Published

What three welfare guidance documents define, and what they leave to statistics

Three animal-welfare guidance documents define exactly when a study must end and what is recorded, but none says whether the measurement it prevented still exists.

Current status

This is a source-bounded research note, not a peer-reviewed paper and not a Protocol decision.

Two independent close reviews inspected the frozen report against the same three source documents. Both accepted its direction. They disagreed on four points of documentation, and each disagreement was settled against the sources rather than by counting reviewers. The wider FND-1 research gate remains open, and no animal-welfare or missing-data convention has been selected for nomue Protocol.

The finding

An animal study can be required to stop before its planned measurements are taken. Three widely used welfare guidance documents say precisely when that must happen, what triggers it, what action follows, and what has to be recorded. None of them says what becomes of the measurement that was never taken.

That silence is not a defect to repair here. It marks where one body of guidance ends and statistical reasoning begins. The practical risk is that an analysis reads a recorded welfare decision as though it had already answered the data question, and so adopts an assumption that no source supplies.

What the guidance defines

The investigation read the complete text of three documents, each recorded by file digest, byte size, and page map before reading. All three describe the end of a study as a structure with separable parts: a rule fixed in advance, an event with a time, and a graded action.

DocumentWhat it defines directly

OECD

Guidance Document on the Recognition, Assessment, and Use of Clinical Signs as Humane Endpoints for Experimental Animals Used in Safety Evaluation. Environment Directorate, ENV/JM/MONO(2000)7, Series on Testing and Assessment No. 19.

  • A humane endpoint as the earliest indicator of severe pain, distress, suffering, or impending death (paragraph 9, p. 10).
  • Three graded responses: humane killing, termination of exposure that may be temporary, or dose reduction (paragraph 10, p. 10; paragraph 62, p. 22).
  • Impending death defined against the observation schedule: a moribund state, or death expected before the next planned observation (paragraph 12, pp. 10–11).
  • Endpoints fixed at planning time in the protocol and standard procedures, observation frequency increased once abnormal signs appear, and the timing of signs documented (paragraph 11, p. 10; paragraph 41, p. 17).
  • The welfare decision bounded by the scientific objective: pain is treated when treatment does not interfere with the study, and a decision to kill weighs the potential loss of valuable data (paragraph 46, pp. 18–19; paragraph 42, p. 18).
  • Separate record classes for an animal found dead and an animal humanely killed (Table 1 note 2, p. 26).
  • Humanely killed animals counted as dosage-dependent deaths (paragraph 54, p. 21) — a counting rule, not a statement about an unmeasured value.

CCAC

CCAC guidelines: Identification of scientific endpoints, humane intervention points, and cumulative endpoints. Canadian Council on Animal Care, March 2022.

  • Scientific endpoints, at which the approved live animal use is complete, kept structurally separate from intervention points, where the required action is not necessarily euthanasia (glossary, pp. 38–39; introduction, p. 4).
  • A graded ladder of interventions: environmental change, supportive care, antibiotics, pain relief, temporary or permanent removal, then euthanasia (p. 4).
  • Anticipated intervention points distinguished from unexpected negative outcomes such as unrelated illness or a life-support systems failure (pp. 4–5).
  • Pilot studies directed at determining welfare-appropriate endpoints rather than at generating usable scientific data (guideline 2, p. 2).

Workman et al.

Guidelines for the welfare and use of animals in cancer research. British Journal of Cancer (2010) 102(11), 1555–1577.

  • Humane endpoints developed alongside the requirements for a valid scientific outcome, with earlier endpoints able to increase precision (p. 1569).
  • Intentional use of death as an endpoint treated as unacceptable (p. 1570).
  • Quantified, domain-specific triggers: for example 20% weight loss at any time, or 15% maintained for 72 hours against pre-treatment or age-matched vehicle controls, and ulceration persisting beyond 48 hours (Box 5, p. 1571).

Across all three, the trigger rule, the event and its time, and the action taken are separately recorded facts. Where the documents agree, the agreement was treated as independent evidence converging on the same scope, never as a vote.

What it leaves open

None of the three defines whether a scheduled measurement that euthanasia or death prevented still conceptually exists, and none defines a statistical mechanism for how such values go missing.

Word-boundary searches over the complete text of all three documents returned no occurrence of MCAR, MAR, or MNAR. The one occurrence of the word censor is in the CCAC text (p. 27), where the guidance reports a cited suggestion that truncated longevity studies can be analyzed with proportional-hazards models on censored data. That is a design option attributed to other literature; it defines neither existence nor a mechanism. The one occurrence of the word missing in the OECD text is in a dental-observation list in an annex (p. 38).

The vocabulary of clinical estimands is absent as well. The terms estimand, ICH, E9, treatment policy, hypothetical, composite, while-on-treatment, and principal stratum returned no hits in any of the three documents. The word intercurrent occurs twice, both in an OECD annex, in the ordinary veterinary sense of a concurrent infection rather than the clinical-trial sense.

How that absence was established is recorded rather than asserted. It rests on complete text extraction, a full reading of every definitional, procedural, endpoint, monitoring, and record section, inspection of page images where the text layer was degraded, and targeted word-boundary searches. It does not rest on a verified line-by-line reading of every sentence. The absence is therefore graded one step below the direct verification used for the definitions above, and the report says so.

What the documents do define next to that gap shows the boundary is real rather than accidental. The event and its time are recorded facts distinct from the observation stream. An animal found dead and an animal humanely killed produce differently sourced final records. An observation lost to a life-support systems failure is classed as an unexpected negative outcome rather than an anticipated intervention point.

A terminology split

The three documents do not share one word for the same idea. The 2022 CCAC guidelines deliberately replace humane endpoint with humane intervention point, because the required action is not necessarily the end of anything. The OECD and Workman documents use humane endpoint throughout.

The investigation preserved that split instead of smoothing it. Any later vocabulary decision has to face the disagreement as the sources state it.

Why this matters

The narrow technical result is that the endpoint, its time, and the action taken are directly evidenced as separable declared facts across three separately issued documents, from at least two institutionally independent issuing bodies, while the existence status of an unmeasured outcome is carried entirely by statistical sources.

The wider point concerns how a verification system should treat such a record. A statement that an animal reached a humane endpoint is a well-defined and well-recorded fact. It is not a statement about whether the measurement that did not happen exists, and treating it as one adopts an assumption that no inspected source provides. This is the kind of fact a Record has to carry explicitly rather than leave an analysis to infer. The finding is research input for that design question. It changes no current capability.

Limits

The conclusions hold for three artifacts at recorded digests, in the specified editions and page maps. They do not extend to other renditions or to guidance that was not inspected.

The originally commissioned 1998 CCAC text could not be obtained, and the 2022 CCAC guidelines were inspected in its place. That substitution rests on the same issuing body and the same stated role, not on a formal notice of supersession. A claim that one document supersedes the other would need an official CCAC record.

Two further leads, Morton (2000) and Toth (2000), were not inspected, and no conclusion here rests on them. Network routes to the publishers and repositories were blocked in the investigating environment, so the three documents were supplied directly, and their digests, byte sizes, and page maps were recorded before any content was read.

This note does not recommend an endpoint, a threshold, a euthanasia rule, or an analysis method. It does not rank the three documents, and it establishes no missing-data assumption for any study.

Relation to nomue Protocol

The finding is research input for future identity and interpretation design. It adds no Record field, vocabulary term, method identifier, schema, public check, supported procedure, or implementation behavior, and it selects no animal-welfare or missing-data convention.

Four of eight recorded FND-1 holds are closed in narrowed form. Four holds remain open, so the complete FND-1 research gate is still open, and adopting any result here would require a separate Protocol proposal and review.

Public evidence

  • Source commission — the bounded question, required sources, and closure conditions set before the work
  • Primary-source result — claim ledger with page anchors, the absence method record, and eight falsification cases
  • First close review — independent inspection of the frozen report against the same documents
  • Second close review — a separate independent inspection reporting four documentation corrections
  • Review variance adjudication — each disputed point settled against the sources, with the corrections carried forward
  • Steward disposition — the accepted narrowing, the preserved terminology split, and the explicit non-decisions