Data dictionary

Share us_regulatory_feed, schema gold. Four tables, delivered as Delta Sharing (Parquet) and CSV.

Load one with delta_sharing.load_as_pandas("config.share#us_regulatory_feed.gold.recalls").

Source: U.S. FDA via openFDA (public domain, CC0). Apex is not affiliated with or endorsed by FDA, HHS or any U.S. government agency. FDA says openFDA results should be treated as unvalidated, and the data must not be used to make medical-care decisions. Not medical, legal or regulatory advice.

How to read these tables

Coverage. Enforcement reports by report_date and 510(k)/PMA decisions by decision_date, from 2024-01-01 onwards. Drug shortages: FDA's full current list. Device classification: FDA's full list.
Refresh. Weekly. ingest_date is the date of the weekly pull that produced the row.
History starts at our first snapshot (2026-11-04). We don't invent changes from before that date. Records FDA had already published by then have first_seen = 2026-11-04.
Change history, not recall tracking. Every table records changes to FDA's published records. FDA does not update a recall's status after classification, so status in recalls can be stale. Don't use this data for public alerts or to track a recall's real-world lifecycle.
Privacy. FDA's personal contact fields (names, emails, phone numbers of contacts) are removed. Phone numbers inside free text are replaced with [phone removed]. GMDN terms are not included.
Types. string or date (ISO YYYY-MM-DD). Empty source values are null. Codes are kept exactly as FDA publishes them.

recalls

Drug, device and food enforcement reports, cleaned and unified into one table. One row per recall_number (the latest published version). Device recalls are enriched with the product code and root cause from FDA's device recall database.

Column	Type	Meaning
`product_type`	string	`drug`, `device` or `food` (the enforcement report it came from)
`recall_number`	string	Key. FDA's recall number, e.g. `D-0119-2025` (drug), `Z-1737-2026` (device)
`event_id`	string	FDA recall event ID. One event can cover several products (several recall numbers)
`status`	string	`Ongoing`, `Completed`, `Terminated` or `Pending`, as published. Can be stale: FDA doesn't update it after classification
`classification`	string	Health-hazard class: `I` (most serious), `II`, `III`, or `Not Yet Classified`. The "Class " prefix is removed
`recalling_firm`	string	Firm that initiated the recall
`city`, `state`, `country`	string	Recalling firm's location
`voluntary_mandated`	string	E.g. `Voluntary: Firm initiated` or `FDA Mandated`
`initial_firm_notification`	string	How the firm first notified its customers (Letter, E-Mail, Press Release, ...)
`product_description`	string	Product recalled, as described in the report
`product_quantity`	string	Quantity in commerce, free text
`code_info`	string	Lot numbers, serial numbers and other identifying codes, free text
`reason_for_recall`	string	Why the product was recalled
`distribution_pattern`	string	Where the product was distributed, free text
`report_date`	date	Date FDA published the record in its weekly Enforcement Report
`recall_initiation_date`	date	Date the firm first notified the public or its consignees
`center_classification_date`	date	Date the FDA center classified the recall
`termination_date`	date	Date FDA terminated the recall (null if not terminated)
`device_product_code`	string	Device recalls only: three-letter FDA product code, from the device recall database. Joins to `device_clearances.product_code`
`root_cause_description`	string	Device recalls only: FDA's root-cause category, e.g. `Device Design`
`first_seen`	date	First weekly snapshot that contained this record
`last_seen`	date	Latest weekly snapshot that contained this record (equals `ingest_date` for every row in the current table)
`ingest_date`	date	Date of the weekly pull

device_clearances

Device 510(k) clearances (including De Novo numbers) and PMA approvals and supplements, with FDA's device classification joined on product_code. One row per submission.

Column	Type	Meaning
`submission_id`	string	Key. 510(k)/De Novo number (e.g. `K252083`, `DEN230064`), or PMA number and supplement number joined by `\|` (e.g. `P030017\|S396`)
`submission_type`	string	`510k` or `PMA`
`submission_number`	string	510(k) number or PMA number
`supplement_number`	string	PMA supplement number (null for 510(k) and original PMAs)
`applicant`	string	Company that made the submission
`device_name`	string	510(k): device name. PMA: trade name
`generic_name`	string	PMA only: generic device name
`product_code`	string	Three-letter FDA product code
`decision_code`	string	FDA decision code, e.g. `SESE` (substantially equivalent), `APPR` (approved), `OK30` (30-day notice accepted), `DENG` (De Novo granted)
`decision_description`	string	510(k) only: decision in words, e.g. `Substantially Equivalent`
`review_type`	string	510(k): clearance type (`Traditional`, `Special`, `Abbreviated`). PMA: supplement type (`30-Day Notice`, `Real-Time Process`, `Normal 180 Day Track`, ...)
`supplement_reason`	string	PMA only: why the supplement was filed
`advisory_committee`	string	Two-letter review panel code, e.g. `CV`
`advisory_committee_description`	string	Review panel name, e.g. `Cardiovascular`
`city`, `state`	string	Applicant's location
`country_code`	string	510(k) only: applicant's country code
`date_received`	date	Date FDA received the submission
`decision_date`	date	Date of FDA's decision
`device_class`	string	From device classification: `1`, `2`, `3`, `U` (unclassified), `N` (not classified) or `f` (HDE)
`medical_specialty`	string	From device classification: two-letter specialty code
`medical_specialty_description`	string	From device classification: specialty name
`regulation_number`	string	From device classification: 21 CFR regulation number, e.g. `878.4810`
`classification_device_name`	string	From device classification: FDA's name for the product code
`first_seen`	date	First weekly snapshot that contained this record
`last_seen`	date	Latest weekly snapshot that contained this record
`ingest_date`	date	Date of the weekly pull

Classification columns are null when the product code has no match in FDA's classification list.

drug_shortages

FDA's published drug shortage list, one row per presentation (package). Week-to-week changes appear in weekly_changes.

Column	Type	Meaning
`shortage_key`	string	Key. package NDC, generic name, company and presentation joined by `\|`, plus `#n`. FDA publishes no single ID, so `#n` separates rare exact duplicates
`package_ndc`	string	National Drug Code of the package
`generic_name`	string	Generic drug name and form
`company_name`	string	Manufacturer or labeler
`presentation`	string	Product, form, strength and package, as FDA lists it
`dosage_form`	string	E.g. `Injection`, `Tablet`, `Capsule`
`therapeutic_category`	string	One or more categories, separated by `;`
`status`	string	`Current`, `Resolved` or `To Be Discontinued`
`availability`	string	E.g. `Available`, `Limited Availability`, `Unavailable`, or FDA's free text
`shortage_reason`	string	Reason given for the shortage
`related_info`	string	Additional information FDA publishes, free text
`resolved_note`	string	Note published when the shortage is resolved
`update_type`	string	FDA's last update type: `New`, `Revised` or `Reverified`
`initial_posting_date`	date	Date FDA first posted the shortage
`update_date`	date	Date of FDA's last update
`change_date`	date	Date FDA last changed the record
`discontinued_date`	date	Date the product is or was discontinued, if any
`first_seen`	date	First weekly snapshot that contained this record
`last_seen`	date	Latest weekly snapshot that contained this record
`ingest_date`	date	Date of the weekly pull

weekly_changes

One row per record that was added, changed or removed in FDA's published data since the previous weekly snapshot. The first snapshot (2026-11-04) lists every record as new.

Column	Type	Meaning
`row_key`	string	Key of the changed record: `recall_number`, `submission_id` or `shortage_key`
`table_name`	string	`recalls`, `device_clearances` or `drug_shortages`
`change_type`	string	`new`, `updated`, `terminated` (a recall whose `termination_date` became set) or `removed` (no longer published by FDA)
`changed_fields`	string	For `updated` and `terminated`: the comma-separated columns whose values changed. Null otherwise
`ingest_date`	date	Weekly snapshot in which the change was detected

Example, everything that changed this week in recalls: filter table_name = 'recalls' and ingest_date = the latest date, then join row_key to recalls.recall_number.

Every table is a change history of FDA's published records. It does not track the real-world status of a recall. FDA does not update a recall's status after classification.