AHA Scientific Sessions 2026 Abstract · Nov 2, 2026
Operationalizing AI for Cardiovascular Care: An AI and Centralized Outreach-Enabled Workflow for Symptomatic Peripheral Artery Disease Detection
By Perez Guerrero E, Jain SS, Banda J et al.
Background: Peripheral artery disease (PAD) remains underdiagnosed in routine clinical practice despite its association with cardiovascular morbidity, mortality, and limb-related complications. Artificial intelligence (AI) models have shown promise for identifying patients at elevated risk for PAD, but little is known about how such tools perform when integrated into actual clinical workflows. Objectives: To evaluate the feasibility and implementation of an AI-augmented workflow for identification and evaluation of symptomatic, undiagnosed PAD in primary care, and to assess diagnostic yield, workflow completion and barriers, and downstream cardiovascular prevention actions. Methods: We developed a supervised machine-learning model using structured electronic health record data from Stanford Health Care to identify patients at high risk for undiagnosed PAD. The model was prospectively deployed across 4 primary care community clinics involving 59 clinicians. High-risk patients with upcoming visits were assigned a brief symptom questionnaire regarding claudication; symptomatic patients triggered an EHR-based non-interruptive advisory recommending referral for vascular evaluation (Figure 1). When no clinician action occurred, centralized outreach facilitated further engagement. Primary outcomes included PAD or borderline PAD identification by ankle-brachial index testing, completion of diagnostic evaluation, and changes in cardiovascular prevention management. Results: On retrospective test-set evaluation, the model achieved a sensitivity of 0.85, specificity of 0.97, and an area under the receiver operating characteristic curve of 0.80. During the deployment period, 1,204 high-risk patients completed the symptom questionnaire, of whom 166 (13.8%) were symptomatic. Clinician action in response to the EHR advisory occurred in less than 10% of cases, with most completed evaluations occurring through centralized outreach. 62 patients completed the PAD evaluation pathway and underwent vascular assessment (Table 1, Figure 2). Among these, 5 patients (8.1% [95% CI 3.5-17.5%]) were diagnosed with definite PAD and 13 (21.0% [95% CI 12.7-32.6%]) had borderline PAD. In addition, 30 patients (48.4% [95% CI 36.4-60.6%]) had documented changes in cardiovascular prevention management. 39 of 62 patients (62.9% [95% CI 50.5-73.8%]) had at least one actionable finding (definite PAD, borderline PAD requiring follow-up, a change in preventive management, or a surveillance recommendation). Conclusions: An AI-augmented workflow for PAD detection in primary care was feasible to implement and identified diagnostic and preventive opportunities among patients completing evaluation. Clinical impact depended not only on model performance, but on workflow design, change management, and healthcare delivery innovation via centralized outreach infrastructure.