The first time I watched a CRO hand over a SEND package, I assumed it was a formality — convert the tox study, ship the .xpt files, move on. Then FDA's loader rejected it. Wrong controlled terminology in the LB domain, a missing variable in DM, and pooled reference intervals that didn't match the study design. Two weeks lost on a Phase 1 IND that was already on a tight clock.
If you're at a Series A-C biotech preparing your first IND, SEND is the kind of thing that looks like a checkbox until it isn't. Here's what it actually is, when it bites, and how to avoid the errors I see most often.
What SEND is, and why FDA wants it
SEND stands for Standard for Exchange of Nonclinical Data. It's the CDISC implementation of SDTM for animal studies — a fixed structure that tells FDA reviewers exactly where to find body weights, clinical observations, lab results, and histopathology findings, regardless of which CRO ran the study.
The legal hook is 21 CFR 312.23(a)(8) for INDs and 21 CFR 314.50 for NDAs, layered with FDA's Providing Regulatory Submissions in Electronic Format — Standardized Study Data guidance (originally CDER, December 2014, with periodic Data Standards Catalog updates). The catalog is the document that actually tells you which standards are required, for which study types, starting which date.
The reason FDA cares: reviewers used to get tox data as locked PDF tables. Comparing a sponsor's high-dose group against a historical control range meant retyping numbers into a spreadsheet. SEND lets the reviewer load the dataset into JReview or a similar tool and run the analysis themselves. It's not a courtesy — it's how the modern review division works.
Which studies need SEND
Not every nonclinical study triggers SEND. The ones that do, per the current Data Standards Catalog:
- General toxicology — single-dose and repeat-dose, all species. This is the biggest category and the one most sponsors get wrong first. See single-dose toxicity and repeat-dose toxicity for what FDA expects in the underlying study reports.
- Carcinogenicity studies — full lifetime rodent bioassays.
- Safety pharmacology — cardiovascular, respiratory, and CNS core battery studies. Details on what these studies need to contain are in safety pharmacology.
- Genetic toxicology — Ames, in vitro mammalian, in vivo micronucleus. Background on the standard battery: genotoxicity.
- Reproductive and developmental toxicology — DART studies were added in later catalog versions. Check the catalog for the cutoff date that applies to your study start.
- Toxicokinetics as part of any of the above.
What's not required: pharmacology efficacy studies, ADME studies for non-toxicology purposes, and most exploratory work. But if a study is referenced in your CTD Module 4 as supporting the safety package, assume the reviewer will want it in SEND unless you can point to the catalog and prove otherwise.
Timeline: when does SEND actually apply?
This is the question I get most often, and the answer trips people up.
For NDAs and BLAs: SEND has been mandatory for studies started on or after December 18, 2016 (small molecules) and December 18, 2017 (biologics, with caveats).
For commercial INDs: SEND has been required for studies started on or after December 18, 2017. "Study start" means the first dose of the first animal — not protocol approval, not contract signature.
So if you're filing a Phase 1 IND in 2026 and your pivotal 28-day rat tox study dosed its first animal in 2025, you owe FDA a SEND package. The exemption people sometimes invoke — "it's just a Phase 1, we don't need SEND" — has been wrong for almost a decade. The only legitimate carve-outs are studies that started before the cutoff, or study types not in the catalog.
If you're working with a CRO that says "we'll add SEND later," push back. Generating SEND retroactively from a finalized Pathology Peer Review report is painful, and the costs creep.
How to actually prepare a SEND dataset
SEND datasets are built per the CDISC SEND Implementation Guide (currently SENDIG v3.1.1 for general tox, with separate IGs for DART and genetic tox). The guide defines domains — each domain is one .xpt file with a fixed set of variables.
The domains you'll touch on almost every general tox study:
| Domain | What it holds |
|---|---|
| DM | Demographics — one row per animal: species, sex, age at start, arm assignment |
| EX | Exposure — every dose administered, with date/time, route, dose level |
| BW | Body weights — every measurement, scheduled and unscheduled |
| CL | Clinical observations — cageside and detailed clinical signs |
| LB | Laboratory tests — hematology, clin chem, urinalysis, coag |
| MA | Macroscopic findings — gross pathology at necropsy |
| MI | Microscopic findings — histopathology, with severity grading |
| OM | Organ measurements — organ weights, absolute and relative |
| PC / PP | Pharmacokinetic concentrations and parameters (if TK is in the study) |
| TS | Trial summary — study-level metadata |
| TX | Trial sets — arm definitions and dose groups |
Each domain uses CDISC controlled terminology for test codes, units, severity grades, and finding names. "Alanine aminotransferase" must be ALT, not SGPT. Severity for histopath has a fixed scale (MINIMAL, MILD, MODERATE, MARKED, SEVERE) — your pathologist's "1+ to 4+" scale gets mapped, not preserved.
The package ships with a define.xml file (machine-readable metadata describing every variable in every dataset) and a Study Data Reviewer's Guide (nSDRG) — a PDF that tells the reviewer about anything unusual: pooling decisions, missing data conventions, deviations from the IG. Skipping the nSDRG is one of the fastest ways to earn an information request.
For the broader Module 4 context — how SEND fits inside the rest of your nonclinical package — see the nonclinical overview and pharmacokinetics.
The errors FDA actually flags
I've watched a lot of SEND packages get loaded, rejected, and reloaded. The same five problems show up over and over.
1. Missing required variables in DM or TS. The Trial Summary domain has a list of required parameters — SPECIES, STRAIN, SEXPOP, AGE, TCNTRL, STSTDTC. Skip one and the loader throws an error before it even reads the study data. DM gets it too: animals without a SUBJID, or with ARMCD that doesn't match anything in TX.
2. Wrong domain structure. Putting clinical signs in MA instead of CL. Putting organ weights in MI instead of OM. The IG is explicit about which finding goes where, but pathology reports don't always map cleanly, and CROs sometimes guess. The FDA loader doesn't guess — it rejects.
3. Incorrect pooled reference intervals. This is the subtle one. If you report a control range, FDA wants to know: is this concurrent control from this study, or historical control from the testing facility? The variables LBSTRESN, LBSTNRLO, LBSTNRHI need to be consistent with how the range was derived, and the nSDRG needs to explain it. Mixing concurrent and historical without flagging it is a finding waiting to happen.
4. Controlled terminology drift. Using lab test codes that aren't in the current CDISC CT release. The CT package updates quarterly. If your CRO built the dataset against a 2023 CT release and you're submitting in 2026, you'll get warnings — sometimes hard errors — for terms that have been retired or renamed.
5. define.xml that doesn't match the data. A variable described in define.xml that doesn't exist in the .xpt, or a value-level metadata block that points to the wrong domain. Reviewers run automated checks that compare define.xml to the actual datasets, and any mismatch generates a Pinnacle 21 finding. Pinnacle 21 reports are part of how FDA evaluates SEND quality — get the report from your CRO before you submit and read it line by line.
What to ask your CRO before the study starts
A few things I now insist on, contractually, before the first animal gets dosed:
- The CRO commits to a specific SENDIG version and CDISC CT release.
- The SEND deliverable includes the .xpt files, define.xml, nSDRG, and a clean Pinnacle 21 Community report.
- The sponsor (you) gets a draft SEND package within 30 days of the final study report, not bundled with it.
- Any deviations from the IG — pooling, special domains, custom terminology — are documented in the nSDRG with a rationale.
That last one matters more than people think. FDA is reasonable about deviations as long as you explain them. It's silent deviations that get flagged.
FAQ
Do I need SEND for a pre-IND meeting? No. Pre-IND briefing books don't require SEND datasets. But if your pivotal tox studies are already running, the SEND obligation kicks in for the IND itself, so plan early.
What if my study started before the SEND cutoff date? You can submit the legacy report without SEND, but you should state this clearly in your nSDRG or cover letter and reference the catalog cutoff. Don't leave the reviewer guessing why a package is missing.
Can I generate SEND in-house instead of using my CRO?
Yes, and some sponsors do, especially if they switch CROs mid-program. Tools exist (PointCross, Instem, in-house Python pipelines built on xport). But you still need source data, audit trails, and a clear chain from raw collection to final dataset. It's rarely cheaper than letting the CRO that ran the study build the package.
What's the difference between SEND and SDTM? SDTM is the broader CDISC standard for clinical trial data. SEND is the nonclinical implementation — same structural philosophy (domains, controlled terminology, define.xml), different domains and IGs. If you're hiring someone with SDTM experience to build SEND, make sure they've actually worked with SENDIG, not just clinical SDTM.
How does FDA actually use the SEND data once it's loaded? Reviewers in CDER's Office of New Drugs use tools like JReview to query across animals, treatment groups, and findings. They can compare your high-dose group to historical control databases the agency maintains. They can also flag inconsistencies between your written study report and the underlying data — which is why the report and the SEND package need to tell the same story.
What's a Pinnacle 21 report and do I need one? Pinnacle 21 (Community is free, Enterprise is paid) is the de facto validation tool for SEND and SDTM datasets. FDA runs equivalent checks on its end. You should get a Pinnacle 21 report from your CRO before submission and resolve every error and as many warnings as practical. A clean report doesn't guarantee acceptance, but a dirty one almost guarantees an information request.
If you're building the nonclinical section of your IND and want a second pair of eyes on whether your SEND package and your Module 4 narrative actually match, that's exactly what we built Regfo to check. The rules engine reads your study reports against the same standards FDA reviewers use, and tells you where the gaps are before the loader does.