Industries / Health / Medical supply distribution
Recalls
A recall pulls a faulty drug, device or supply back through the chain that delivered it, from maker and wholesaler to every hospital, clinic and pharmacy shelf that holds it.
7
Milestones achieved, from 1978 to 2025.
2
To watch for: announced, not yet confirmed outside.
3
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1978
How deep a recall goes
FDA guidelines published on 16 June 1978 require each recall plan to name its depth, the level of the distribution chain it must reach: wholesale, retail, or the consumer or user. The firm must then run effectiveness checks, by visits, calls or letters, to verify that every customer at that depth received the notice and acted on it. The same part of FDA's rules sorts recalls into three classes, from Class I, where use of the product could cause serious harm or death, to Class III, where harm is not likely.
Credit
US Food and Drug Administration (21 CFR Part 7)
2020
A supplier's unseen move
In January 2020 Cardinal Health, a large US distributor that also sells products under its own name, recalled 9.1 million surgical gowns of AAMI Level 3, a protection class set by the Association for the Advancement of Medical Instrumentation, and took action on 2.9 million Presource procedure packs that contained them. One of its FDA-registered suppliers in China had moved some gown production to unapproved sites with uncontrolled environments, so the gowns could not be assured sterile. FDA's record of the recall lists 7,659,634 units shipped across the US and to 13 other countries and gives the root cause as "vendor change control"; Cardinal booked a $37 million charge.
LessonA distributor that sells under its own brand answers for factories it does not run.
Credit
Cardinal Health (recall and annual report); FDA (recall record)
2020
Recalls mapped, mostly by hand
In a six-month FDA pilot in 2019, drug makers, wholesalers including McKesson, pharmacies including CVS Health, and health systems including Novant Health and Yale New Haven traced how a drug recall moves. Their February 2020 report called the process manual, time-consuming and error-prone: a notice could take from a few days to a couple of weeks to work through a pharmacy chain or health system, returning the product could take three to six months, and because lot numbers were often missing from inventory records, many pharmacies and hospitals removed every lot of a drug, including lots that were not recalled.
LessonWhere stock records do not hold lot numbers, a recall removes more medicine than it needs to.
Credit
participants in FDA's Pilot Project Program under the Drug Supply Chain Security Act
2023
Eye drops through two distributors
On 30 January 2023 Global Pharma Healthcare began recalling preservative-free artificial tears sold under the EzriCare and Delsam Pharma names, 400,008 bottles in all, which had gone to two distributors and from them across the US; FDA tests found bacteria in unopened bottles. Investigators found 81 patients in 18 states carrying an extensively drug-resistant strain of Pseudomonas aeruginosa linked to the drops: 4 of 54 with clinical cultures died within 30 days, and 4 of 18 with eye infections had an eye removed. FDA classified the recall Class I on 14 July 2023, five and a half months after it began.
LessonThe investigators called for stronger requirements on US importers of over-the-counter products.
Credit
M. K. Grossman and colleagues (Clinical Infectious Diseases, 2024); FDA recall records
2024
Distributors warned over syringes
In November 2023 the FDA said it was evaluating leaks, breakage and other failures in plastic syringes made in China. On 18 March 2024 it sent warning letters to the Chinese maker Jiangsu Shenli and to two US distributors, Medline and Sol-Millennium, over syringes not cleared for sale in the US, and on 24 April 2024 a further letter to Cardinal Health; it advised hospitals to move away from syringes by the Chinese makers it named. Separately, in September 2023 Cardinal had recalled 3,113,846 Monoject syringes because some syringe pumps did not recognize them or work with them, and FDA's record gives the root cause as "vendor change control".
LessonA change of maker behind a distributor's own-brand product can break its fit with the hospital equipment it is used in.
Credit
FDA Center for Devices and Radiological Health
2025
Australia rewrites its recall rules
On 5 March 2025 Australia's Therapeutic Goods Administration (TGA) replaced its Uniform Recall Procedure for Therapeutic Goods with a new Procedure for Recalls, Product Alerts and Product Corrections. Each action has a class, and a Class I letter must carry the word "Critical" and a Class II letter the word "Urgent"; each also has a level, the furthest point it must reach: wholesaler, hospital, retail or consumer. The TGA passes approved customer letters to the state and territory recall coordinators but no longer checks the accuracy of companies' customer lists.
Credit
Therapeutic Goods Administration
2025
Device warnings in weeks
FDA's device center began a pilot on 21 November 2024 to publish early alerts about likely high-risk device recalls in five areas, among them cardiovascular and general hospital devices, and on 29 September 2025 extended it to all devices. Reviewing the alerts, lawyers writing on the FDA Law Blog found that public notice of such recalls, which had generally come 2 to 3 months after a maker's first letters to customers, now came about 1 to 3 weeks after.
Credit
FDA Center for Devices and Radiological Health; review by Steven J. Gonzalez and Anne K. Walsh (FDA Law Blog, October 2025)
Watch for
Recalls run in minutes
Honeywell says its TrackWise Recall Management, an AI-assisted cloud system for drug and device makers, bounds a recall to the affected batches more accurately and can cut execution time "from weeks to minutes" by replacing spreadsheets with automated workflows.
Achieved whena maker publishes, with an outside check, the time from a recall decision to confirmed notice at every distributor and hospital customer, before and after adopting the system.
Claimed by
Honeywell, May 2025
Recall notices weeks earlier
Inmar Intelligence, whose OneRecall service handles recalls, says a new generation of AI recall platforms delivers notices "weeks earlier than traditional sources" and matches them against a health system's purchase history and stock.
Achieved whena health system or researcher measures the time from a maker's letter to quarantine on the shelf, with and without such a platform, over many recalls.
Claimed by
Inmar Intelligence, August 2025
Open
01
A public recall-matching test
A test set of at least 200 real FDA drug and device recall records, each with its product codes and lot list, paired with made-up hospital and pharmacy stock files that contain the hard cases: look-alike lots (the letter O against the digit 0), stock lines with no lot recorded, pack and case codes for the same product, and nearby lots that are not recalled. A scoring script grades software and AI models on finding every affected stock line and nothing else.
The briefCredit · unclaimed02
Lot lists, checked for gaps
An audit of how completely regulators' public recall records list the affected lots, compared with the maker's own recall letter or notice, for at least 300 drug and device recalls in FDA's enforcement report from 2022 to 2026 and 100 in Australia's Database of Recalls, Product Alerts and Product Corrections, with a public table of records that are complete, partial, marked "all lots" or missing lots.
The briefCredit · unclaimed03
A clock for drug recalls
A public table, updated monthly, of the days from a firm's start of each US drug recall to FDA's classification and to its appearance in the weekly enforcement report, by class and year from 2012 to 2026, naming the slowest cases.
The briefCredit · unclaimed
Open 01 · the brief
A public recall-matching test
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A test set of at least 200 real FDA drug and device recall records, each with its product codes and lot list, paired with made-up hospital and pharmacy stock files that contain the hard cases: look-alike lots (the letter O against the digit 0), stock lines with no lot recorded, pack and case codes for the same product, and nearby lots that are not recalled. A scoring script grades software and AI models on finding every affected stock line and nothing else.
Why it is still open
The 2019 FDA pilot found that lot numbers are often missing from inventory records, so many pharmacies and hospitals pull every lot of a recalled drug. A 2021 account by Joan Melendez of Xcelrate UDI describes distributors that do not record the lots they sell and a recall database that lacked the relevant lots, after which recalled bone cement had to be pulled from several hospitals. Vendors such as TraceLink sell recall workflow software without published accuracy figures, and Lotline, a Devpost project dated September 2026, matches exact lots on fictional clinic stock and leaves lot ranges and serial-number recalls unsupported. No public scored test was found.
Built with what exists
openFDA's drug and device enforcement endpoints, whose Code Information field holds lot and serial numbers; Python; AI models such as Claude to read recall letters (the Note "Seven more flagged" covers a Claude-based document reviewer).
First step
Pull 50 Class I and II recalls from openFDA for 2025, sort them by how their lot lists are written (single lots, ranges, "all lots"), and build the first stock file by hand.
Done looks like
At least 200 recalls and 2,000 stock lines with an answer key checked by two people; at least 3 tools or models scored on precision and recall per stock line; data, key and code published.
Open 02 · the brief
Lot lists, checked for gaps
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An audit of how completely regulators' public recall records list the affected lots, compared with the maker's own recall letter or notice, for at least 300 drug and device recalls in FDA's enforcement report from 2022 to 2026 and 100 in Australia's Database of Recalls, Product Alerts and Product Corrections, with a public table of records that are complete, partial, marked "all lots" or missing lots.
Why it is still open
Hospitals and pharmacies match their stock against the lots a notice gives, and FDA defines its Code Information field as a list of all lot and serial numbers on the product. Yet FDA's own pilot found missing lot data a main reason recalls are slow, an industry author describes a recall database that lacked the relevant lots, and in December 2025 FDA told food firms that some retailers did not know which lots a recall covered. No study measuring lot completeness for drug or device recall records was found.
Built with what exists
openFDA enforcement endpoints; the TGA's recall database; makers' recall notices on their websites, collected with a web agent such as Browser Use 1.0 (see the Note "Deleting 80% to start over"); a language model to pull lot lists out of letters and PDFs, checked by hand.
First step
Take 30 Class I drug recalls from 2025, find each maker's own notice, and compare the lot lists by hand.
Done looks like
400 recalls audited, with the share of complete lot lists by class, product type, year and regulator; a second person rechecks a random 10%; code that reruns from the public records.
Open 03 · the brief
A clock for drug recalls
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public table, updated monthly, of the days from a firm's start of each US drug recall to FDA's classification and to its appearance in the weekly enforcement report, by class and year from 2012 to 2026, naming the slowest cases.
Why it is still open
For devices, FDA's early alerts have brought public notice of high-risk recalls to about 1 to 3 weeks after makers' letters, by the FDA Law Blog's review; the program covers devices. Studies of FDA drug recalls report causes, classes and the average 1.3 years from start to close, not the wait for classification. The artificial-tears recall began on 30 January 2023 and was classified Class I on 14 July 2023. By this file's count from openFDA, the 316 drug recall events FDA classified in 2024 waited a median of 19 days from the firm's start to classification, the 45 that included a Class I product 33 days, and the longest 785 days; no running series of these waits was found.
Built with what exists
openFDA's drug enforcement endpoint, whose records carry the recall start date, the classification date and the report date; Python; a small static website.
First step
Pull every drug recall classified in 2024 from openFDA and compute both waits for each recall event.
Done looks like
Every US drug recall event from 2012 to 2026 with both waits, medians by class and year, and a second person checking 50 random events against the weekly enforcement reports.