Multi-guideline decision support for vascular surgery
ClinicalGuidelines.io is being re-engineered from a single-guideline GraphRAG assistant into a context-aware vascular surgery agent that routes across 14 guidelines, asks for missing clinical details, and responds with citation-grade evidence.
ESVS guideline books aggregated into one unified vascular decision-support environment.
Identifies the correct guideline pathway instead of forcing clinicians to search manually.
Intercepts underspecified cases before recommendation output is allowed to proceed.
Recommendation-grade evidence is surfaced with preserved source alignment and citation logic.
The core shift is not only better retrieval. The platform now decides what kind of clinical question it has received, whether the case is answerable, and whether critical context is missing before advice is returned.
Pure guideline questions proceed directly to routed retrieval, producing concise answers grounded in the most relevant source sections.
When mandatory clinical variables are present, the case is validated, routed, and answered with traceable recommendation-level justification.
A dedicated Context Gate intercepts unsafe or underspecified scenarios and generates structured clarification prompts instead of premature recommendations.
PHI de-identification and multi-turn change detection keep the pipeline focused on clinical reasoning while reducing downstream privacy and state-tracking risks.
Before the move to a multi-guideline agentic system, the original guideline-locked application was benchmarked against expert-level answers and demonstrated non-inferior performance with explicit citation control.
This published validation work is the foundation for the new platform layer now being extended into multi-guideline, context-aware vascular decision support.
AI Performance Score
vs 77.8% for committee exemplarsAcross All 12 Metrics
p < 0.001 safety marginHallucinations
100% citation lock-inThe new system moves beyond passive retrieval. It interprets intent, checks whether the case is sufficiently specified, and only then decides how to retrieve and frame the evidence.
The incoming query is classified as a knowledge request, a complete case, or an incomplete case requiring clarification.
The request is directed across the correct guideline domain rather than forcing the user to know where the answer lives.
Mandatory clinical parameters are checked before advice is produced, preventing unsafe recommendations from underspecified cases.
Narrative chunks provide nuance while structured recommendation chunks carry the citation-grade decision signal.
Outputs return with source-linked justification, preserved traceability, and resistance to cross-guideline contamination.
The next stage is designed to move from internal technical validation to real-world case testing in conjunction with the Hellenic Society of Vascular and Endovascular Surgery.
Clinical validation pathway
The proposed prospective program is centered on multicenter assessment against real vascular surgery cases, with the explicit goal of testing whether an auditable, multi-guideline agent can support decision quality at the bedside without sacrificing transparency or clinician control.
The strategic value is simple: an agent that asks for missing details before it speaks is materially closer to the workflow of an experienced consultant than a static chatbot that only retrieves text.
Peer-reviewed work tracing the evolution from general-purpose LLM benchmarking to guideline-locked clinical decision support in vascular surgery.
Ann Vasc Surg. 2025 Feb;111:260-267. doi: 10.1016/j.avsg.2024.11.014. Epub 2024 Nov 24. PMID: 39586530.
Established the early benchmark for how foundation models handle vascular surgery decision-making before the guideline-locked architecture was introduced.
Ann Vasc Surg. Published 2026 Jul;128:473-483. doi: 10.1016/j.avsg.2026.03.020. PMID: 41895597.
Published in the Annals of Vascular Surgery, this study validated standardized expert-level interpretation of vascular clinical practice guidelines and forms the launchpad for the next agentic, multi-guideline phase.
ClinicalGuidelines.io has also been presented across major vascular meetings, moving the platform from publication into live international discussion.
ExCeL London, United Kingdom
Thursday, 23 April 2026
Istanbul, Türkiye
23-26 September 2025
Barcelona, Spain
15-17 October 2025
Showcased the ClinicalGuidelines.io concept within the endovascular meeting circuit, reinforcing interest in guideline-aware vascular AI beyond the manuscript setting.
Larissa Imperial, Larissa, Greece
19-21 March 2026
A multidisciplinary leadership group spanning vascular surgery, academic medicine, and translational clinical AI.
Vascular Surgeon & Developer
University Hospital of Ioannina, Greece
Vascular Fellow
Attikon University Hospital, Athens
Professor of Surgery
Yale University School of Medicine, USA
Professor of Vascular Surgery
National and Kapodistrian University of Athens
Professor of Vascular Surgery
University Hospital of Patras, Greece
Request access to the beta or register interest as a clinical validation collaborator. Approved clinicians will be redirected to the secure chat interface.