Why an animal might stop taking part in a study

Separate absence from failure

A missing animal in a study’s final dataset doesn’t automatically mean the study “went wrong” for that animal. Attrition is its own category, separate from a negative result or a design flaw.

Reading a study well means asking why a subject didn’t finish before assuming the answer says anything about welfare or effectiveness.

Researchers plan for this ahead of time. The NC3Rs Experimental Design Assistant guidance on inclusion and exclusion describes how investigators are expected to set the conditions, before an experiment starts, for whether an animal stays in the study.

That upfront planning exists precisely because removal happens for identifiable reasons, not as an unexplained gap in the record.

Some of those reasons are procedural. A sample can be damaged during collection. A blood draw can turn out too small to measure reliably. Equipment can fail to record.

None of them say anything about the animal’s behavior, health trajectory, or the intervention being tested.

Other reasons are clinical or humane. The same guidance points to cases like an animal developing a movement disorder that would distort behavioral measurements, or reaching a defined humane endpoint.

Those are welfare-driven removals, built into the protocol as a safeguard. They don’t count as evidence that the study failed.

The distinction matters for anyone reading a study’s results section: a shrinking sample size is a data point about the process. On its own, it says nothing about the animal or the treatment.

What counts as an exclusion?

Exclusion criteria belong on paper before data collection begins. A careful reader looks for that specification directly, since its absence doesn’t mean nothing was excluded. The NC3Rs guidance on inclusion and exclusion gives the reason without dressing it up: decide these before your experiment begins, to prevent unconsciously biasing your results by deciding as the experiment progresses.

The guidance separates two levels of exclusion. One concerns animals or experimental units themselves, for example if an animal develops surgical complications or reaches a humane endpoint and is removed from the study and euthanized.

The other concerns individual data points that remain in an otherwise-continuing animal, like a body weight value that is biologically impossible, or a recording lost because someone forgot to switch on equipment.

Both levels can produce the same visible effect in a results table: a missing entry. But they document different things happening to different units of analysis.

Readers checking a study’s rigor can look for whether it states these criteria explicitly, and whether it accounts for how much attrition it expected. The guidance works a specific example: a surgical intervention carrying a 10 percent attrition rate means adding an extra 10 percent to the calculated sample size at the start.

A study that mentions this kind of planning is telling readers it built dropout into its numbers rather than discovering it as a surprise.

What does “dropout” mean in a study?

The precise definition of “dropout” in a given study is worth checking directly, because the term can describe several different measured events. The NC3Rs page on inclusion and exclusion criteria lists distinct categories that a results section might collapse into a single missing row: a humane endpoint reached, a complication following surgery, a sample judged unusable during processing, or a data value excluded for falling outside a biologically plausible range.

Each of those is a different observed event with its own trigger. A humane endpoint is a threshold defined ahead of time, often tied to a measurable sign like tumor volume.

A damaged histological section is a handling problem discovered during processing. A negative body-weight value is a data-entry or measurement error caught during analysis, unrelated to the animal’s continued participation at all. And outliers can go by rule rather than by judgment, removed with a built-in method in a statistics package, which is why the analysis record matters alongside the animal record.

Conflating these categories risks a specific mistake: treating a data-cleaning exclusion (a typo, a contaminated sample) as though it were evidence about an animal’s condition or response to an intervention. The source distinguishes exclusion of animals or experimental units from exclusion of individual data points, and it points to ARRIVE guidelines item 3b for reporting standards on data cleaning specifically.

Before drawing a conclusion from an attrition figure, it helps to identify which category the study places it in. A study that reports its exclusions by type, rather than as one undifferentiated dropout count, gives readers the material to make that distinction themselves.

Avoid an invented motive

A study record does not tell you what an animal wanted, felt, or decided, so any account of “why” an animal stopped participating should stay within what the criteria actually measured. The NC3Rs guidance on inclusion and exclusion defines exclusion in terms of prespecified conditions, things like a movement disorder affecting measurement, a humane endpoint, or a sample failing quality control.

None of these are descriptions of animal intent or preference. It’s tempting to narrate a dropout in motivational language, as though an animal “gave up” on a task or “chose” not to continue. The cited criteria don’t support that framing.

A behavioral training threshold not met, for example, is listed as an inclusion criterion about eligibility to start a task, not a claim about willingness or effort. Describe the exclusion using the same terms the protocol used to define it.

If a study removed an animal for reaching a humane endpoint, that is the reportable fact. Extending it into a claim about distress, motivation, or welfare beyond what the endpoint measured turns a defined criterion into speculation.

This applies equally to readers summarizing a study for others. Restating an exclusion criterion accurately is straightforward. Reconstructing the animal’s internal experience from it goes beyond what the source material supports, and treats an operational threshold as though it were a window into the animal’s mind.