The complete response letter atlas

What is in the atlas

Every complete response letter the FDA has published through openFDA (2002–2026), each matched to its application, its sponsor and the sponsor's SEC filings. A complete response letter is the FDA's decision not to approve an application in its current form. Each letter is tagged with the deficiency categories it cites and linked to the public PDF; approvals come from Drugs@FDA.

Letters by year

Year	Letters	Publication
2013	12	published once approved
2014	8	published once approved
2015	9	published once approved
2016	20	published once approved
2017	25	published once approved
2018	33	published once approved
2019	37	published once approved
2020	36	published once approved
2021	47	published once approved
2022	25	published once approved
2023	30	published once approved
2024	69	published as issued
2025	59	published as issued
2026	32	published as issued

Before 2024 the FDA published a letter only once the product was later approved; from 2024 it publishes letters as issued. The two periods are not comparable as rates, and the workspace keeps them apart wherever an outcome is computed.

Letters published as issued (2024+)

What the 160 letters cite

Category	Letters	Share
Facility inspection	92	58%
Product quality (CMC)	68	42%
Efficacy evidence	35	22%
Safety	14	9%
Clinical pharmacology	20	12%
Device / human factors	8	5%
Nonclinical	23	14%
Accelerated approval	8	5%

Where the 125 original applications stand

Status as of 2026-10-09	Applications	Share
Approved	33	26%
Resubmitted (public source)	9	7%
No public outcome	75	60%
Discontinued (public source)	1	1%
CBER biologic, not in Drugs@FDA	7	6%

Facility citations across consecutive letters

When a letter cited a facility inspection and the application was rejected again, the next letter cited a facility in 56 of 71 pairs (79%), against 19 of 47 (40%) when the earlier letter did not. All eras, consecutive letters of the same application.

Method

• Letter. A complete response letter (CRL) in the SEC Ephemeris corpus: every letter openFDA publishes, matched to its sponsor.
• Application. An NDA or BLA number. Letters naming the same number belong to one application; cycle is a letter's order within it.
• Letter type. Original application or supplement: the two coders' call for 2024+ letters, else the openFDA file's supplement number, else 'supplement' when Drugs@FDA shows the application approved before the letter.
• Era. issued: the application's first letter is dated 2024 or later, when the FDA began publishing letters as issued. after_approval: earlier, when a letter was published only once the product was approved.
• Categories. Taxonomy v1 (FAC, CMC, EFF, SAF, CLP, DEV, NCL, AAP, REG). Two-coder coding for 2024+ letters; section-aware rules over the knowledge graph for the rest (agreement in rule_agreement).
• Outcome. approved: Drugs@FDA lists an original approval of the same application after its first letter. resubmitted / discontinued: a public source says so (2024+ only). not_covered: a CBER biologic (BLA 125xxx) with no Drugs@FDA record. no_public_outcome: none of these, which is not a rejection. supplement: the application's letters are all on supplements.
• Time to approval. Days from the first original-application letter to approval; applications not approved are censored at outcomes_as_of. Kaplan-Meier with log(-log) Greenwood 95% intervals.
• Pair. Two consecutive letters of the same application on different dates.

How far the rule-read categories can be trusted

For letters dated before 2024, categories are read by rules over the EphemerisX knowledge graph. Each build measures those rules against the two-coder coding of the 2024+ letters:

Category	Coded letters	Recall	Precision
Facility inspection	93	0.96	1.0
Product quality (CMC)	68	0.93	0.8
Efficacy evidence	35	0.69	0.92
Safety	14	0.57	0.73
Clinical pharmacology	20	0.75	0.88
Device / human factors	8	1.0	0.67
Nonclinical	23	0.57	0.68
Accelerated approval	8	1.0	0.67

Filter by category source to keep only coded letters. Your filters, comparisons and layout change the view; they never change these definitions, the outcome rules or how missing outcomes are treated. Every export carries the dataset version and the exact view.

Sources

openFDA complete response letters · Drugs@FDA · SEC EDGAR. The atlas stores only public URLs, sentence locators and short quotes, never the documents.