Electronic Health Reporter

By Angela Adams, RN, BSN, CEO, Inflo Health.

There is a version of a radiology report I have seen hundreds of times. The imaging is done well. The finding is documented. And the report ends with a sentence along the lines of: “Please correlate clinically. Further imaging may be warranted.”

That sentence sounds reasonable. It is also, in many cases, clinically useless. It does not say what should happen next. It does not indicate urgency. It does not tell the ordering provider whether this is a finding that needs action in two weeks or two years. This is called “hedging.” It’s defensive language designed to limit liability without committing to a specific recommendation. And it is one small symptom of a much larger structural problem in imaging.

Radiology-specific AI has made significant gains in detection. Algorithms are getting better at identifying lung nodules, incidental lesions, and abnormalities that might have been missed a decade ago. That progress is real, and it matters. But most of the industry conversation around imaging AI has focused on the front end of the workflow—what the model can find, what it misses, and the usefulness of AI enabled detection—while largely ignoring the back end: what happens after the finding hits the report.

That back end is where care actually breaks down.

The Downstream Problem Nobody Designed For

Every flagged finding is the beginning of a workflow. A follow-up study needs to be ordered. A patient needs to be contacted. A referral may need to be placed. A timeline needs to be tracked. When the finding is serious, those steps carry genuine clinical urgency. When they do not happen, patients get lost.

The data on this is sobering. Research published in the Journal of the American College of Radiology found that overall adherence to recommendations for additional imaging of incidental findings was just 39.1%. Other studies put the figure closer to 50 percent. However you measure it, the gap between what is found and what gets followed up on is enormous, and it widens as imaging volume grows.

And volume is growing. The Neiman Health Policy Institute projects that imaging utilization could increase by as much as 26.9% by 2055, while radiologist supply is expected to grow at a roughly comparable rate, meaning the current shortage is unlikely to improve without deliberate intervention. Radiologist attrition has accelerated since the pandemic, with departure rates up 50% from pre-COVID levels. Under that kind of pressure, report language gets less specific, recommendations get more vague and the downstream infrastructure (which was never adequate to begin with) absorbs more volume than it can handle.

This is the paradox at the center of imaging AI right now. Better detection tools surface more findings. More findings generate more downstream work. And the hard task of translating a finding into actual care relies on a workforce and systems already running at capacity.

More Dashboards Will Not Solve This

Health systems have tried to address the follow-up gap with worklists, tracking spreadsheets, and manual processes. I have watched care navigators spend hours every morning reconciling data from radiology systems against the EHR to figure out which patients still need to be contacted. In many organizations, that manual reconciliation is not a temporary workaround. It is the process. It is also the reason people fall through the cracks.

The limitation of most existing approaches is that they create visibility without creating accountability. A dashboard can tell you that a lung nodule was flagged. It cannot often tell you whether the follow-up was ordered, whether the patient was contacted, whether the appointment was scheduled, or whether the result came back. Those are different operational problems, and each one requires a different handoff.

What is needed is infrastructure that connects detection to completed care. Not just a view of what was found, but an operational layer that routes findings based on actual clinical risk, manages outreach, tracks completion, and escalates when something stalls.

What Completed Care Actually Looks Like

Radiology is often the starting point for a patient’s journey through the health system. From the moment an image is captured to the next step in care, each pathway is different. Patients have different needs, priorities, and resources. Technology workflows have different gaps that require different levels of support to close. Providers serve different populations with different barriers to care. There is no single worklist, workflow, or outreach strategy that can reliably solve every situation, every time.

In an environment defined by high variability and high stakes, high reliability becomes essential. It requires layered processes that apply the right tools to the right problem, with the goal of ensuring that no patient falls through the cracks. This means shifting our focus from task completion to patient outcome. Instead of asking, “Did the provider receive a notification?” we ask, “Did the patient receive the right next step in care?”

That distinction matters. True follow-up requires accounting for the complexity of the patient journey, including the reality that the right next step may change as new information, barriers, or circumstances emerge. High reliability is not measured by whether a task was checked off a list. It is measured by whether the system produced the intended action and result for the patient.

The Real Question for Imaging AI

The radiology AI market has spent the last several years racing to build better detection. That was the right starting point. But the industry is now at a point where the bottleneck is no longer whether a finding can be identified. The bottleneck is whether a finding, once identified, reliably reaches the right clinician, generates the right action, and results in completed care.

Health systems that invested heavily in AI detection tools are beginning to discover that the return on those investments depends almost entirely on what happens after the algorithm runs. A finding that surfaces in a report but never reaches the patient is not a detection success. It is a care failure that started with accurate imaging.

The next chapter of imaging AI needs to be about care completion: building the infrastructure between the radiology report and the EHR, between the finding and the follow-through, between what was identified and what was actually done about it. That is where patient safety lives. And right now, for too many health systems, it is also where patient safety breaks down.