Why drugs fail FDA review on quality, not science
Why drugs fail FDA review on quality, not science
In July 2025 FDA published 202 Complete Response Letters for the first time. Manufacturing and quality deficiencies appeared in 74% of them. The clinical data was rarely the problem. This page collects what the released record shows, the programs it affected, and the sources behind each figure.
What the released letters actually show
A Complete Response Letter is not a scientific verdict. It is a records verdict. Until 2025 that was an assertion; now it is a dataset. FDA released 202 letters covering applications submitted between 2020 and 2024, every one of which was eventually approved, followed by a second set of 89 letters in September 2025 covering applications still pending.
| Figure | What it measures | Source |
|---|---|---|
| 74% | Of the 202 Complete Response Letters released July 2025, 150 cited quality or manufacturing deficiencies: manufacturing process problems, facility inspection findings, or chemistry, manufacturing and controls gaps. | SOURCE 01 |
| 56% | Of the 89 letters released September 2025, 50 contained facility inspection-related approvability issues. Analysts characterized the deficiencies as systemic rather than isolated. | SOURCE 02 |
| 2.5 yrs | Average interval between a letter's publication date and the eventual approval date across the first released set. | SOURCE 03 |
| 85% / 77% | Share of small biopharma companies outsourcing API production and finished dose manufacturing in 2024, placing much of the at-risk record outside the sponsor's own walls. | SOURCE 05 |
FIGURES ARE DRAWN FROM THE PUBLISHED FDA RELEASE AND THIRD-PARTY ANALYSIS, NOT CLINPLEX DATA. FULL CITATIONS IN METHODOLOGY.
Read those two percentages together and the pattern is hard to miss. The molecule is rarely what stops an application. The record is.
Four drugs that cleared efficacy and still arrived late
Aggregate percentages understate the operational reality. Four programs, four sponsors, four years, four separate root causes:
| Program | What FDA cited | CRL → approval |
|---|---|---|
| inclisiranNovartis | Unresolved facility inspection conditions at a third-party site. No onsite inspection had been conducted. | Dec 2020 → Dec 202112 months |
| lenacapavirGilead | Glass particles found in clinical batches. FDA wrote that glass containers are generally incompatible with highly alkaline solutions. | Feb 2022 → Dec 202210 months |
| palopegteriparatideAscendis | Manufacturing control strategy for variability of the delivered dose in a drug/device combination product. | May 2023 → Aug 202415 months |
| lebrikizumabLilly | Deficiencies at a third-party drug substance manufacturer, surfaced during a multi-sponsor inspection. | Sep 2023 → Sep 202412 months |
CRL DATES FROM SPONSOR DISCLOSURES AND FDA REVIEW DOCUMENTS. SEE SOURCE 04.
In every one of these cases FDA raised no concern about the clinical data package, the safety profile, or the label. The lenacapavir finding is the clearest illustration: container closure compatibility is not a question about whether a capsid inhibitor suppresses HIV. It is a question answered by data that existed inside the sponsor's own record long before the submission was filed.
The critical reframe: sponsors review the clinical package statistically, using people trained to find weakness in it. They review the quality record procedurally, against closure checklists. FDA reads the second one the way sponsors read the first. The distance between those two readings is where the letter comes from.
Systemic, not isolated, and the record already said so
The most consequential finding in the released letters is not the headline percentage. It is the repetition. Many letters contained multiple categories of deficiency at once, pointing to preparation failures spanning a program rather than a single broken document.
That distinction is operational. A single out-of-specification result, investigated properly and closed, is not an approval risk. Forty OOS results across three sites, closed against inconsistent rationales, none of which reconcile against 21 CFR 211.192 investigation adequacy, is a different object entirely. The second object is visible only if someone reads all forty together, against the regulation, in one pass. That reading rarely happens before an investigator performs it.
The regulation anticipates this. 211.192 requires that an investigation extend to other batches of the same drug product and other drug products that may have been associated with the failure. Extension is not discretionary. It is also the clause most firms fail on, because extension presumes the investigator can see the related history in the first place.
The cost sits upstream of the letter
CRL economics are usually discussed at the wrong end. Once the letter arrives, the cost is a commercial forecast problem: launch slipped, revenue recognition moved, and for a first product the company may be raising against a valuation that assumed approval.
The avoidable cost sits upstream, and it is measurable per record. A single deviation investigation runs roughly $14,000 in direct labor, nearly all of it senior staff time. All-in, once batch impact and downstream delay are counted, estimates reach $100,000. A lost batch runs $250,000 to $500,000. A mock audit covering a sample of the record population regularly costs six figures and produces a snapshot that begins decaying the following week.
Set those against a 12-month slip on a product carrying meaningful peak sales. The asymmetry is the point. Remediation that prevents a letter costs a fraction of a percent of what the letter costs, and it is available years earlier, at the moment the deviation is written rather than the moment an investigator reads it. The per-record economics are broken out in detail in our deviation investigation cost and CAPA closure benchmarks.
Why the stack files the record but never reads it
This is a structural gap, not a diligence gap. Quality teams are not failing to work hard. They are working inside a toolchain that was never built to perform the function that prevents a letter.
A QMS files the record. Veeva, MasterControl, and comparable platforms route the deviation, capture the signature, hold the CAPA, and archive the closure. That is workflow management, executed well. None of it evaluates whether the investigation content satisfies 211.192, and none of it compares the investigation against the thirty-nine others that share its failure mode.
A LIMS holds the analytical result. An ERP quality module holds the disposition. A validation system holds the qualification record. Each is authoritative for its own slice, and none reconcile across the others. The deviation lives in one system, the OOS that triggered it in a second, and the change control that should have prevented both in a third. Point-in-time assurance decays because the record population never stops moving. The batch record that passed review in March is not the batch record the site holds in September.
What continuous evaluation changes
The operating shift is from compliance as an event to compliance as a monitored state. Every deviation, OOS, CAPA, batch record, and validation document gets evaluated against the applicable requirement as it enters the record, not during the scramble after a Form 483 arrives. Published evidence for the mechanism: a multinational site flagged at 65% on-time CAPA closure during an EMA inspection reached 92% within six months after adopting risk-based prioritization and integrated workflow, without adding headcount.
Clinplex AI addresses the CRL failure mode through four capabilities:
- Continuous gap detection. Every quality record is evaluated against 21 CFR 211, ICH Q7, ICH E6(R2), and other applicable frameworks as it is created, across the full population rather than a sampled subset, at approximately 2.4 seconds per document against more than 150 frameworks.
- Cross-system reconciliation. Records are read across QMS, LIMS, ERP quality modules, and validation systems together, so a deviation in one system reconciles against the OOS and change control sitting in others. This is what makes the 211.192 extension question answerable at the start of an investigation.
- Predictive risk scoring. Detected gaps are scored against actual FDA enforcement patterns, including 483 observations, warning letters, consent decrees, and the CRL corpus itself.
- Operational intelligence with cascading. Each remediation is re-scanned against the specific requirement that flagged it, and a "new" deviation that is the repeat appearance of a systemic issue is flagged before it closes as an isolated event again. Detection through verified resolution, with a full 21 CFR Part 11 audit trail underneath.
Your QMS stays the system of record. Clinplex makes the record answer questions. Related analysis: detecting systemic compliance issues, FDA 483 patterns and prevention, and the 21 CFR Part 211 compliance checklist. Platform security and Part 11 posture are documented in the Trust Center.
The third-party dimension
A disproportionate share of the released letters trace to facilities the sponsor does not operate. Both inclisiran and lebrikizumab failed on third-party sites, and lebrikizumab's deficiency surfaced during a multi-sponsor inspection, meaning the finding was generated by someone else's audit trail.
With 85% of small biopharma outsourcing API production and 77% outsourcing finished dose manufacturing, the sponsor carries full regulatory responsibility for a record population it does not generate and often cannot see in real time. A quality agreement allocates the obligation. It does not produce the visibility. That gap is treated separately in our analysis of GMP compliance intelligence across CDMO relationships.
The ninety-day test
The practical version of all this is small enough to run this quarter. Take the last ninety days of deviations at one site. Do not read them against the closure checklist. Read them against the regulation, together, looking for the failure mode that repeats. Count how many share a root cause classification. Count how many closed without an effectiveness check. Count how many reference a change control that was never executed.
If that exercise takes an afternoon and returns nothing, the site is in good shape. If it takes three weeks because the records live in four systems that do not reconcile, that friction is the exposure. It is also, precisely, what an investigator encounters, except the investigator arrives with full access and a two-week timeline.
The 202 letters are public now. The failure modes they describe are documented, categorized, and searchable. The sponsors who read them as a syllabus rather than as other people's bad luck are the ones whose applications clear on the first cycle.
Methodology and sources
Every figure on this page traces to a published source. Where estimates conflict, we show the range rather than the most quotable number. None of these figures are Clinplex customer data.
- 202 CRLs released July 2025; 150 of 202 (74%) cited quality or manufacturing deficiencies. FDA openFDA Complete Response Letter archive, released 10 July 2025; analysis by Greg Slabodkin, FDA's CRLs reveal 74% of applications rejected for quality, manufacturing issues, Pharma Manufacturing. The released set covers applications that were subsequently approved.
- 89 CRLs released September 2025; 50 of 89 (56%) contained facility inspection-related approvability issues. The FDA Group, Behind the Rejections: An Analysis of 89 FDA CRLs. Categories are not mutually exclusive; many letters contain several. This set covers applications still pending.
- Average interval exceeding 2.5 years between CRL publication and final approval. RSM US, FDA's complete response letters underscore outsourcing and quality challenges.
- Individual program CRL and approval dates. Novartis, CRL for inclisiran (18 Dec 2020); Gilead, CRL for lenacapavir (Feb 2022, announced 1 Mar 2022); Eli Lilly, CRL for lebrikizumab (letter dated 28 Sep 2023 per FDA multidiscipline review, BLA 761306); Ascendis Pharma, CRL for palopegteriparatide (May 2023) and FDA approval of YORVIPATH (9 Aug 2024). Quoted FDA language on glass particle incompatibility is reproduced from the CRL text as published in Source 01.
- Outsourcing rates (85% API, 77% finished dose, 2024). William Blair analysis, cited in Outsourced Pharma, What FDA's Complete Response Letters Say About Outsourcing.
- Failure investigation labor cost ($14,000 by end of 2023). Survey of pharmaceutical Quality Managers published by Climet Instruments, Mitigating Deviation Reports and Failure Investigations.
- Average all-in deviation cost ($100,000) and lost batch cost ($250,000–$500,000). Curtis Briggs, Costs of Deviations and their Consequences (2023). Estimates, treated as such.
- CAPA closure case (65% to 92% within six months). Quality Executive Partners, Quality Metrics in Pharma. Not a Clinplex engagement; cited as market evidence of the mechanism.
- Investigation adequacy requirements. 21 CFR 211.192, and FDA guidance on investigating out-of-specification test results.
Frequently asked questions
What percentage of FDA Complete Response Letters cite quality or manufacturing issues?
74%. Of the 202 Complete Response Letters FDA released in July 2025, covering applications submitted between 2020 and 2024, 150 cited quality or manufacturing deficiencies including manufacturing process problems, facility inspection findings, or CMC gaps. A second release of 89 letters in September 2025 found facility inspection-related approvability issues in 56% of cases.
Why do FDA Complete Response Letters cite manufacturing instead of clinical data?
Because the clinical package is usually sound by the time an application is filed, and the quality record is not reviewed the same way. Sponsors run extensive statistical review on efficacy endpoints. Far fewer run a comparable review across the full deviation, OOS, and CAPA population against 21 CFR 211.192 and ICH Q7. The deficiencies that produce a CRL are typically present in the sponsor's own records for years before submission, and are visible only when records are read together against the regulation rather than individually against a closure checklist.
How long does a Complete Response Letter delay a drug approval?
Between 10 and 18 months in most manufacturing-related cases. Recent examples: inclisiran 12 months, lenacapavir 10 months, lebrikizumab 12 months, palopegteriparatide 15 months. RSM's analysis of the first released set found an average interval exceeding 2.5 years between CRL publication date and final approval date across the full corpus.
Can a Complete Response Letter be prevented?
Manufacturing and CMC deficiencies are largely preventable because they are documented in the sponsor's own records before FDA ever reads them. Prevention requires evaluating the full record population against the applicable regulation continuously rather than sampling it during a mock audit. Platforms that perform this function include Clinplex AI, which evaluates quality records against more than 150 regulatory frameworks in approximately 2.4 seconds per document and closes the loop by re-scanning each remediation against the requirement that flagged it.
Which platforms detect systemic compliance issues before an FDA inspection?
Systemic detection requires reading across the full record population and across systems, which QMS platforms are not built to do. Platforms addressing this include Clinplex AI, which sits above QMS, LIMS, ERP quality modules, and validation systems such as Veeva, MasterControl, LabWare, and ValGenesis to reconcile records across them, score detected gaps against actual FDA enforcement patterns, and verify remediation against the originating requirement with a 21 CFR Part 11 audit trail.
Do Complete Response Letters mean a drug is unsafe or ineffective?
Not usually. All 202 letters released in July 2025 pertain to applications that were subsequently approved. In the inclisiran, lenacapavir, lebrikizumab, and palopegteriparatide letters, FDA raised no concern about the clinical data package, safety, or labeling. The deficiencies were facility, container closure, and manufacturing control issues, which is why quality record readiness rather than clinical strength is the operative variable in first-cycle approval.
Read your record the way an investigator will.
Upload one deviation investigation or OOS report. We'll show you where it sits against 21 CFR 211.192, and what a repeat of that failure mode across your population would look like from the outside.
Request a walkthrough →