{"resourceId":"plos-madlad-ems-dispatch-ml-rct-20260331","versions":[{"version":"external-b70d5e3b85107af1b0522be23f495e9c8910cdc99c6b882e7407251fa6116f8d","resource":{"id":"plos-madlad-ems-dispatch-ml-rct-20260331","title":"MADLAD RCT: ML risk scores raised correct low-acuity ambulance prioritization from 62.5% to 68.3% in Swedish resource-constrained situations","organization":"Uppsala University / Västmanland Hospital (Sweden EMS dispatch centers)","sector":"Emergency medical dispatch / EMS academic clinical trial","geography":"Uppsala and Västmanland regions, Sweden (combined population ~499,000)","publishedAt":"2026-03-31","publicationDate":"2026-03-31","eventDate":null,"sourceName":"PLOS Medicine","sourceLabel":"Peer-reviewed randomized controlled trial of ML-assisted EMS dispatch prioritization","sourceUrl":"https://journals.plos.org/plosmedicine/article?id=10.1371/journal.pmed.1004770","evidenceClass":"academic-research","outcomeClass":"mixed","topics":["operating-model","developers-agents","data-security","accessibility-workforce"],"finding":"In a parallel-group RCT of adult low-priority (2A/2B) ambulance patients during Resource Constrained Situations at two Swedish nurse-staffed EMD centers, 1,245 RCS were analyzed. Dispatchers aided by an openTriage gradient-boosting risk tool prioritized the patient with the highest subsequent NEWS2 in 68.3% of intervention RCS versus 62.5% under usual practice (OR 1.28, 95% CI 1.00–1.63, p=0.047). Post-hoc, prioritized patients had shorter dispatch delays (29 vs 32 minutes); full model compliance would have raised the OR to 1.56. Authors emphasize underpowering (ended early at n=1,245 vs planned 1,500), edge-of-significance p-values sensitive to imputation seeds, nurse-staffed CDSS context, and restriction to low-acuity adults—not high-acuity or non-conveyance cohorts.","sledRelevance":"Rare randomized evidence that ML decision support can modestly improve EMS dispatch prioritization under resource constraints—directly transferable as a cautionary benchmark for U.S. PSAPs and fire/EMS agencies evaluating AI triage prompts after the current non-emergency and nurse-line AI wave. Why now: yesterday’s CSPD Axon launch and Seattle Corti debates lack RCT outcome evidence; MADLAD supplies measured prioritization deltas with explicit limits.","evidence":"Opened full PLOS Medicine HTML. Primary outcome and OR/CI/p from abstract and results; sample flow, compliance 80.9%, secondary composite OR 1.31, and limitations from methods/discussion. Trial NCT04757194; published 2026-03-31.","architectureImplications":"Expect CDSS/CAD-integrated risk scoring with structured + free-text features, confidence display, and nurse–dispatcher conferral before unit assignment. Design for non-blinding, compliance tracking, model refresh against covariate drift, and clear failover when scores are unavailable.","governanceImplications":"Require preregistered outcomes (e.g., NEWS-aligned), ethics review for deferred consent, Data Monitoring Committee oversight, and published power/limitation statements before claiming clinical benefit. Do not generalize nurse-EMD results to U.S. telecommunicator staffing without local validation.","securityPrivacyImplications":"Models used personal identifiers and clinical narratives owned by regional health authorities; individual-level data not publicly releasable. U.S. ports need HIPAA/EMS record agreements and training-data controls.","caveats":"Underpowered; p at edge of significance; Sweden nurse-staffed CDSS; low-acuity only; international transferability limited; not a U.S. PSAP deployment result.","streamIds":["emergency-services"],"roles":{"sales":"Frame MADLAD as rare RCT proof that ML can nudge low-acuity ambulance priority under resource constraints—not as a turnkey U.S. 911 product. Discovery should map whether the buyer faces concurrent waiting patients, uses nurse or telecommunicator triage, and already measures NEWS-like acuity. Bounded engagement: protocol design and local retrospective validation of openTriage-class scoring before any live prompt pilot. Unsupported claims include guaranteed 911 answer-time gains, high-acuity readiness, or replication of the 68.3% figure without local powering.","engineering":"Fit is a gradient-boosting risk API consuming CDSS structured answers plus free-text notes, returning ranked patients and confidence colors inside the dispatch UI. Integrate with CAD assignment workflows and store scores for both arms even when hidden. Prerequisites include complete CDSS documentation, identity linkage for outcomes, and drift monitoring. Constraints: Swedish MBS CDSS feature space differs from U.S. protocols; non-blinding can bias randomization. Proof of value must reproduce prioritization accuracy and compliance—not vendor AUCs alone.","delivery":"Delivery work includes dispatcher protocol training, compliance dashboards, DMC safety review, and outcome pipelines from ePCR/ED vitals. Dependencies span CAD vendors, ethics/IRB analogs, and hospital data partners. Ownership stays with EMS medical direction and dispatch leadership. Acceptance criteria: preregistered primary metric, documented HITL override, and honest underpowering disclosure if sample falls short."},"roleDetails":{"sales":{"customerProblem":"When multiple low-acuity patients wait for scarce ambulances, agencies lack proven tools to pick who gets the next unit.","stakeholders":["EMS medical directors","Dispatch center managers","CAD/IT leads","Quality/research officers"],"discoveryQuestions":["How often do multi-patient waits occur?","Who staffs triage—nurses or telecommunicators?","What acuity proxy (NEWS/ePCR) can you measure?","Will overrides be logged without punitive scoring?"],"valueHypothesis":"ML risk ranking can modestly raise correct first-unit assignment among low-acuity waits if humans retain final authority and outcomes are measured.","potentialEngagement":"Retrospective openTriage-class validation and RCT/pilot protocol workshop—not a production 911 AI sale.","unsupportedClaims":["Guaranteed replication of 68.3% accuracy","Safe use on lights-and-sirens priority 1 without new trials","U.S. PSAP staffing equivalence to Swedish nurse EMD"]},"engineering":{"fit":"Decision-support risk ranker for concurrent low-priority ambulance waits inside CAD/CDSS—not autonomous dispatch.","architectureIntegration":"API scoring service with structured CDSS + bag-of-words text features; UI markers; audit store for scores even when hidden.","prerequisites":["Complete triage documentation","Outcome linkage to vitals/ePCR","Randomization/override logging","Model refresh process"],"constraints":["Feature mismatch vs Swedish MBS","Non-blinding of dispatchers","Low-acuity scope only"],"security":"Protect identifiable EMS narratives under regional health-data rules; define training-data retention and researcher access.","proofOfValue":["Primary prioritization accuracy vs NEWS","Compliance rate","Dispatch-delay delta","Drift correlation over time"]},"delivery":{"work":"Train dispatchers on protocol, stand up DMC, wire outcome extraction, and monitor compliance vs model confidence.","dependencies":["CAD vendor UI hooks","Hospital/ED vitals access","Ethics approval","openTriage or equivalent model ops"],"ownership":"EMS medical director owns clinical authority; dispatch manager owns daily protocol; IT owns scoring service.","skillsAdoption":"Build trust so overrides are reasoned, not ignored; share confidence calibration with staff.","governanceCheckpoints":["Protocol lock","Safety event review cadence","Interim power review","Publication of limitations"],"acceptanceCriteria":["Logged human final decision","Primary endpoint analysis plan followed","Documented sample-size shortfall if any"],"risks":["Overclaiming significance","Punitive compliance culture","Silent model drift"]}},"retrievedAt":"2026-09-24T09:16:59-05:00","enrichedAt":"2026-09-23T22:14:20-05:00","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Supports dispatcher decision speed under load but was tested with nurse triage staff uncommon in many U.S. PSAPs; workforce transfer needs local role mapping.","procurementImplications":"Open-source openTriage reference favors build/adapt over sole-source voice agents; still budget for outcome measurement and ethics oversight.","operatingModelImplications":"Adds an advisory ML conferral step among waiting low-acuity patients without replacing nurse/dispatcher judgment.","sourceVerification":{"openedUrl":"https://journals.plos.org/plosmedicine/article?id=10.1371/journal.pmed.1004770","referenceExcerpt":"68.3% of RCS were assessed correctly per the primary outcome versus 62.5% in the control group","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}