Health Care Supplies · U.S. FDA openFDA
openFDA Device Recalls API
Datadory delivers openfda device recalls api data covering 58,999 FDA medical device recall event records from November 2002 to present - recalling firm with FEI number, device description and product codes, reason for recall, categorized root cause, corrective action, affected quantity and distribution pattern, with 510(k) and PMA numbers joined where they exist, delivered daily, weekly, or hourly.
API, files, or your warehouse. Daily, weekly, or hourly.
What is the openFDA Device Recalls API dataset?
The enforcement side of the device world, one event at a time. openFDA Device Recalls is the FDA's running record of medical device recalls - formally, a firm's removal or correction of a marketed product that the agency considers in violation of the laws it administers, reported under 21 CFR 806.10. At the August 2026 research pass the corpus held 58,999 recall event records, each carrying the recalling firm with its FEI number and address, a device description down to brand and catalog number, the reason for recall, a categorized root cause, the corrective action taken, the affected quantity, and a distribution pattern recording how far the product traveled.
Two structural facts make the corpus more than a news archive. First, the timeline runs deep: classified recalls reach back to November 1, 2002, and from January 3, 2017 the record also takes in corrections or removals firms initiated before FDA review. Second, the rows join upward: where a device cleared premarket review, its 510(k) or PMA numbers sit on the same record, so an enforcement event connects to the clearance trail that put the device on the market. Get a sample of this dataset cut to your device categories.
What do sample rows look like?
Four real recall events spanning two decades of enforcement, flattened for reading:
res_event_number : 86352
recalling_firm : Ansell Healthcare Products LLC
product : MICROFLEX Diamond Grip Examination Gloves, MF-300
reason : Examination gloves were shipped inadvertently,
without [testing] to verify barrier integrity.
status : Terminated initiated : 2020-08-19
quantity : 1312 Cases
distribution : U.S. Nationwide distribution incl. IL, OK, WI, PA, TX, OH and others.
----------------------------------------------------------------------------------------------
res_event_number : 71763
recalling_firm : Toshiba American Medical Systems Inc
product : Celesteion PCA-9000A/2 PET/CT System
reason : Raw data may not be saved during specific multi-phase helical
scanning operations due to a software problem.
root_cause : Device Design status : Terminated
initiated : 2015-05-08 product_code : KPS k_numbers : K140651
----------------------------------------------------------------------------------------------
res_event_number : 27549
recalling_firm : Sedecal USA, Inc
product : Sedecal SP-HF 4.0 Portable X-Ray System
reason : Minimum source-skin-distance below 30 cm and missing tube
manufacturer identification; failed the Federal performance standard.
root_cause : Other status : Terminated
initiated : 2003-10-27 product_code : IZL k_numbers : K020436
----------------------------------------------------------------------------------------------
res_event_number : 29815
recalling_firm : DeRoyal Wound Care
product : Aquasorb Border, Hydrogel Wound Dressing ... Catalog # 46-511
code_info : Lot 360025 product_code : MGQ state : VA
initiated : 2004-07-08 status : TerminatedRead together, three habits of the corpus surface. Root cause is a taxonomy, not a guess - the PET/CT scanner lands on Device Design while the older X-ray record falls back to Other, so treat the category as curated but incomplete. Scope lives in code_info: a lot number, serial range or catalog code decides whether a recall touches one case of gloves or a nationwide fleet of imaging systems. And distribution_pattern is the exposure map - the glove recall names its states outright, which is precisely the column a supply-chain screen joins against. Your sample pins down current totals before any commitment.
What fields does the dataset include?
Twenty-nine verified fields cover the full arc of a recall event - who recalled, what and how much, why, what was done, and when each milestone was recorded. Definitions map to the published field reference gathered during the August 2026 research pass, so nothing here is inferred:
Where does coverage run, and at what grain?
Geography - United States enforcement actions against FDA-registered firms, with distribution_pattern preserving the states, territories and international destinations the affected product actually reached. A recall can therefore expose foreign shipments while remaining an account of actions under FDA jurisdiction.
Temporal - classified recalls from November 1, 2002 to present, with firm-initiated corrections and removals added from January 3, 2017. Five separate date columns - initiated, created, posted, terminated, plus record creation - let you measure lag directly instead of guessing at it.
Granularity - one record per recall event/product registration entry, keyed by res_event_number with cfres_id beneath it. Set against the wider Datadory catalog - where the average quality score across all cataloged datasets is 7.81 - this record scores 10/10, carried by verified field definitions, complete date coverage and direct premarket-submission joins.
How is the data delivered?
API, files, or your warehouse. Daily, weekly, or hourly.
Cadence is yours to set, and to change when your risk model changes - take one full snapshot back to November 2002, or keep a warehouse current so newly recorded recall events diff cleanly into yesterday's rows.
Deliveries arrive normalized to the field dictionary above, with
res_event_numberkept intact as the event-level key and the harmonized annotation block riding beside the raw fields, so no column needs inferring after the fact. Every shipment carries validation rows and coverage notes mapped to whichever device categories you named - the schema you see in the sample is the schema you ship against.
Who uses this data, and for what?
Two decades of dated enforcement earns its keep on specific jobs:
- Supplier and supply-chain risk screening - vendors and SKUs get checked against enforcement history before they reach an approved-supplier list;
distribution_patternshows whether your regions were reached. - QA/RA precedent research -
reason_for_recallandroot_cause_descriptiongive regulatory teams precedent language for CAPAs, design changes and submission responses. - Investor and diligence monitoring - recall frequency, root-cause mix and termination lag by firm become operational-quality screens.
- Compliance and litigation timelines - five dated milestones reconstruct how long each action took, event by event.
- Model training on failure modes - 58,999 labeled events pairing reason narratives with a categorical root cause train classifiers that flag risky designs before the next field action.
- Procurement and category management - buyers spot device classes with recurring enforcement patterns before committing to a multi-year agreement.
Which personas get the most value?
Data Scientists & Analysts get a corpus that behaves: labeled events, a stable event-number key, a categorical target and twenty-plus years of depth. Competitive Intelligence & Product Teams get documented, dated enforcement history across competitor catalogs - positioning ammunition no win/loss interview matches. Investors & Quants read operational quality straight off recall counts, root-cause mix and time-to-termination. Sales & Growth Teams get trigger events worth timing outreach against. Journalists, Academics & Students get a quotable public-interest record with named firms, dated events and stated causes.
Which notes and datasets pair with it?
- Provenance - compiled during the August 2026 research pass from U.S. FDA openFDA documentation; field definitions map to the published field reference rather than inferred conventions.
- Status has a shelf life by design - once FDA classifies a recall, the record is not revised beyond corrections to Enforcement Report entries, so termination dates are authoritative even where late-breaking detail never arrives. Plan monitoring accordingly.
- Enforcement history, not an alerting feed - FDA itself cautions the data should not drive medical decisions or public alerts; treat it as the record of what happened, not a substitute for safety communications.
- Where it pairs - complaints come from the openFDA Device Adverse Events (MAUDE) API, facility identity from the openFDA Device Registration & Listing API, item-level master data from the openFDA Unique Device Identifier (GUDID) API, and the openFDA Device 510(k) Clearances API supplies the premarket baseline. The Device 510(k) Clearances vs Device Recalls comparison frames clearance promise against enforcement reality, and the best health-care-supplies datasets ranking places the whole shelf.
Source: U.S. FDA openFDA.
Field dictionary
Every field below is documented against real records. The full dictionary ships with the sample.
| field | type | definition | example |
|---|---|---|---|
cfres_id | string | Internal FDA identifier for the recall event record. | <returned in your sample> |
res_event_number | string | Recall event number assigned by FDA; the event-level key multiple product entries can share. | 86352 |
product_res_number | string | Product-specific recall number in the Z-####-## format published in enforcement reports. | Z-0001-04 |
event_date_initiated | date | Date the recalling firm initiated the recall. | 2003-10-27 |
event_date_created | date | Date the recall record was created in FDA's system. | <returned in your sample> |
event_date_posted | date | Date the recall was posted publicly. | <returned in your sample> |
event_date_terminated | date | Date FDA terminated the recall after corrective action was verified. | <returned in your sample> |
recall_status | enum | Lifecycle stage of the recall - Ongoing, Completed or Terminated. | Terminated |
recalling_firm | string | Name of the firm conducting the recall. | Sedecal USA, Inc |
firm_fei_number | string | FDA Establishment Identifier of the recalling firm; the join key into facility registries. | <returned in your sample> |
address_1 | string | Primary street address of the recalling firm. | <returned in your sample> |
address_2 | string | Secondary address line of the recalling firm. | <returned in your sample> |
city | string | City of the recalling firm. | <returned in your sample> |
state | string | State of the recalling firm. | VA |
postal_code | string | Postal code of the recalling firm. | <returned in your sample> |
country | string | Country of the recalling firm. | <returned in your sample> |
additional_info_contact | string | Contact person and phone number recorded for further recall information. | <returned in your sample> |
reason_for_recall | text | Unstructured description of the defect or violation prompting the recall. | Examination gloves were shipped inadvertently, without testing to verify barrier integrity. |
root_cause_description | string | Categorized root cause assigned to the recall - Device Design, Other and peers. | Device Design |
action | text | Corrective action taken or planned by the recalling firm. | <returned in your sample> |
product_description | text | Description of the recalled device including brand, model, catalog number and manufacturer details. | MICROFLEX Diamond Grip Examination Gloves, MF-300 |
code_info | string | Lot, serial or model codes identifying the affected units. | Lot 360025 |
product_code | string | Three-letter FDA regulation product code for the device type. | KPS |
k_numbers | text | Structured text field. 510(k) clearance numbers associated with the recalled device. | K140651 |
pma_numbers | text | Structured text field. Premarket Approval application numbers associated with the recalled device. | <returned in your sample> |
other_submission_description | string | Description of other premarket submissions related to the device. | <returned in your sample> |
product_quantity | string | Quantity of product subject to the recall. | 1312 Cases |
distribution_pattern | text | Geographic distribution of the recalled product - states, territories or international. | U.S. Nationwide distribution including IL, OK, WI, PA, TX, OH and others. |
openfda | text | Structured text field. Harmonized annotation block joined by openFDA where the device could be resolved. | <returned in your sample> |
What teams do with it
- Supplier and supply-chain risk screening Screen vendor lists and stocked SKUs against enforcement history; distribution_pattern tells you whether the regions you buy and ship in were actually reached.
- QA/RA precedent research reason_for_recall and root_cause_description hand regulatory teams precedent language for CAPAs, design changes and submission responses.
- Investor and diligence monitoring Recall frequency, root-cause mix and termination lag by recalling firm feed screens on manufacturer operational quality.
- Compliance and litigation timelines Five dated milestones from initiation to termination reconstruct exactly how long each action took, event by event.
- Model training on failure modes 58,999 labeled events pairing unstructured reason narratives with a categorical root cause give classifiers both signal and target.
- Procurement and category management Category managers spot device classes with recurring enforcement patterns before signing a multi-year supply agreement.
Questions buyers ask
What does the openFDA device recalls dataset contain?
One record per recall event and product entry: recalling firm with FEI number and full address, device description with brand, model and catalog detail, lot or serial code information, reason for recall, categorized root cause, corrective action taken, product quantity, distribution pattern, five dated lifecycle milestones from initiation to termination, and 510(k) or PMA numbers where a premarket submission exists.
How far back does device recall coverage go?
Classified recalls reach back to November 1, 2002 - more than two decades of enforcement history for longitudinal work. From January 3, 2017 onward the record also captures corrections or removals firms initiated before FDA review, so post-2017 analyses see a wider population of actions than earlier eras - worth flagging in any trend chart that spans the boundary.
Can recall records be joined to clearance or adverse-event data?
Yes, on three keys. product_code carries the three-letter FDA device category shared across the FDA device stack, k_numbers and pma_numbers tie a recalled device back to its premarket submissions, and the harmonized annotation block adds device name, class and regulation number where the device could be resolved. Those joins connect recalls to clearances, adverse-event reports and identifier records.
Is the data suitable for training machine learning models?
It trains classification and screening models well: 58,999 labeled examples pairing an unstructured reason narrative with a categorical root cause and a corrective-action account give supervised learning both signal and target. Treat two caveats as design features - status reflects the lifecycle stage recorded when FDA acted, and the narratives carry the nuance the coded fields compress.
Does the data cover recalls outside the United States?
The enforcement actions themselves are United States recalls by FDA-registered firms, but the distribution_pattern field records how far affected product traveled, including states, territories and international destinations. A recall can therefore surface foreign exposure - useful when screening a supplier network that ships across borders - while remaining an account of actions taken under FDA jurisdiction.
Can I get a sample cut to my device categories?
Yes. Name the product codes, recalling firms, root causes or date windows and the sample arrives in exactly the schema shown above, extended across whichever slice you need. Delivery runs through API, files, or your warehouse on a daily, weekly, or hourly cadence, with validation rows and coverage notes documented alongside your sample.
Datasets that pair with this one
- openFDA Device Adverse Events (MAUDE) API The complaint layer over the same device world - injury and malfunction reports that often precede or accompany these enforcement actions.
- openFDA Device Registration & Listing API Facility-to-product registry that tells you who made the device; FEI number joins the two records directly.
- openFDA Unique Device Identifier (GUDID) API Item-level device master records to hang identifiers and labels off the recall events.
- openFDA Device 510(k) Clearances API The premarket baseline - k_numbers on a recall row point straight back to the clearance that put the device on the market.
- Health Care Supplies data hub The pooled industry view across enforcement, identifiers, trials and trade data.
- Sample policy Samples ship in the exact schema shown above, cut to the product codes, firms, root causes or date windows you name.
See the rows before you pay anything.
Name this dataset and we send real records from it — scoped to the fields you asked for.