Incidental finding follow-up is the process of making sure a clinically significant finding discovered unexpectedly on an imaging study — like a lung nodule seen on a cardiac CT — is documented, communicated to the right people, tracked, and completed. It fails far more often than most people realize. In the U.S. alone, more than 370 million radiology studies are performed each year, roughly 10% include a recommendation for additional imaging, and nearly half of those go unfulfilled — meaning as many as 18.5 million patients a year may miss a crucial follow-up. Because missed incidental findings are a leading source of diagnostic error, malpractice claims, and preventable patient harm, closing this gap is one of the highest-impact quality problems in radiology today. AI-driven, closed-loop workflows are now doing exactly that — detecting findings reliably, assigning ownership, and confirming each one is resolved.
What is an incidental finding? A finding discovered unexpectedly, outside the reason the scan was ordered — for example, a chest CT ordered for chest pain that also reveals a pulmonary nodule, or an abdominal scan for appendicitis that uncovers a breast lesion. These findings are clinically important but sit at the margins of the workflow, which is exactly why they get lost.
Why are incidental finding follow-ups so often missed?
Short answer: because follow-up depends on manual handoffs with no clear owner and no system tracking completion. A recommendation can be worded inconsistently, buried in a long report, passed between radiologist, referring provider, and patient through fragmented channels, and never reliably confirmed as done. Any one break in that chain loses the patient.
The problem is systemic, not a matter of individual performance. The most common failure points are:
- Radiology reports vary in how recommendations are worded, so simple keyword searches miss them.
- Findings are buried in long narrative reports without standardized formatting.
- Responsibility for follow-up is unclear across teams.
- Patients never receive clear, plain-language instructions.
- Health systems lack a reliable mechanism to track whether follow-up was completed.
- The follow-up provider is outside the health system’s EHR, so the referral becomes a fax or phone call no one can confirm was completed.
- The appointment happens, but nothing reconciles the result back to the original report; the finding stays “open” indefinitely.
The published baselines are stark: before any intervention, completion rates for recommended follow-up imaging can be as low as 43%, and for incidental findings identified in emergency departments, follow-up occurs in only about 17% of cases (Wandtke & Gallagher, 2017).
And the stakes are high: in malpractice analyses, failure to follow up on radiology findings is a recurring theme — the medical liability insurer Coverys has reported that diagnostic errors are the most expensive malpractice category in physician offices, averaging roughly $661,000 per claim, with many tracing back to findings that were documented but never pursued.
Inflo Health’s own analysis of more than six million radiology records shows how uneven the problem is. About 15% of studies carried a follow-up recommendation. Structured breast and lung programs — where dedicated navigators exist — closed 64% and 60% of their follow-ups, respectively. But 88% of imaging volume falls outside those formal programs, into an “other” category typically handed to primary care without a structured handoff, where closure drops to 51%.
How does AI detect incidental findings in radiology reports?
Short answer: radiology-tuned language models read every report and flag actionable findings regardless of how they’re worded — something keyword rules and manual review can’t do reliably. This is the first and most important link in the chain, because a finding that isn’t reliably captured can’t be followed up.
Modern radiology-tuned large language models (LLMs) and image-recognition algorithms analyze reports and imaging data with high accuracy, catching findings whether a radiologist wrote “3 mm nodule,” “small pulmonary opacity,” or any other phrasing. Once a finding is captured, AI classifies its urgency — routing a potentially life-threatening abnormality for immediate review while queuing lower-acuity findings appropriately — and generates a structured summary (location, size, clinical significance, recommended interval) that lands in the patient’s record rather than in a free-text paragraph nobody re-reads.
It is also worth noting that detection has gotten very good. FDA-cleared models routinely report 95%+ sensitivity on narrow indications, and peer-reviewed studies show AI reducing false positives in lung cancer screening on chest X-ray (Megat Ramli et al., 2025) and in prospective multicenter breast screening (Chang et al., 2025). There’s a catch, though: better detection increases the volume of downstream follow-up recommendations, and that surge can quickly outpace what staff and manual systems can manage. Some health systems have even hesitated to turn detection capabilities on, fearing the liability of findings they can’t reliably act on. Detection without follow-up infrastructure doesn’t reduce risk; it documents it.
How does AI make sure the follow-up actually gets completed?
Short answer: it closes the loop — every finding becomes a tracked task with an owner and a due date, and the system escalates anything that isn’t completed. Detection tells you a finding exists. Follow-through ensures the patient actually gets care. If your measure of success is “the AI flagged the nodule,” you’re measuring the first ten percent of the workflow and calling it done. Follow-through is the other ninety percent: the ordering provider seeing and acting on the recommendation, the patient being contacted, the appointment being scheduled and kept, and the result being reconciled back to the original finding. Every one of those steps is a handoff, and every handoff is a place where patients get lost.
In practice, that looks like:
- Smart worklists. Each finding is entered into a worklist sorted by urgency, risk, and due date, so coordinators can see exactly who needs action and when.
- Automated reminders. A six-month pulmonary-nodule recheck is tracked and flagged when due, with reminders to both clinicians and patients.
- Closed-loop tracking. The platform monitors whether the follow-up is ordered, scheduled, and documented — and if the loop breaks, it alerts the care team before the patient is lost.
- Reconciliation. When the follow-up visit happens, the result is matched back to the index finding and the loop is formally closed — so nothing sits “open” and unmonitored months later.
This is the core difference between an AI-enabled system and a traditional workflow: it moves follow-up from a fragmented, hope-for-the-best process into a fully accountable one.
Does AI-driven follow-up actually improve outcomes?
Short answer: yes — organizations using it have moved follow-up completion from roughly 50% to 80%+ without adding staff. The evidence includes both published research and Inflo’s own client results.
- Working with East Alabama Medical Center through the ACR Learning Network’s ImPower program, Inflo Health helped boost recommendation follow-up completion by 74%, improve guideline adherence by 20%, and reduce manual tracking time by 95%.
- Ochsner Health unified more than 40 hospitals on a single follow-up program and replaced manual spreadsheets with automated, closed-loop tracking.
- Across Inflo’s client portfolio, teams maintain 80%+ follow-up completion rates and 100% identification of explicit findings, aligned with ACR-validated methods.
- Independent research points the same direction: RSNA reviews report AI reduces time-to-diagnosis and improves detection sensitivity versus manual review, and Columbia University researchers found AI-driven tracking reduced patients lost to follow-up.
- Completed follow-ups also recover revenue that missed ones forfeit: Inflo clients have consistently found that the imaging revenue from completed follow-up exams far exceeds the cost of the tracking system itself, and can drive a 27X return on investment
Professional guidelines are consistent on one point: AI should complement physician oversight, not replace it.
What does reliable incidental finding follow-up look like in practice?
This isn’t hypothetical. Angela Adams, RN, CEO of Inflo Health, has shared the story of Jill, who went to the ER with appendicitis. Her CT also revealed a breast lesion — which was never communicated to her. A year later, that oversight became a fatal, inoperable cancer diagnosis.
In a closed-loop workflow, the system flags the lesion as actionable, creates a follow-up task, alerts the provider, and enters the patient into a reminder sequence. Follow-up imaging is scheduled within weeks, the lesion is biopsied early, and treatment starts before the disease progresses. That is the practical difference between a missed finding and a caught one.
How do you build a reliable incidental finding follow-up program?
Short answer: pair AI automation with clear ownership, physician oversight, and completion tracking — technology alone isn’t a program. A safe, effective implementation rests on five practices:
- Clinical validation — tools tested against large, diverse datasets and benchmarked against gold-standard methods.
- Physician oversight — every AI-identified finding reviewed and confirmed by a qualified clinician.
- Workflow integration — systems that fit inside the EHR and existing radiology processes rather than adding a separate inbox.
- Continuous monitoring — outcomes tracked over time to sustain accuracy and refine performance.
- Patient-centered communication — findings and next steps explained in plain language so patients act on them.
If you can’t answer all four from a single system of record, you’re measuring detection and hoping — not measuring completion.
As Inflo Health CEO Angela Adams, RN, put it: “We didn’t just focus on the math and the AI problem. We focused on taking that information that we identified and making sure that it worked within the clinical workflow.”
Wondering where to begin? Start with a simple audit. Pull a recent month of flagged incidental findings and ask four questions about each:
- Was the ordering provider’s receipt of the recommendation confirmed?
- Is there a scheduled follow-up event tied to the finding?
- Did that follow-up event actually happen?
- Is the result reconciled against the original report?
Want to benchmark where your program stands today? Take Inflo Health’s 6-question Follow-Up Maturity Assessment. Health systems ready to put a high-reliability follow-up solution in place can see how Inflo Health builds it into the radiology workflow on the Clinical Programs and Solution for Hospitals pages, or book a working session.
Referneces
Wandtke, B. & Gallagher, S. (2017). Reducing delay in diagnosis: Multistage recommendation tracking. American Journal of Roentgenology, 209(5), 984–991. https://doi.org/10.2214/AJR.17.18332
Chang, Y.W., Ryu, J.K., An, J.K., et al. (2025). Artificial intelligence for breast cancer screening in mammography (AI-STREAM). Nature Communications, 16, 2248. https://doi.org/10.1038/s41467-025-57469-3
Megat Ramli, P.N., et al. (2025). The Role of Artificial Intelligence in Lung Cancer Screening in Detecting Lung Nodules on Chest X-Rays: A Systematic Review. Diagnostics, 15(3), 246. https://doi.org/10.3390/diagnostics15030246
