BRENNEN

Industries / Health / Medical devices

Tracking after sale

After a device is sold, makers, hospitals and regulators collect reports of faults and harm, trace devices by code, and recall or correct them when a problem is found.

7

Milestones achieved, from 1975 to 2025.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1975

    A national register of knees

    Göran Bauer, head of orthopedics in Lund, Sweden, saw knee replacement as a large human experiment that should be watched. In 1975 Sweden started the first national arthroplasty register, a record of joint replacements and of every later operation to replace a failed implant (a revision), so that implants could be compared. It covered all of Sweden by the early 1990s and held 150,000 knee replacements and 10,000 revisions by the end of 2008.

    pmc.ncbi.nlm.nih.gov

    Credit

    Göran Bauer, Lars Lidgren and the Swedish Orthopedic Society

  2. 1984

    Makers must report harm

    Since December 1984, FDA rules have required device makers that receive complaints of malfunctions, serious injuries or deaths to report them to the FDA. Hospitals and other user facilities had to report from November 1991. The reports sit in MAUDE, a public FDA database updated every month, which the FDA warns cannot be used to work out rates because events are under-reported and reports can be wrong.

    Lessonthe 1984 rule did not reach foreign makers, which were not required to register; a 1995 rule changed that.

    fda.govfda.gov

    Credit

    US Food and Drug Administration (FDA)

  3. 2010

    A registry catches a hip

    Australia's national joint replacement registry named DePuy's ASR metal-on-metal resurfacing hip an outlier in its 2007 report, meaning its revision rate was more than twice that of other resurfacing hips. An expert group advising Australia's regulator put this down to the difficulty of fitting the device. DePuy withdrew the ASR from Australia in December 2009 and, after data from the National Joint Registry of England and Wales, recalled it worldwide in August 2010, by when about 93,000 patients had received it.

    Lessona registry can spot a failing implant years before a worldwide recall, but only if its signal is acted on.

    mja.com.auaph.gov.au

    Credit

    Australian Orthopaedic Association National Joint Replacement Registry; National Joint Registry of England and Wales

  4. 2013

    A code on every device

    The FDA's unique device identification (UDI) rule of 24 September 2013 requires most device labels and packages to carry a UDI, in plain text and in a form machines can read, such as a barcode. The code holds a device identifier for the model and a production identifier for the lot or serial number. Makers load each model's details into a public database, the Global Unique Device Identification Database (GUDID), and reports of adverse events must include the UDI.

    govinfo.gov

    Credit

    US FDA

  5. 2018

    Journalists map recalls worldwide

    In the Implant Files, more than 250 journalists in 36 countries gathered recalls, safety alerts and field safety notices (letters in which a maker tells users about a fault and what to do) into a public database, first released on 25 November 2018 and later grown past 120,000 records. They found that fewer than 20% of the world's countries put device safety alerts and recalls online, and that some devices were recalled or banned in some countries but not others.

    medicaldevices.icij.org

    Credit

    International Consortium of Investigative Journalists (ICIJ) and 58 media partners

  6. 2021

    One recall, 15 million machines

    In June 2021 Philips Respironics recalled CPAP and BiPAP machines, which keep a sleeping person's airway open, and ventilators, because a polyurethane foam used to damp noise could break down into pieces and chemicals that users might breathe in or swallow. The recall covered about 15 million devices worldwide. On 9 April 2024 a US federal court entered a consent decree that restricts Philips' production of new CPAP and BiPAP machines at several US plants until it meets set conditions.

    Lessona part chosen for comfort can fail with age, and the replacement part needs its own testing, which the FDA said was still under review.

    fda.gov

    Credit

    Philips Respironics (recall); US FDA and a US federal court (consent decree)

  7. 2025

    A third of reports late

    Researchers led by Alexander Everhart of Washington University School of Medicine studied 4.4 million first reports that makers sent to MAUDE from September 2019 to December 2022. 71% arrived within the 30 days the FDA allows; 9.1% came more than 180 days late, and 15.5% had missing or impossible dates. Three makers sent 54.8% of the late reports, and late reports often came in bursts, in one case in the month the device was recalled.

    Lessona rise in reports can appear after a problem is known, not before.

    pmc.ncbi.nlm.nih.govraps.org

    Credit

    A. Everhart and colleagues, in The BMJ

Watch for

  1. EU incident data in EUDAMED

    The European Commission made the first four parts of EUDAMED, the EU device database, mandatory from 28 May 2026: registration of companies, registration of devices with their UDIs, notified-body certificates, and market surveillance. It says the vigilance part, which will hold reports of serious incidents and field safety notices, is still in development and will be released only when its use becomes mandatory.

    Achieved whenthe vigilance part is declared working in the Official Journal of the EU and serious-incident reports and field safety notices can be searched on EUDAMED's public site.

    health.ec.europa.eu

    Claimed by

    European Commission, November 2025

Open

  1. 01

    Recall warnings backtested

    For every Class I device recall (the FDA's most serious class) reported from 2018 to 2024, a record of the MAUDE reports that describe the recalled fault, and the first date a standard disproportionality test would have flagged it. Such a test checks whether one device type's share of a problem is unusually high compared with other devices. Publish the lead time, or "no signal", for each recall, once by the date FDA received each report and once by the date the maker learned of the event.

    The brief
    Credit · unclaimed
  2. 02

    World recall list, refreshed

    An open table, refreshed weekly, of device recalls and field safety notices from at least ten regulators' public pages, translated into English. The same action is matched across countries by maker, model and UDI where given, and each match shows the days between the first and the last country's notice.

    The brief
    Credit · unclaimed

Open 01 · the brief

Recall warnings backtested

Credit · unclaimed

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

What to build

For every Class I device recall (the FDA's most serious class) reported from 2018 to 2024, a record of the MAUDE reports that describe the recalled fault, and the first date a standard disproportionality test would have flagged it. Such a test checks whether one device type's share of a problem is unusually high compared with other devices. Publish the lead time, or "no signal", for each recall, once by the date FDA received each report and once by the date the maker learned of the event.

Why it is still open

The tests of this found cover one device family, the wires (leads) of implanted defibrillators: FDA staff led by Hesha Duggirala showed in 2012 that a disproportionality test on MAUDE would have flagged Medtronic's Sprint Fidelis lead earlier than it was actually caught, and Lisa Ensign's dissertation and Krish Rajashankar's student project repeated such tests for leads. Open tools predict which maker and product type will face a recall (Benedict Brunker's model) or flag weekly spikes in reports (MaudeSignal), without measuring the lead time recall by recall. Everhart and colleagues found that late reports often arrive in bursts, which can hide or shift a signal.

ebi.ac.ukgithub.comgithub.comgithub.comraps.org

Built with what exists

openFDA's device adverse event endpoint (26.8 million reports) and its enforcement endpoint, which lists 1,921 Class I recall entries reported from 2018 to 2024; vigipy, an open-source Python library for disproportionality tests; a language model such as Claude or Gemini to decide whether each report describes the recalled fault, with a person checking a sample (yes-or-no questions at scale are the subject of the Note "A thousand questions, one pass"); DuckDB.

  1. First step

    Pick 20 Class I recalls from 2022, pull three years of MAUDE reports for each one's product code, and mark by hand which reports describe the recalled fault.

  2. Done looks like

    Every Class I recall reported from 2018 to 2024 has a lead time or "no signal". A second person checks 100 random report-to-recall matches and agrees with at least 90%. The code reruns from openFDA's public files.

Open 02 · the brief

World recall list, refreshed

Credit · unclaimed

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

What to build

An open table, refreshed weekly, of device recalls and field safety notices from at least ten regulators' public pages, translated into English. The same action is matched across countries by maker, model and UDI where given, and each match shows the days between the first and the last country's notice.

Why it is still open

ICIJ's International Medical Devices Database, the only multi-country set found, added its final batch on 25 November 2019 and its data run only through April 2019. The EU's own list will wait for EUDAMED's vigilance part, still in development. ECRI, a US nonprofit, sells a recall alert service that matches notices to a hospital's inventory and says its alerts often arrive ahead of the FDA's, but its data is for paying members only. Regulators publish in different forms: Britain's MHRA posts weekly lists of field safety notices, and Malaysia's Medical Device Authority posts monthly listings of field corrective actions.

ecri.orgmedicaldevices.icij.orghealth.ec.europa.eugov.ukmda.gov.my

Built with what exists

Python scrapers (requests and Beautiful Soup, or Playwright for pages built by script); the openFDA recall endpoint (59,705 records); a language model such as Claude or Gemini to translate notices and pull out maker, model and lot (see the Note "The jobs agents repeat"); GUDID to match UDIs; DuckDB.

  1. First step

    Collect one year of MHRA's weekly field safety notice lists and match each notice to openFDA recall records by maker and model.

  2. Done looks like

    Ten or more regulators covered and refreshed weekly for three months. A second person checks 200 random cross-country matches and agrees with at least 95%. The median days from first to last country are published, with code and data under an open license.