JMIR Publications

ABSTRACT

Background:

Incidental detection of abdominal aortic aneurysms (AAAs) has increased with widespread imaging, while traditional surveillance workflows remain fragmented and clinician-dependent. We describe the implementation and system-wide performance of the System to Track Abnormalities of Importance Reliably (STAIR™), a centralized, artificial intelligence–assisted program designed to identify AAAs, assign guideline-based surveillance, and ensure longitudinal tracking within an integrated healthcare system.

Objective:

To evaluate the implementation and performance of a centralized, artificial intelligence–enabled surveillance program designed to identify, risk stratify and longitudinally track patients with abdominal aortic aneurysms across an integrated healthcare system.

Methods:

This descriptive cohort study included all patients enrolled in the STAIR™ AAA surveillance program following its implementation in December 2022. Case identification was performed using rule-based natural language processing of radiology reports, structured electronic health record queries, clinician referral, and automated lost-to-follow-up searches. All cases underwent centralized clinical review, with surveillance intervals assigned according to Society for Vascular Surgery guidelines. Patients were followed until a predefined administrative or clinical endpoint was reached. Outcomes were descriptive and included identification pathways, surveillance assignments, endpoint resolution, imaging utilization, and operative activity.

Results:

A total of 8,464 patients were enrolled. Identification occurred via problem list queries (59%), radiology natural language processing (29%), clinician referral (7%), and automated lost-to-follow-up searches (5%). Following centralized review, 3.7% required immediate imaging, 45.3% of patients were assigned biennial duplex surveillance, 9.5% were assigned five-year surveillance, and 20.6% were referred for vascular surgery evaluation. Prior AAA repair at enrollment was identified in 20.6% of patients. Among 4,718 patients who reached a definitive endpoint, all had documented final disposition, including transfer of care outside the health system (57.4%), no further follow-up required (13.2%), prior repair, death, patient refusal, or inability to establish contact. Duplex ultrasonography accounted for approximately 80% of surveillance imaging. Elective AAA repair volume averaged approximately 135 cases annually during the study period.

Conclusions:

In a large integrated healthcare system, a centralized, artificial intelligence–assisted surveillance infrastructure was operationally feasible and supported comprehensive identification, guideline-based surveillance assignment, and complete endpoint adjudication for patients with AAAs. These findings describe a scalable, workflow-focused approach to population-level AAA surveillance that is independent of care setting and emphasizes clinical oversight rather than autonomous decision-making. Clinical Trial: NA