{"resourceId":"hou-ems-workflow-ai-interviews-2026","versions":[{"version":"external-19b1c8be472418e0ceaf2e7c840922462c46e60e07e887514dae5994a638fabf","resource":{"id":"hou-ems-workflow-ai-interviews-2026","title":"U.S. EMS interviews identify workflow and autonomy constraints for AI support","organization":"Emily Hou and colleagues; Wellesley College, University of New Hampshire and University of Haifa","sector":"Emergency services","geography":"United States","publishedAt":"June 15, 2026 arXiv version; interviews June–July 2025","publicationDate":"2026-06-15","eventDate":null,"sourceName":"From 911 to Hospital: Challenges and Opportunities for AI Integration in Emergency Medical Services","sourceLabel":"CHIWORK 2026 paper, author manuscript on arXiv","sourceUrl":"https://arxiv.org/html/2606.16984v1","evidenceClass":"academic-research","outcomeClass":"cautionary","topics":["knowledge-work","infrastructure","data-security","accessibility-workforce","operating-model","governance-procurement"],"finding":"Interviews identify coordination, attention and autonomy concerns surrounding prospective EMS AI.","sledRelevance":"Interpretation: Directly relevant to U.S. municipal, private and volunteer EMS workflow discovery; no deployment efficacy is established.","evidence":"Researchers interviewed 25 U.S. EMS clinicians in June–July 2025 and used thematic analysis. Tables map technology interactions and hypothetical applications across response stages. No controlled baseline or measured AI response-time benefit.","architectureImplications":"Interpretation: Prototype one handoff adapter with source timestamps and an explicit unavailable state. Compare local, hosted and hybrid options through field connectivity tests.","governanceImplications":"Interpretation: Let medical leadership approve intended use and pause criteria; require visible uncertainty and usable override.","securityPrivacyImplications":"Interpretation: Restrict recordings and draft reports by role; review consent, retention and vendor reuse before any pilot.","caveats":"Interview and hypothetical-system evidence, not direct observation. Some ideation assumed perfectly accurate AI. Section 5.5.4 conflicts with 5.4.1 about assessment/treatment demand; avoid that stage-specific conclusion.","streamIds":["emergency-services"],"roles":{"sales":"Interpretation — Engage the EMS chief, field clinicians, clinical quality lead and records owner about information lost at handoffs. Ask which details require repeated entry and whether another prompt would interrupt care. A bounded engagement could map one documentation task and test a draft-only assistant in training. The value hypothesis is better record completeness with less correction work, subject to comparison against current practice. The interview evidence helps identify needs; it cannot establish demand for a particular vendor, savings, safe diagnosis or workforce reductions. Confirm that the agency has time and reviewers for evaluation before proposing a live pilot.","engineering":"Interpretation — Fit is a reversible documentation or handoff prototype. Map CAD, ePCR and approved source records; require attributable inputs, timestamps and human confirmation before filing. Prerequisites include representative test cases, reviewer capacity and a medical owner. Test contradictory notes, missing fields, noisy speech and loss of connectivity, keeping established communication available. Protect patient information through limited access and controlled retention. Proposed proof of value should compare omissions, invented facts, correction time and task interruption against ordinary documentation. Evaluate each current model configuration locally; hypothetical clinician preferences do not validate a product or justify an autonomous clinical agent.","delivery":"Interpretation — Start in training with field representatives and a named EMS quality owner. Inventory interfaces, approve test data and teach staff how to reject drafts. Records and IT teams must resolve retention and access dependencies before adoption. Governance checkpoints should cover baseline scoring, simulation review and each material model change. Proposed acceptance requires an agreed improvement in completeness without added critical errors, measured correction burden and successful fallback exercises. These criteria are proposed, not observed. Preserve staff feedback and non-use reasons, and stop when the tool distracts from care. Expansion requires prospective evidence beyond the interview study."},"retrievedAt":"2026-09-09T03:00:46Z","enrichedAt":"2026-09-09T03:04:34Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Interpretation: Test with gloved hands, noisy scenes and disabled staff; preserve practiced manual routines.","procurementImplications":"Interpretation: Contract for representative usability testing, export and change notification; do not purchase claimed clinical efficacy from a design study.","operatingModelImplications":"Interpretation: EMS supervisors own workflow, medical directors own clinical boundaries and IT owns service restoration. Developer agents have limited direct relevance; integration code still needs ordinary review.","updateExplanation":"New to all 119 archived resources inspected at offsets 0 and 100. Fills U.S. frontline workflow evidence beyond the existing simulation and Danish trial; no claim of a new September announcement.","sourceVerification":{"openedUrl":"https://arxiv.org/html/2606.16984v1","referenceExcerpt":"Finally, we examined perceptions of hypothetical AI systems rather than evaluating deployed technologies.","promptVersion":"sled-research-v3.1","model":null,"basis":"agent-reported inspection"}}}]}