BRENNEN

Industries / Health / Healthcare

Records and paperwork

Writing down, coding and sharing what is known about each patient, so that the next clinician, the patient and public health statistics can use it.

7

Milestones achieved, from 1893 to 2025.

2

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1893

    A shared list of causes

    Jacques Bertillon, head of statistics for the City of Paris, presented a classification of causes of death to the International Statistical Institute in Chicago in 1893, and the Institute adopted it. Several countries and many cities took it up, and in 1898 the American Public Health Association recommended it for Canada, Mexico and the United States and proposed revising it every ten years. Its successor, the International Classification of Diseases (ICD), passed to the World Health Organization in 1948. The current version, ICD-11, came into effect on 1 January 2022, and ICD codes now underpin health records, cause-of-death certificates and health statistics.

    who.intwho.int

    Credit

    Jacques Bertillon and the International Statistical Institute; later the World Health Organization

  2. 1907

    One record per patient

    On 1 July 1907 the Mayo Clinic gave its first patient number one and began keeping each patient's whole medical history in a single file, a paper dossier, in one place. For many years the files moved around the clinic by pneumatic tube and mechanical carriers. Mayo has registered more than 8 million patients under the system.

    history.mayoclinic.org

    Credit

    Henry Plummer and Mabel Root

  3. 1968

    Records organized by problem

    Lawrence Weed set out in the New England Journal of Medicine a way to organize a patient's record around a list of the patient's specific problems, with the notes and plans for each problem kept together. The problem-oriented record was widely adopted in the United States in the 1970s, and its parts, from the problem list to the SOAP note (subjective, objective, assessment, plan), now appear in millions of records.

    europepmc.orgeuropepmc.orgeuropepmc.org

    Credit

    Lawrence Weed

  4. 2009

    Records go electronic

    The US HITECH Act of 2009 set up federal programs to promote electronic health records (EHRs) and secure exchange of health information. By the count of the Office of the National Coordinator for Health IT (ONC), 9% of US hospitals had a basic EHR in 2008; by 2024 more than 99% of non-federal acute care hospitals and 91% of office-based doctors had a certified one.

    LessonMoving records onto computers moved clerical work onto doctors: a 2016 time-and-motion study of 57 US doctors found nearly 2 hours of EHR and desk work for every hour spent face to face with patients.

    healthit.govhealthit.goveuropepmc.org

    Credit

    US Congress; ONC

  5. 2011

    A national program stalls

    England's National Programme for IT in the NHS aimed to give every patient an electronic care record by 2010. In May 2011 the National Audit Office reported that the core aim was no longer expected to be met, that only 4 of 97 planned systems had reached acute hospital trusts in the North, Midlands and East in seven years, and that the £2.7 billion spent on care records systems "does not represent value for money".

    LessonA national records system can absorb billions and still fail to reach most hospitals.

    nao.org.uk

    Credit

    National Audit Office, which reviewed it

  6. 2018

    A common format for records

    HL7 International, a standards body for health data, published Release 4 of FHIR (Fast Healthcare Interoperability Resources) on 27 December 2018. It was the first FHIR release with normative content, meaning parts that are held stable so software built on them keeps working. FHIR splits a record into small standard pieces that software can request over the web. A 2020 US rule under the 21st Century Cures Act calls on the industry to adopt standardized interfaces so that patients can get their records through smartphone apps.

    hl7.orghealthit.govhealthit.gov

    Credit

    HL7 International; ONC

  7. 2025

    AI scribes in a trial

    An AI scribe listens to a visit and drafts the clinician's note. In a randomized trial of 238 outpatient doctors in 14 specialties at UCLA, those given the Nabla scribe spent 9.5% less time writing notes than those without one, while those given Microsoft's DAX Copilot showed no significant change. Users of either scribe reported less burnout and task load, and clinically significant errors in the drafts were reported "occasionally".

    LessonDrafts still need checking: in a 2026 randomized trial in one surgeon's practice, letters drafted by an AI scribe held 19 medical errors against 9 in conventionally written letters.

    europepmc.orgeuropepmc.org

    Credit

    P. J. Lukac and colleagues, NEJM AI, November 2025

Watch for

  1. AI charting inside the record

    Epic Systems, whose software holds 42% of the US acute hospital records market by STAT's figures, released AI Charting in February 2026 as part of Art, its AI for clinicians. It listens during a visit, drafts the note and queues orders and diagnoses for the clinician to review and sign; a version for nurses followed. Epic reports that clinicians at John Muir Health save 34 minutes a day on notes, and that UPMC cut after-hours charting ("pajama time") by nearly two hours a day.

    Achieved whenan outside study measures time saved and note errors for Epic's tool, as the 2025 UCLA trial did for Nabla and DAX Copilot.

    epic.comstatnews.com

    Claimed by

    Epic Systems and its customers, February 2026

  2. Paper forms turned digital

    Health Samurai, which makes the Aidbox health-data platform, sells a feature that reads a PDF of a medical form and writes it as a FHIR Questionnaire, the standard format for digital health forms. Its engineers wrote that early versions "often lacked full fidelity", including missing skip rules, and that the current pipeline "can now reliably handle complex, multi-page PDF forms". The company has published no accuracy figures or test set.

    Achieved whenan outside test on a published set of forms reports how many questions, answer lists and skip rules the tool gets right.

    health-samurai.iohealth-samurai.io

    Claimed by

    Health Samurai, May 2025, with current product documentation

Open

  1. 01

    A named, repeatable scribe test

    A public set of at least 50 staged clinic visits, played by actors from written scripts across common specialties and published as audio. Run every AI scribe that can be bought on the same recordings, have clinicians rate the notes blind, name the products, and repeat the test every six months.

    The brief
    Credit · unclaimed
  2. 02

    A test for form converters

    A public answer key of at least 100 blank clinical forms, such as new-patient intake and symptom questionnaires, each converted by hand into a FHIR Questionnaire with the right question types, answer lists and skip rules. Then score every tool that converts forms, commercial ones such as Health Samurai's and an open baseline built on a general AI model, question by question against the key.

    The brief
    Credit · unclaimed

Open 01 · the brief

A named, repeatable scribe test

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public set of at least 50 staged clinic visits, played by actors from written scripts across common specialties and published as audio. Run every AI scribe that can be bought on the same recordings, have clinicians rate the notes blind, name the products, and repeat the test every six months.

Why it is still open

Head-to-head tests so far are small and anonymous. An April 2026 study in the Australian Journal of General Practice compared four commercial scribes on four staged consultations and did not name them "because of commercial sensitivities"; its authors list the small sample and the point-in-time test among its limits. Randomized trials have compared at most two products at a time.

www1.racgp.org.aueuropepmc.org

Built with what exists

scripts adapted from ACI-Bench, an openly licensed set of visit dialogues and notes; actors and ordinary recorders; the modified PDQI-9, the 10-item note-quality form used in the Australian study; a speech-to-text model such as ElevenLabs Scribe v3 to make reference transcripts (see the Note "Fewer word errors"); free trials or maker permission for each scribe.

europepmc.org

  1. First step

    Adapt 10 ACI-Bench dialogues into scripts, record them with two volunteer actors, and ask three scribe makers for test access.

  2. Done looks like

    At least 8 named products scored on all 50 visits by at least 3 blinded clinician raters, with agreement between raters reported, every omission and invented statement counted, and the audio, scripts and scores published under an open license.

Open 02 · the brief

A test for form converters

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public answer key of at least 100 blank clinical forms, such as new-patient intake and symptom questionnaires, each converted by hand into a FHIR Questionnaire with the right question types, answer lists and skip rules. Then score every tool that converts forms, commercial ones such as Health Samurai's and an open baseline built on a general AI model, question by question against the key.

Why it is still open

Health Samurai sells a converter but publishes no accuracy figures. Published tests of language models on FHIR map data fields between research databases, not forms; in a July 2026 test, three models made 314 errors in 2,583 matchings. A search found no published test set for turning clinical forms into FHIR Questionnaires.

health-samurai.ioeuropepmc.org

Built with what exists

HL7's Structured Data Capture implementation guide, version 4.0.0, the standard for FHIR forms; the LHC-Forms viewer and NLM Form Builder from the US National Library of Medicine's FHIR tools; the HL7 FHIR validator; a model that reads scanned documents, such as Claude or Gemini, for the open baseline (see the Note "Seven more flagged", on a Claude-based reviewer of drug-making records).

hl7.orglhcforms.nlm.nih.gov

  1. First step

    Collect 20 blank forms from hospital websites, convert 5 by hand to make the first keys, and run one converter on all 5.

  2. Done looks like

    100 forms taken from hospital and clinic websites, with their hand-made keys checked by a second person. At least three tools scored on the share of questions with the right wording, type and answer options, the share of skip rules that work, and whether each output passes the FHIR validator. Forms, keys, outputs and scores are published.