{"resourceId":"txp-hearts2-himeji-llm-ems-records-20261001","versions":[{"version":"external-04decfdfe6d6ed8074a816397cd4f3673d9923fc5598d461a5a804483cdbf70c","resource":{"id":"txp-hearts2-himeji-llm-ems-records-20261001","title":"TXP Medical: Himeji and two Hyogo fire departments launch HEARTS 2.0 with LLM voice summarization of EMS records","organization":"Himeji City Fire Bureau, Nishi-Harima Fire Union, Ako City Fire Department / TXP Medical","sector":"Municipal fire and EMS / prehospital information systems","geography":"Japan (Hyogo Prefecture)","publishedAt":"2026-10-01","publicationDate":"2026-10-01","eventDate":"2026-10-01","sourceName":"iza! (PR TIMES press release)","sourceLabel":"Vendor press release on HEARTS 2.0 full operation across three fire departments","sourceUrl":"https://www.iza.ne.jp/pressrelease/prtimes/MIFN4OKMYNJ2TC5YZKL5JNKR3Q/","evidenceClass":"vendor-claim","outcomeClass":"emerging","topics":["knowledge-work","data-security","operating-model"],"finding":"TXP Medical's Oct 1, 2026 release says three Hyogo fire departments (about 790,000 residents, about 48,800 ambulance runs a year) began full operation of HEARTS 2.0 in October. Crews speak summaries and an LLM maps them into record fields, described as a national first. Operating rules bar LLM use for personal information and limit it to urgent cases where typing is hard. Monitor OCR reached 92.3% fully correct across six items; offline mode is included.","sledRelevance":"International EMS example of generative AI for field documentation and hospital-notification speed, with explicit usage limits; transferable as a design pattern, not as evidence of outcomes.","evidence":"Vendor press release; the only accuracy figure is OCR (92.3%); no LLM summary accuracy, time savings, or transport-time outcomes are reported. Expected effects are stated as goals.","architectureImplications":"Speech-to-text plus LLM field mapping into a structured EMS record, OCR from ID cards and monitors, ICD-10 dictionary, offline mode, and links to fire department systems and hospitals.","governanceImplications":"Written rules limiting where the LLM is used, especially excluding personal data, are a concrete control U.S. agencies can copy.","securityPrivacyImplications":"Personal information is kept out of LLM input by rule; how that is enforced technically is not described.","caveats":"Vendor claims; Japan's EMS and hospital-notification system differs from U.S. structures; outcomes not yet measured.","streamIds":["emergency-services"],"roles":{"sales":"Frame HEARTS 2.0 as a bounded documentation use case: three departments, about 48,800 runs a year, LLM summarization limited to urgent cases and kept away from personal data. That scoping is the selling point for U.S. EMS agencies wary of generative AI in patient records. Be clear the release offers no measured time savings or summary accuracy, so any U.S. engagement starts as a pilot with measurement.","engineering":"The architecture pairs speech-to-text with LLM field mapping, high-accuracy OCR (92.3% fully correct on six monitor items), an ICD-10 dictionary, and offline mode for poor coverage areas. The notable control is a usage rule that keeps personal data out of the LLM. U.S. engineers should ask how that is enforced in software, how summaries are verified by crews, and how it maps to ePCR and NEMSIS fields.","delivery":"If replicating, write the usage rules first: which call types allow LLM summarization, what data is excluded, and who signs off on each record. Pilot with a crew subset, measure documentation time and correction rates, and keep hospital notification on the existing path until summaries prove reliable. Joint deployment across departments needs shared governance from the start."},"roleDetails":{"sales":{"customerProblem":"EMS crews face rising run volumes and long documentation and hospital-notification times.","stakeholders":["Fire/EMS chiefs","EMS medical director","Receiving hospitals","Privacy officer"],"discoveryQuestions":["Where do crews lose time documenting?","Which data must never enter an LLM?","How are hospitals notified today?"],"valueHypothesis":"Narrowly scoped voice summarization may cut documentation burden on urgent calls; benefit is unmeasured.","potentialEngagement":"Usage-rule design and pilot measurement plan for LLM-assisted ePCR entry.","unsupportedClaims":["Measured time savings","LLM summary accuracy","Transport-time improvements"]},"engineering":{"fit":"Field documentation assist for EMS with OCR and voice summarization.","architectureIntegration":"Mobile app speech input to LLM field mapping; OCR from IDs and monitors; sync to fire systems and hospitals; offline mode.","prerequisites":["ePCR field mapping","Data-exclusion rules","Crew verification step"],"constraints":["Japan-specific system","Vendor-only claims","No U.S. NEMSIS mapping described"],"security":"Enforce personal-data exclusion from LLM input technically, not only by policy.","proofOfValue":["Documentation time per run","Crew correction rate on LLM summaries","Hospital notification time"]},"delivery":{"work":"Define LLM usage rules, pilot with selected crews, and measure corrections and time.","dependencies":["Vendor","Hospitals","Medical director","IT/privacy"],"ownership":"Fire/EMS operations with medical direction.","skillsAdoption":"Train crews on spoken summaries and checking AI-filled fields.","governanceCheckpoints":["Usage rules approval","Pilot review","Multi-department governance board"],"acceptanceCriteria":["Zero personal data in LLM inputs on audit","Crew sign-off on every AI-filled record","Correction rate tracked per field"],"risks":["Summarization errors in records","Policy-only data controls","Overstated vendor benefits"]}},"retrievedAt":"2026-10-04T22:00:49-05:00","enrichedAt":"2026-10-04T22:00:49-05:00","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Aims to cut documentation load on crews after run volumes rose to 160% of 2007 levels and on-scene time grew five minutes over ten years (agency background).","procurementImplications":"Joint procurement across three departments and a regional medical area shows a shared-platform model.","operatingModelImplications":"LLM used narrowly for urgent, hard-to-type cases rather than as default documentation.","sourceVerification":{"responseId":"curl-iza-hearts-2026-10-04T22:00:49-05:00","openedUrl":"https://www.iza.ne.jp/pressrelease/prtimes/MIFN4OKMYNJ2TC5YZKL5JNKR3Q/","referenceExcerpt":"個人情報の入力にはLLMを使用しない、テキスト入力必須箇所のみに使用する等の運用ルールを明確化したうえで","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}