Ask any first-time IND sponsor what they're scared of, and they'll say the tox package. The two-species study. The NOAEL math. The genotox battery.
Now look at what FDA actually puts INDs on hold for.
CMC. Then clinical. Then nonclinical, in third place.
That's not me guessing. That's FDA's own published analysis of 1,410 initial CDER INDs from FY2013, run by Getz and colleagues and published in 2016. The hold rate was 8.9%. CMC was the number one trigger. Toxicology came third (PubMed 26911627).
This matters because most preclinical teams plan against the wrong threat. They build a beautiful tox program and then get held up by stability data, an impurity spec, or an inconsistency in their dosing rationale that the chemistry reviewer caught in week three.
I'm not telling you to deprioritize nonclinical. I'm telling you to read the data before you decide where the risk actually lives.
What FDA actually publishes about hold rates
There isn't much, which is part of the problem. The most cited cross-therapeutic dataset is still Getz 2016, looking at FY2013. Nothing equivalent has been published for 2020–2024.
Here's what we know:
- 8.9% of initial CDER INDs went on clinical hold in FY2013 (n=1,410). More than half were resolved within 12 months (Getz et al., 2016).
- Under 10% of oncology INDs were placed on hold between 2014 and 2017. The deficiency ranking was clinical first, CMC second, nonclinical third (Bhatt et al., 2019).
- Around 20% of CGT INDs went on hold between 2021 and 2023, more than double the cross-therapeutic average (Liao et al., 2023).
- CDER processed 1,855 new IND receipts in 2024 against 14,870 active INDs. That's the denominator most people forget.
The Getz paper has one line that I keep coming back to:
"The vast majority of initial applications proceeded without clinical holds. Among the 8.9% placed on hold, most achieved active status within one year, suggesting many holds could be prevented through adherence to existing guidelines."
Translation: most holds aren't surprises. They're known requirements that someone didn't check.
Where nonclinical gaps actually do matter
Third place isn't no place. When tox does cause a hold, it tends to cluster in the same handful of categories. Bhatt 2019 named them, and they line up with what we see in our own rules engine.
1. Inadequate species justification. ICH M3(R2) wants two species, one rodent and one non-rodent, unless you can scientifically justify one. "We only had budget for rats" is not a justification. FDA has held INDs over this for years. If you're going single-species, your nonclinical overview needs the relevance argument front and center, with PK and receptor data backing it. See the nonclinical overview module for what reviewers expect there.
2. Genotoxicity battery gaps. ICH S2(R1) asks for a standard battery: bacterial reverse mutation, in vitro mammalian assay, in vivo assay. Teams skip the in vivo piece, or run it non-GLP, or use a pilot study to substitute. None of those work for an IND. Full requirements are in the genotoxicity section of the library.
3. Safety pharmacology under-scoped. ICH S7A wants the core battery — cardiovascular, respiratory, CNS. The hERG assay specifically needs to be GLP-compliant for IND. Non-GLP hERG is one of the cleanest, most preventable hold triggers I've seen. The safety pharmacology section walks through the full battery.
4. Dose justification math. This is where the Nonclinical-Clinical Interface 2021 study (n=58) gets uncomfortable. 46% of FIH trials had exposure margins more than 100× below NOAEL — wildly conservative. 25% exceeded NOAEL at the highest planned clinical dose. Both extremes get caught: one because the sponsor is leaving enormous clinical headroom unexplained, the other because the reviewer sees a patient safety problem before the protocol runs. The math has to be defensible in both directions, and it usually isn't when someone treats the NOAEL calculation as a formality rather than the actual argument it is.
5. PK bridging that doesn't bridge. If your tox species PK doesn't predict human exposure adequately, your safety margins are paper. The pharmacokinetics module covers what reviewers expect to see linking the two.
6. Repeat-dose duration mismatch. ICH M3(R2) tables tell you the minimum repeat-dose duration based on planned clinical dosing. Sponsors routinely propose Phase 1 multiple-dose regimens supported by single-dose tox. That's a mismatch the reviewer will catch in week one.
None of these require new science. They require somebody whose job is to read the requirement and check whether the study actually meets it.
GLP enforcement: what FDA warning letters reveal
If hold data tells you what reviewers reject, GLP warning letters tell you what inspectors see when they walk into a facility. We pulled 17 letters citing 21 CFR Part 58 violations between 2019 and 2026. The pattern is consistent and ugly.
| Violation | % of letters | CFR |
|---|---|---|
| Study Director failures | 88% | 21 CFR 58.33 |
| Quality Assurance Unit failures | 82% | 21 CFR 58.35 |
| SOPs missing or inadequate | 59% | 21 CFR 58.81 |
| Data integrity / falsification | 53% | 21 CFR 58.130 |
| Final report deficiencies | 47% | 21 CFR 58.185 |
Study Director and QAU failures co-occur in over 80% of letters. That's not a coincidence — it's the structural core of GLP. When the Study Director isn't actually directing and the QAU isn't actually auditing, every other Part 58 requirement starts to slip.
Two trends in the recent letters worry me more than the totals.
Foreign labs dominate 2024–2026 enforcement. Five of the seven most recent letters went to facilities in China or India. Mid-Link Testing (China, 2024) had animal weights so identical across animals that they were physically impossible. CCIC Huatongwei (China, 2025) had staff who couldn't explain their own GLP responsibilities. Jiangsu Kerbio (China, 2025) had inspectors directly observe live data fabrication.
Final report fraud is escalating. Vedic Lifesciences (India, March 2026) was cited for falsifying study director names in final reports — hiding which facility actually ran the work. When that happens, FDA doesn't just reject the study. They refuse all data from the facility.
If your CRO is offshore and you haven't audited them in person in the last 18 months, you're carrying a risk that doesn't show up in any tracker until it's a 483.
The CGT problem
Cell and gene therapy is the one place where the "tox is third" rule breaks down. Sort of.
Liao 2023 looked at CGT INDs from 2021–2023. The hold rate was around 20%. Of the 33 hold deficiencies they categorized:
- 70% were clinical (mostly safety reporting and adverse event handling)
- 21% were CMC
- 9% were preclinical
So preclinical is still third. But the headline statistic is harder to ignore:
"NIH-listed CGT clinical trials total less than 2% of all listed clinical trials, yet they are responsible for approximately 40% of clinical holds." — Liao et al. (2023)
Less than 2% of trials, around 40% of all FDA holds. CGT is a different risk class entirely. CMC holds in this category take an average of 8.4 months to resolve, the longest of any IND type FDA tracks.
If you're running a CGT program, the cross-therapeutic 8.9% hold rate isn't your number. Your number is 20%, and your CMC team needs to be staffed accordingly.
What this means for first-time IND sponsors
Bhatt 2019 said it directly:
"INDs placed on hold were mostly for first-in-human trials or submitted by sponsors with limited regulatory experience."
FIH plus first-time sponsor is the highest-risk profile FDA tracks. That's not a soft observation. That's the agency's own analysis.
A few things follow from that.
First, the Phase 1 cliff is real. Phase 1 success rates dropped from over 75% (2006–2008) to under 40% in recent Citeline/Norstella tracking. Phase 1 LOA is now 6.7%. Translation: by the time you submit your IND, you've already lost most of your starting cohort to preclinical attrition. The IND review is the last filter before you spend real clinical money.
Second, "we'll fix it in the response" is a bad plan. Half of held INDs do resolve within 12 months. That's also a 6-month average runway burn for a Series A biotech with one program.
Third, the gaps aren't exotic. Almost everything that gets caught is named in an existing ICH or FDA guidance document. Getz said it in 2016 and it's still true: most holds are existing requirements that nobody checked.
That last point is the entire reason we built Regfo.
Where we fit
Our preclinical compliance checker reads your study reports and scores them against 24 ICH and FDA guidelines, broken into 1,054 individual rules. Species justification, GLP status of every required study, hERG conditions, genotox battery completeness, dose margin math, repeat-dose duration against planned clinical regimen — all of it.
It's not a replacement for a regulatory consultant. It's the thing that runs before you call one, so the conversation starts with "here are the three gaps" instead of "let's spend two weeks reading PDFs."
If you want to see what a clean check looks like for your own studies, try it on one report. You don't need to upload your whole package.
FAQ
What percentage of IND submissions actually get clinical holds? About 8.9% of initial CDER INDs in FY2013 — that's the most recent cross-therapeutic dataset we have, from Getz et al. (2016). Oncology runs under 10%. CGT runs around 20%, which is a whole different conversation.
Are preclinical gaps the most common reason for IND clinical holds? No, and that's the whole point of this article. CMC is first, clinical is second, nonclinical is third. I've seen this pattern hold across the FY2013 cross-therapeutic data and the 2014–2017 oncology data. The only exception is CGT, where clinical issues drive 70% of holds — but even there, preclinical isn't the lead problem.
Which preclinical gaps cause the most holds when they do happen? The ones I see come up repeatedly: species justification that doesn't hold up, an incomplete genotox battery (usually the in vivo piece), safety pharmacology gaps — especially a non-GLP hERG study, dose justification at either extreme of NOAEL, PK bridging that doesn't actually bridge, and repeat-dose duration that doesn't match the planned clinical regimen. None of these are obscure. They're all in the guidelines.
What's the most common GLP violation in FDA warning letters? Study Director failures under 21 CFR 58.33 — cited in 88% of the 17 letters I reviewed. QAU failures under 58.35 follow at 82%. The two almost always show up together, which makes sense: when oversight is broken at the top, everything else follows.
How long does it take to get a clinical hold lifted? More than half of general INDs resolve within 12 months. For CGT, Liao 2023 puts the average at 6.2 months overall, with CMC-driven holds at 8.4 months — the longest category FDA tracks (PMC10597781).
Why does CGT have such a high hold rate? CGT INDs face roughly 20% holds versus 8.9% cross-therapeutic average. The driver isn't preclinical — it's clinical safety reporting and AE handling, which account for 70% of CGT hold issues. If you're running a CGT program and you're focused mainly on the tox package, you're probably looking in the wrong place.
Sources: Getz et al. 2016 (PubMed 26911627), Bhatt et al. 2019 (PubMed 31678263), Liao et al. 2023 (PMC10597781), FDA IND Activity Reports 2024, Nonclinical-Clinical Interface 2021, Norstella Phase 1 LOA tracking. FDA warning letter analysis: 17 letters citing 21 CFR Part 58 violations, 2019–2026.