Industries / Health / Healthcare / HealthTech
Sharing data between systems
Sharing data between systems is the work of agreeing on message formats, codes, networks and rules so that one hospital's or clinic's software can send records, test results and images to another's and have them understood at the other end.
7
Milestones achieved, from 1987 to 2023.
3
To watch for: announced, not yet confirmed outside.
3
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1987
One message format for hospitals
In 1987 a new standards group, HL7 (Health Level Seven), began writing one format for the data that hospital computers send each other, such as admissions, orders and results. Before it, every link between two systems had to be negotiated from scratch. Donald Simborg, one of HL7's founding organizers, recalls that connecting a lab system to a hospital mainframe then cost about $100,000 in software fees, and that his company gave its StatLAN protocol to HL7 as the basis for versions 1 and 2. HL7 version 2 is still the most used exchange standard in the United States: in 2024 HL7 reported it in use at 95% of US healthcare organizations and in more than 35 countries, as the US National Library of Medicine relays.
Credit
Donald Simborg and Simborg Systems; Wes Rishel, who led the version 1 team; Sam Schultz, HL7's first president
1994
A national messaging network
Denmark set up MedCom in 1994, a publicly funded nonprofit, to agree national standards for the electronic messages that pass between family doctors, hospitals, labs and pharmacies. Uptake was slow at first: a 1998 Copenhagen Business School study found that on average only 15% of discharge letters, 7% of lab results and 10% of prescriptions went by EDI (electronic data interchange, structured messages between computers). By 2003 MedCom staff reported that more than 70% of health documents between hospitals and primary care, about 25 million a year, went electronically, and that only MedCom-certified systems were in use. By 2010 more than 5 million clinical messages a month passed through the network, and one standard letter format was used in over 5,000 institutions with 50 vendors' systems, by Denis Protti and Ib Johansen's count for the Commonwealth Fund.
Lessonone national body that tested and certified every vendor's messages gave a whole country a single way to exchange.
Credit
MedCom, whose certification project was financed by the Danish Ministry of Health, the Danish County Association and others
1998
Vendors test connections together
The radiology society RSNA and the health IT society HIMSS held the first IHE (Integrating the Healthcare Enterprise) workshop on 13 and 14 October 1998, with 13 companies, to ask why manufacturers could not make standards work in clinical practice. IHE writes profiles, recipes that fix exactly how existing standards are used for one job, and tests them at Connectathons, where vendors connect their systems to each other's under supervision. The first, in radiology in 1999, led to a demonstration at RSNA 1999 with 23 vendors and 47 systems; Connectathons now run every year in Asia, Europe and North America, and results are published. In 2015 the European Commission (Decision 2015/1302) named 27 IHE profiles, including ones for finding a patient and fetching documents across communities and for pharmacy prescriptions, that public buyers in the EU may cite in tenders.
Lessonproducts written to the same standard still failed to connect until vendors tested them against each other.
Credit
RSNA and HIMSS
2006
A model exchange shuts down
The Santa Barbara County Care Data Exchange, a regional health information exchange (a shared service that passes records between a community's hospitals, clinics and labs), was one of the most ambitious and publicized in the US. It closed in December 2006, about eight years after it began and several months after it started providing some data; participants found no compelling value in its first services. Its founder, David Brailer, had meanwhile gone on to lead US federal health IT efforts. Regional exchanges kept closing: a national survey counted 89 in operation in 2019 and 76 in 2023.
Lessonwithout some mix of grants, incentives and mandates, regional exchanges may not be able to pay for services that are socially valuable but unprofitable.
Credit
David Brailer, founder; Robert H. Miller and Bradley S. Miller, who analyzed the closure
2014
Certified records full of errors
From 2014 the second stage of the US Meaningful Use incentive program required certified record systems to exchange C-CDA (Consolidated Clinical Document Architecture) documents, a standard XML summary of a patient's problems, medicines, allergies and results. The SMART C-CDA Collaborative contacted 107 health IT organizations and collected 91 sample documents from 21 different technologies. Parsing, the certification validator, an automated scoring tool and manual review turned up 615 errors and variations in how the same data were written, and the authors named 11 areas that would stop documents from being understood at the other end.
Lessonpassing certification did not make two systems read a document the same way.
Credit
John D'Amore, Joshua Mandel, Kenneth Mandl, Isaac Kohane, Rachel Ramoni and colleagues
2019
Prescriptions cross a border
In January 2019 an electronic prescription written in Finland could be filled at a pharmacy in Estonia, the first use of the EU's cross-border ePrescription service, now called MyHealth@EU, by a 2023 study's account; Estonian prescriptions followed in Finnish pharmacies in 2020. The patient consents in advance, and each country's national contact point (in Finland, run by Kela) passes the prescription across. In a 2021 survey of pharmacists at the 664 pharmacies that had filled such prescriptions, 88% of Estonian and 81.8% of Finnish respondents agreed access to medicines had improved, while 61% and 42.8% had met ambiguities or errors, most often a wrong pharmaceutical form or total amount. The EU's European Health Data Space regulation, in force since 26 March 2025, requires every member state to exchange patient summaries and ePrescriptions by March 2029, and medical images, lab results and discharge reports by March 2031.
Lessona prescription that arrives intact can still be hard to fill when forms, pack sizes and products differ between countries.
Credit
the EU eHealth Network and European Commission; Finland's Kanta Services and Kela
2023
A national network of networks
The US 21st Century Cures Act of 2016 called for a trusted exchange framework: one common agreement under which separate health information networks answer each other's requests for records. TEFCA went live on 12 December 2023, when the first Qualified Health Information Networks (QHINs), networks approved to connect directly to each other, were designated, among them eHealth Exchange, Epic Nexus, Health Gorilla, KONZA National Network and MedAllies. It followed private frameworks such as Carequality, which first carried records between an Epic health system, Sutter Health, and athenahealth practices on 1 July 2016 and handled more than 6 billion documents in 2023. By the US health IT office's count, about 10 million documents crossed TEFCA before 2025 and 464 million by the end of 2025, and by November 2026 more than 71,000 sites took part through 11 QHINs.
rss.globenewswire.comhealthit.govmedcitynews.comtechtarget.com
Credit
the Office of the National Coordinator for Health IT (now ASTP/ONC); The Sequoia Project, its coordinating contractor
Watch for
Penalties for blocking records
Information blocking, under the US Cures Act, means practices that interfere with the access, exchange or use of electronic health information. Fines of up to $1 million per violation have applied to health IT developers and exchanges since 1 September 2023, and penalties for providers since 1 July 2024, yet by February 2026 nearly 1,600 complaints had been filed and no enforcement action taken since 2021. On 3 September 2025 HHS said Secretary Robert F. Kennedy Jr. had directed it to put more resources into enforcement, and in February 2026 Thomas Keane, head of ASTP/ONC, said his office was issuing notices of investigation of potential nonconformity to health IT developers.
Achieved whenthe HHS Office of Inspector General or ASTP/ONC publishes a finished penalty, disincentive or loss of certification against a named developer, network or provider.
Claimed by
HHS and ASTP/ONC, September 2025 and February 2026
Scans without the CD
Epic says its Care Everywhere Diagnostic Image Exchange, released across its customers in August 2026, lets a clinician pull full diagnostic-quality CT, X-ray and MRI images from another health system's image archive with one click, using open image-exchange standards. Until now Epic sites shared written reports and lower-detail reference images, and Epic says millions of US patients still carry their scans on CDs. Epic calls itself "the first to embrace open image-sharing standards at scale" and says its existing duplicate-order checks already prevent more than 21,000 repeat scans a year.
Achieved whena health system or researchers outside Epic report how many outside studies were retrieved this way, how often a CD or manual transfer was still needed, and whether repeat scans fell.
Claimed by
Epic (Matt Doyle, director of interoperability), August 2026
A billion linked records
India's National Health Authority says its Ayushman Bharat Digital Mission (ABDM) had linked more than 100 crore (1 billion) health records to ABHA accounts, 14-digit personal health IDs, by May 2026, double the 50 crore of February 2025, with more than 450 public and private software products connected; in July 2026 the government gave 104 crore records and 93 crore accounts. The figures count records linked to an ID; they do not say how many were sent from one provider to another. An outside study at three government hospitals in Agra found ABHA registration at 23.96% of 537,278 outpatient visits between September 2024 and April 2025.
Achieved whenan independent study reports how often linked records are actually fetched by a different provider with the patient's consent.
Claimed by
National Health Authority and Government of India, May and July 2026
Open
01
Score every C-CDA converter
A public, repeatable scoreboard that runs every available converter from C-CDA, the US summary-of-care format described in the 2014 entry above, to FHIR, the newer standard that splits a record into small pieces software can request over the web. Use paired synthetic records: Synthea writes each made-up patient both as a C-CDA document and as a native FHIR bundle, so the native bundle is the answer key. Name each product and version, and report what each one drops, moves or changes. Synthetic patients only; no real records.
The briefCredit · unclaimed02
Answers for HL7's test messages
Hand-built, validated FHIR bundles for the seven test messages in HL7's guide for converting HL7 version 2 messages to FHIR (an admission, a scheduling message, two kinds of order, a lab result, a clinical document and an immunization), written from the guide's mapping tables and checked by two reviewers. Then about 200 more synthetic v2 messages made from Synthea patients, including the local variations real interfaces add, each with its expected bundle. Score named converters and AI models against all of them. Synthetic messages only.
The briefCredit · unclaimed03
A key for merged records
A public test set of at least 200 synthetic patients whose records arrive from 3 to 5 sources (the hospitals, primary care clinics and urgent care centers Synthea assigns them), each source sending only what it recorded, as C-CDA documents or FHIR bundles. The same drug, allergy or diagnosis then appears several times with different codes, dates and wording. Each patient gets a hand-checked answer key: one correct current medication list, allergy list and problem list. Score deduplication tools and AI models on what they merge wrongly, keep twice or drop. Synthetic patients only.
The briefCredit · unclaimed
Open 01 · the brief
Score every C-CDA converter
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public, repeatable scoreboard that runs every available converter from C-CDA, the US summary-of-care format described in the 2014 entry above, to FHIR, the newer standard that splits a record into small pieces software can request over the web. Use paired synthetic records: Synthea writes each made-up patient both as a C-CDA document and as a native FHIR bundle, so the native bundle is the answer key. Name each product and version, and report what each one drops, moves or changes. Synthetic patients only; no real records.
Why it is still open
The only published test of this kind, an October 2026 preprint, ran one converter, python-fhir-converter 0.3.0 (a Python port of Microsoft's templates), on 331 Synthea patients. Of 276,153 source entries 96.9% survived, but only 89.8% landed where FHIR expects them; no timestamp kept its time of day, no medication or condition kept its status, and every medication failed validation. The author lists "We tested one converter, one version and default settings" among the limits. A 2026 review of 39 studies found that FHIR mapping quality is still judged mostly by hand, with little agreement on method.
Built with what exists
the preprint's MIT-licensed code, which is meant to be rerun on other converters; Synthea with its C-CDA and FHIR exporters switched on; Microsoft FHIR-Converter, the upstream.NET version; python-fhir-converter; the ccda-to-fhir Python library; Metriport's C-CDA-to-FHIR converter API; commercial converters with their makers' permission; HL7's C-CDA on FHIR mapping tables as the rules; the HL7 FHIR Validator and the Inferno US Core test kit.
github.comgithub.comgithub.comgithub.compypi.orgdocs.metriport.combuild.fhir.orghl7.orggithub.com
First step
Rerun the published harness on python-fhir-converter to reproduce its numbers, then add upstream Microsoft FHIR-Converter and ccda-to-fhir.
Done looks like
At least 5 named converters, versions stated, scored on the same 331 or more patients for strict and lenient recall in each clinical family, share of timestamps, statuses and codes kept, and share of resources passing US Core validation. A second run with Synthea's missing time-zone offsets fixed separates faults in the source from faults in the converter. Code, data and scores published and rerun at each new converter release.
Open 02 · the brief
Answers for HL7's test messages
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
Hand-built, validated FHIR bundles for the seven test messages in HL7's guide for converting HL7 version 2 messages to FHIR (an admission, a scheduling message, two kinds of order, a lab result, a clinical document and an immunization), written from the guide's mapping tables and checked by two reviewers. Then about 200 more synthetic v2 messages made from Synthea patients, including the local variations real interfaces add, each with its expected bundle. Score named converters and AI models against all of them. Synthetic messages only.
Why it is still open
HL7's V2-to-FHIR guide (version 1.0.0) says its test messages and "the corresponding agreed to FHIR Bundle (once available)" are meant for validating mapping tools and will later become "the agreed to benchmark", but all seven bundles are marked "To be provided", in both the published guide and the continuous build. A 2026 review notes that local customization of v2 has long left messages semantically incomplete between sites. Searches of Europe PMC, arXiv and GitHub found no published comparison of v2-to-FHIR converters; the nearest AI benchmark, FHIRBench (July 2026), tests how FHIR formatting affects model answers, not conversion.
Built with what exists
the guide and its mapping content on GitHub; Microsoft FHIR-Converter, whose HL7v2 templates are based on v2.8 and which also has a preview FHIR-to-HL7v2 mode for drafting messages; the LinuxForHealth hl7v2-fhir-converter; the HL7 FHIR Validator; Synthea; general AI models such as Claude or Gemini, and a small open-weight model fine-tuned on one GPU such as Liquid AI LFM3 (see the Note "One GPU to fine-tune").
First step
Write the expected bundle for the guide's ADT_A01 admission message from its message and segment maps, validate it, and post it for review on chat.fhir.org, the FHIR community's chat.
Done looks like
Seven expected bundles that pass the validator, agreed by two independent reviewers and offered to the HL7 project; 200 more messages with expected bundles; at least 4 converters and 3 AI models scored field by field on the share of mapped fields correct and the share of resources valid, with all files published under an open license.
Open 03 · the brief
A key for merged records
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test set of at least 200 synthetic patients whose records arrive from 3 to 5 sources (the hospitals, primary care clinics and urgent care centers Synthea assigns them), each source sending only what it recorded, as C-CDA documents or FHIR bundles. The same drug, allergy or diagnosis then appears several times with different codes, dates and wording. Each patient gets a hand-checked answer key: one correct current medication list, allergy list and problem list. Score deduplication tools and AI models on what they merge wrongly, keep twice or drop. Synthetic patients only.
Why it is still open
Records pulled from several networks overlap, and the tools that tidy them are not tested in public. Particle Health, which pulls C-CDAs from networks and calls duplicate clinical data "a prevalent issue", says its deduplication removes a resource only when "every non UUID field" is identical to another, so near-duplicates pass through. The nearest AI test, an April 2026 preprint, scored models on one FHIR record per patient for 200 synthetic patients and found omission the main failure; it did not merge records from several sources. A search found no public answer key for merging records from several sources.
Built with what exists
Synthea, which places patients with hospitals, primary care and urgent care providers drawn from CMS data and exports C-CDA and FHIR; the converters from the first item to bring documents into FHIR; NLM's RxNorm API to compare drug codes; general AI models such as Claude or Gemini, and Meta Muse Ember for extraction from long documents (see the Note "The jobs agents repeat"); a scoring script.
First step
Generate 20 Synthea patients with C-CDA and FHIR export on, split each patient's encounters by the provider that recorded them into separate source files, and build the first 20 answer keys by hand.
Done looks like
200 patients with 3 to 5 sources each and answer keys checked by two reviewers, one with pharmacy or nursing training; at least 3 tools or models scored on precision and recall for each list and on false merges, such as two different drugs combined into one; data, keys, code and scores published under an open license.