Every day, radiology reports routinely surface actionable findings unrelated to the reason for the scan, such as a lung nodule on a CT after a car accident or a mass noted in passing on an abdominal study. Once the patient leaves the emergency department and the ordering physician moves on, ownership of the follow-up is often unclear, and many of these findings are never acted on. It's a problem every health system shares.

Cleveland Clinic wasn't willing to accept that outcome for its patients. Committed to making sure every patient gets the care they need, it formed the Actionable Findings Clinic, a team of 10 full-time advanced practice providers (APPs) dedicated to reviewing each patient's chart and seeing every case through to follow-up. However, at Cleveland Clinic's scale, volume soon outpaced what the team could review by hand. The backlog climbed to roughly 60,000 cases—years of work at the team's pace.

Cleveland Clinic was already using Bunkerhill's long-horizon AI agent to identify undiagnosed heart disease on routine CTs. Encouraged by results of that work, the team began exploring other ways the same AI agent could improve patient outcomes, starting with actionable findings. The goal was to clear the backlog and keep pace with every new finding going forward, tracking each one through the weeks and months follow-up can take.

Taking on the chart review behind every open case

Cleveland Clinic's commitment to following up on every finding was specific to the organization, and so was the program it built. Bunkerhill's agent had to learn that way of working. The team shared the protocols and follow-up guidelines their APPs relied on. They walked Bunkerhill through how they worked a case from start to finish and explained the clinical context behind their judgment calls. The agent was then tailored to operate in the same systems as their staff and apply the same logic to every case.

For each finding, the agent works through the questions an APP would. It checks whether follow-up is indicated, what the appropriate interval is, and whether the recommended study was completed. That last question often takes judgment, since an MRI may satisfy a CT recommendation, or a CT may satisfy an X-ray recommendation.

When a follow-up is overdue, the agent looks for a clinical reason:

• A patient admitted to the ICU may reasonably miss a thyroid nodule ultrasound while more urgent care comes first.
• A patient hospitalized for a myocardial infarction may have a lung nodule follow-up deferred until they recover.
• A patient who has moved out of state is no longer under Cleveland Clinic's care.

Each of these once required a full chart review to establish.

The agent then brings the relevant history together and recommends a next step, such as contacting the patient, reaching out to the ordering provider, or closing the case. An APP reviews each recommendation and decides how to act.

In six months, the backlog went from roughly 60,000 cases to zero. The team now has near-real-time visibility into every patient with a recommended follow-up.

Returning APP time to patient care

APPs remain closely involved by design. They review the agent's recommendations, decide how to act on them, and continue to manage their own caseload. Because they no longer build each case from scratch, each APP now manages roughly twice the volume they did before.

That added capacity has gone back to patients. The Actionable Findings Clinic now follows up with 44% more patients, with APPs meeting patients directly to explain their findings and arrange the follow-up they need.

“We ask them to become the safety net for their patients. That's the part that they enjoy,” Dr. Chen shared. “In the past, when they were spending a lot of their time mining the electronic medical record, it was not a very fulfilling career. Over time, we're able to give them back the thing that they signed up for.”

Bunkerhill's impact on Cleveland Clinic's actionable findings program

Cleveland Clinic's program now keeps pace with every finding its radiologists document, and its APPs spend more of their time with patients.

To confirm the agent stays accurate, the team periodically samples cases for blinded review. Reviewers compare the final decisions made from the agent's work against those made from an APP's. The team plans to repeat this review every six months.

Here's what changed:

• No backlog: 60,000 cases cleared in six months
• Real-time visibility: Every recommended follow-up is tracked as it comes due
• Twice the capacity: Each APP manages roughly double the cases
• More patients followed up: 44% more patients through the Actionable Findings Clinic