital on the Anschutz Medical Campus.
“But we saw the potential to grow their technology, and that meant not just the AI component, but also the full end-to-end care coordination,” said Wiler, who is also a professor at the University of Colorado School of Medicine.
Care coordination, in this case, means making sure the four-leaf clovers of incidental findings not only avoid burial in medical notes or going lost in missed handoffs, but also that they find the right providers to ensure proper patient care.
Bigger picture, Eon fit perfectly into the CARE Innovation Center’s guiding philosophy of finding and fostering technologies to solve pressing problems from the medical front lines, adds Dr. Richard Zane, UCHealth’s chief innovation officer, CARE Innovation Center cofounder and chair of emergency medicine at the University of Colorado School of Medicine.
“It was consistent with our strategy of simplifying the delivery of health care by using technology to do what technology can do and allowing humans to do what they do, which is adjudication,” Zane said.
Quick work, striking results
Eon and UCHealth signed their agreement on May 22, 2022. Eon used anonymized UCHealth medical record data to further train its AI, and UCHealth worked with the company on the care-coordination front. Based on nodule size and other factors, the Eon system would automatically categorize incidental findings. Low-risk nodules would trigger an automated referral for patients to follow up with primary care providers – and, if the patients didn’t schedule their follow-ups within a given window, the system would automatically alert providers to remind patients directly.
Incidental findings of high-risk nodules would follow a different automated path. They would be assigned to a pulmonologist in the appropriate UCHealth region who could then fast-track the patient to a lung-nodule clinic for biopsy and diagnosis, Breyer says.
On Sept 22, 2022, just four months after the project’s launch, the Eon system went live (yes, that’s fast). In its first seven months, the lung-nodule incidental findings it mined and managed reduced the time of referral from 34 days to five days and yielded 263 cancer diagnoses.
If data from University of Colorado Hospital are any measure, many of those cancers would have been caught much later. In the first quarter of 2022, before the Eon system went live, experts at the hospital registered 112 new incidental pulmonary-nodule patients. In the first quarter of 2023, the system captured and managed 1,889 patients with incidental findings of lung nodules.
Further, patients with lung-nodule incidental findings are pursuing follow-up care at far greater rates. At Universtiy of Colorado Hospital, just half of those 112 patients in 2022 bothered to follow up. With Eon, about 80% of the 1,889 did.
“A significant portion are not going to end up with cancer, but you’re not going to know that unless you follow them,” Breyer said. “So having that kind of return on scheduled, recommended follow-up for their size and type of nodule is really the key piece of it.”
Eon’s lung-nodule software is now running throughout the UCHealth system, and what works at UCHealth is working at a growing number of institutions. The UCHealth CARE Innovation Center team continues its collaboration with Eon even as UCHealth considers where it will integrate the AI solution next. (Eon also has incidental finding products for abdominal aortic aneurisms, thyroid nodules, pancreatic cancer, breast cancer and liver cancer.)
For Breyer, the ability to capture, categorize, and boost the odds of incidental-finding follow-up amounts to a safety net for patients who might otherwise plunge into a belated cancer diagnosis. It took a great deal more than the luck of the Irish to realize.
“It’s very exciting to be able to provide this and really support our clinicians in caring for their patients,” she said.