What actually repeats across FDA CRLs, pulled from 291 real letters and 5,567 extracted readiness checks.
This is the operational layer most teams miss before filing: safety updates, name reviews, facility readiness, translations, and cross-module consistency checks that should have been closed before the submission clock started.
291
unapproved CRLs analyzed
1,748
structured deficiency rows
5,567
line-item checklist actions
2,802
unique requirement statements
The package is layered on purpose: leadership gets the executive brief, operators get the ranked Top 500, and QA or RA teams still have the full evidence exports behind both.
77%
of the deficiency rows are major or critical
The package is designed to be shared in sequence.
Start with the web report, move to the ranked Top 500 for operational ownership, and keep the full raw files for audit trail and evidence review.
Most repeated requirement
Provide English translations of current approved foreign labeling not previously submitted.
Curated operating file
The ranked checklist condenses 2,802 unique extracted statements into the 500 most recurring pre-submission checks.
Report package
Three CSV exports, all tied to the same 291-letter source dataset.
Top 500 checklist
500
ranked checks
The curated operational file: ranked recurring requirements with owner, module, regulation, pathway coverage, and representative evidence.
Full checklist CSV
5,567
action rows
The complete extracted workload from the letters, kept intact for QA, RA, and submission operations teams that need every line item.
Full deficiency analysis
1,748
deficiency rows
One row per deficiency with cited regulation, affected module, root-cause tags, and recommended response action wording.
Safety update findings
Grouped from exact finding titles that explicitly reference safety updates in the 291-letter dataset.
Proprietary name findings
Grouped from exact finding titles that explicitly reference proprietary name work or resubmission.
Manufacturing facility findings
Grouped from exact finding titles that explicitly reference manufacturing facilities or facility inspections.
Foreign labeling translation findings
Grouped from exact finding titles that explicitly reference foreign labeling translations.
Pathway Breakdown
The risk profile shifts hard by submission path
These three pathways cover 287 of the 291 CRLs in the set. The missing pattern across teams is not just what FDA cites, but where in the application the work piles up.
Standard NDAs still fail first on clinical packaging discipline, not just on new science.
Clinical, labeling, and data integrity dominate. The package is usually close, but the summary and safety work is not yet submission-grade.
Average deficiency load
5.5
finding rows per CRL
Average checklist load
18.5
extracted action lines per CRL
Top issue clusters
Primary module pressure
Comprehensive Safety Update Required (31), Proprietary Name Resubmission Required (27), Missing English Translations of Foreign Labeling (15)
Deficiency Landscape
Where the letters actually concentrate
Clinical still dominates the volume, but the operational spread across labeling, admin, quality systems, and manufacturing is what makes the remediation burden heavier than most teams plan for.
Clinical
753The dominant category. Most of the volume is documentation discipline, safety packaging, and reporting completeness rather than failed efficacy claims.
Labeling
488More pervasive than most teams expect. Prescribing information, carton/container text, translations, and proprietary name loops keep recurring.
Data Integrity
342Cross-module mismatches remain one of the most expensive ways to lose credibility late in review.
Administrative
338Name resubmissions, procedural omissions, and response mechanics still show up as avoidable blockers.
Manufacturing
299Facility issues and process readiness remain central, particularly in 505(b)(2) applications.
Quality Systems
294When the quality system enters the letter, the remediation burden usually expands quickly beyond one isolated finding.
CMC
266Control strategy, specifications, stability logic, and justification gaps still account for a large share of remediation work.
Statistical
221Endpoint, pre-specification, and interpretation issues keep showing up in letters that had already consumed major study budgets.
Six questions to ask before you file
If any answer is no or unclear, the dataset says you already have a documented CRL risk factor.
Is the safety update fully current as of the actual submission date?
This is still the most repeated grouped trigger in the analysis. It gets delayed because everyone assumes someone else is closing it.
Has proprietary name work been cleared early enough to stop a preventable hold?
Name review timelines are slower than most submission plans assume, and sponsors keep treating them like late-stage admin.
Are all manufacturing facilities inspection-ready and already through their open objections?
If the site is not ready, the letter is often already written. Document quality does not offset unresolved inspection risk.
Do Module 2 summaries match the underlying Module 5 and Module 4 datasets exactly?
Cross-module drift is one of the fastest ways to create avoidable data integrity work after submission.
Does every foreign labeling artifact in the package have an English translation and named owner?
Translation gaps are low-complexity misses that still recur because nobody owns them explicitly enough.
If this is a 505(b)(2), has M3 and CMC been stress-tested as hard as the clinical bridge?
This pathway still gets treated as lighter than it is. The letters say the opposite once manufacturing is in scope.
Severity distribution
The letters are not mostly advisory cleanup.
Most cited regulations
Safety and labeling rules still dominate the citation surface.
21 CFR 314.50(d)(5)(vi)(b)
Safety update requirements still dominate the remediation load in the unapproved set.
21 CFR 314.50(l)(1)(i)
Labeling and content format requirements continue to appear across both NDA and BLA letters.
21 CFR 201.56(a)
General labeling rules keep surfacing in letters where teams expected the science to be the only risk.
21 CFR 201.56(d)
The structure and content standard itself remains a recurring cleanup point.
21 CFR 201.57
PI content rules remain consistently visible across repeated labeling comments.
21 CFR 601.14(b)
BLA labeling requirements still matter enough to drive recurring resubmission work.
Activity by year
The 2024-2025 slice is where the extracted workload explodes.
2018
18 letters
Findings
129
Checklist Actions
398
2019
20 letters
Findings
128
Checklist Actions
343
2020
16 letters
Findings
98
Checklist Actions
352
2021
16 letters
Findings
104
Checklist Actions
353
2022
27 letters
Findings
144
Checklist Actions
549
2023
26 letters
Findings
170
Checklist Actions
494
2024
93 letters
Findings
546
Checklist Actions
1,757
2025
75 letters
Findings
429
Checklist Actions
1,321
Checklist Preview
What the ranked 500-check file looks like
This preview comes from the curated Top 500 export, which ranks grouped requirement statements by distinct letter recurrence, total mentions, and dominant severity before surfacing the operational owner and evidence.
Front of the Top 500 checklist
The curated file is meant to be worked from directly. The raw checklist CSV remains available behind it when a team needs every extracted requirement line instead of the ranked shortlist.
| Rank | Requirement | Owner | Letters | Mentions |
|---|---|---|---|---|
| 1 | MajorM1 Provide English translations of current approved foreign labeling not previously submitted. | Regulatory / Medical Writing | 176 | 176 |
| 2 | CriticalM5 Include a safety update as described at 21 CFR 314.50(d)(5)(vi)(b) when responding to deficiencies. | Clinical Safety / Pharmacovigilance | 98 | 98 |
| 3 | MajorM5 Provide updated exposure information for the clinical studies/trials (e.g., number of subjects, person time). | Clinical / Biostatistics | 89 | 89 |
| 4 | MajorM5 Describe any information that suggests a substantial change in the incidence of common, but less serious, adverse events between the new data and the original application data. | Clinical Safety / Pharmacovigilance | 72 | 72 |
| 5 | MajorM5 For indications other than the proposed indication, provide separate tables for the frequencies of adverse events occurring in clinical trials. | Clinical Safety / Pharmacovigilance | 71 | 71 |
| 6 | CriticalM5 Present new safety data from studies/clinical trials for the proposed indication using the same format as in the original submission. | Clinical / Biostatistics | 70 | 70 |
| 7 | MajorM5 Include tables that compare frequencies of adverse events in the original application with the retabulated frequencies. | Clinical Safety / Pharmacovigilance | 66 | 66 |
| 8 | CriticalM5 Provide narrative summaries for serious adverse events. | Clinical Safety / Pharmacovigilance | 63 | 63 |
Methodology and trust
The numbers on this page and the numbers in the downloads come from the same generated summary file.
Scope
291 unapproved CRLs from 2018-2025, all generated from the repository-backed source set.
Source set
data-letters/letters/unapproved_CRLs
Extraction fields
Requirements and deficiencies are pulled from structured JSON fields, not hand-entered page copy.
JSON paths
metadata_overlay.summary.key_requirements_detail / findings.items
Curated ranking
The ranked checklist is prioritized by recurrence first, then total mentions, then severity and operating usefulness.
Rank order
Distinct letter recurrence -> Total extracted mentions -> Dominant severity -> Requirement type priority -> Requirement text
Evidence retained
5,567 full checklist rows and 1,748 deficiency rows remain downloadable alongside the curated file.
Export files
fda-crl-pre-submission-checklist.csv / fda-crl-full-analysis.csv
How to use the package
The assets are meant to move from narrative to action to evidence.
Start with the web report
Use the web report when the audience needs the pattern, not the raw spreadsheet.
Work from the ranked Top 500
The curated checklist is the operational bridge between leadership summary and submission execution.
Keep the raw exports as evidence
The full CSVs stay available so reviewers can trace every claim back to the extracted dataset.