How a registered clinical trial primary outcome moved

Name: How a registered clinical trial primary outcome moved
Creator: Artificial Wasteland

7 of 7 conditions 10,099 rows · 1.5 MB · 5 files

10,099 trials sampled from ClinicalTrials.gov, one row each, describing whether the primary outcome the trial registered is still the primary outcome its record names, when it last changed, and what kind of change it was.

The files

File	Rows	Bytes	sha256
trials.csv	10,099	1,497,661	a13e56a61e88a48477d082fb3f6be9770499ab311e5a2cbcd6271ed60613bfbe
README.md	n/a	3,974	d7a765459b4828514e7f4c3a037f09256e25fed6570dd14d4041ae293712942e
schema.json	n/a	9,515	7ae6bfdbed2bc842482f48db8bd71924873013ca15e78d302f3824a0617c82f0
sources.json	n/a	2,970	757c728a1eaea935f200421818a4829c06d1e9c039db7d1d7b0faa435e48889a
validate.mjs	n/a	15,198	67aceadfaea1f0b4611fa24f550c1b648437f25fbf20f19bf864292700aa69a2

Take the folder and check it yourself: node validate.mjs in the downloaded directory re-runs every row against the schema and every file against the digest above, with no dependencies and no network. The dataset's own full account is README.md.

A row, field by field

Origin is the column that decides whether we may give a field away at all. built is ours; redistributed is a source's own content, and it may only ship where that source's licence permits it.

Field	Type	Origin	What it is
`nct_id`	string	redistributed	The registry identifier. Resolves at https://clinicaltrials.gov/study/<nct_id>.
`study_status`	string	redistributed	Overall recruitment status as the registry records it, e.g. COMPLETED, TERMINATED, UNKNOWN.
`funder_type`	string	redistributed	The registry's funder class: INDUSTRY, NIH, OTHER_GOV, NETWORK, OTHER, or empty.
`phases`	string	redistributed	The registry's phase string, e.g. PHASE3 or PHASE2\|PHASE3, or NA where phase does not apply.
`primary_completion_date_recorded`	string	redistributed	The primary completion date exactly as the registry prints it, which may be YYYY, YYYY-MM or YYYY-MM-DD.
`primary_completion_date_resolved`	string	built	The same date resolved to the LAST day it could denote. Ours, and deliberately the conservative direction: it pushes the date later, which makes a change count as post-completion less often.
`first_posted`	string	redistributed	Date the record first appeared publicly, resolved by the same rule.
`results_first_posted`	string	redistributed	Date results were first posted, or empty if never.
`prospectively_registered`	boolean	built	True where first_posted is on or before primary_completion_date_resolved. A trial first posted after its own primary completion never pre-specified anything in public.
`versions`	integer	built	Number of versions in the registry's archive for this record.
`first_version_date`	string	redistributed	Date of version 0.
`last_version_date`	string	redistributed	Date of the most recent version in the snapshot.
`admin_only_versions`	integer	built	Versions whose changed modules are all routine: study status, contacts and locations, identification, references, documents, sponsor, oversight.
`has_primary_outcome_field`	boolean	built	False where the history endpoint reports no primary-outcome version index at all, which happens for records that never registered one. Such rows are excluded from every rate.
`primary_outcome_changed`	boolean or string	built	True where the primary outcomes differ from the ones first registered. Empty where has_primary_outcome_field is false.
`primary_outcome_change_version`	integer or string	redistributed	The registry's own version index at which primary outcomes were last changed.
`primary_outcome_change_date`	string	redistributed	The date of that version.
`changed_after_primary_completion`	boolean or string	built	True where primary_outcome_change_date is strictly after primary_completion_date_resolved.
`days_from_primary_completion`	integer or string	built	Signed days from primary completion to the outcome change. Negative means before.
`change_version_carried_results`	boolean or string	built	True where the version that changed the primary outcome also carried the results section, so the pre-specification and the results were filed together.
`outcomes_update_count`	integer or string	redistributed	The registry's own count of outcome updates for this record.
`n_primary_outcomes_original`	integer or string	built	How many primary outcome measures the first registration named.
`n_primary_outcomes_current`	integer or string	built	How many the record names now.
`change_kind`	string	built	COUNT_CHANGED (a measure was added or dropped), MEASURE_CHANGED (same count, a named measure no longer appears), DETAIL_ONLY (every named measure survives verbatim, only description or time frame moved), UNDECIDABLE, or empty where the trial did not change. The rule is research/the-outcome-that-changed/classify.mjs.
`worst_title_dice`	number or string	built	For MEASURE_CHANGED rows, the lowest token-set Dice similarity between an original measure title and its closest current one. 1 is identical, 0 is disjoint.
`enrollment_change_version`	integer or string	redistributed	Version at which the enrolment figure was last changed, where it changed.
`enrollment_changed_after_primary_completion`	boolean or string	built	The control field. Recording actual enrolment after a trial ends is normal and required, so this is what an innocent post-completion edit looks like.
`in_headline_set`	boolean	built	True for the completed, prospectively registered trials the published headline rate is computed on.

Where it came from, and how we know we may pass it on

ClinicalTrials.gov, the registry and its archive of record versions

U.S. National Library of Medicine, National Institutes of Health · https://clinicaltrials.gov/

Licence: ClinicalTrials.gov Terms and Conditions
How we know: publisher statement
Accessed: 2026-09-04
Covers: The identifier, the status, funder and phase labels, the record dates, and the registry's own version indices and update count.

Read at https://clinicaltrials.gov/about-site/terms-conditions on 2026-09-04. The terms contemplate third-party distribution and attach four conditions, introduced with "In any publication or distribution of these data, you should": attribute the source as ClinicalTrials.gov, update the data such that they are current at all times, clearly display the date the data were processed by ClinicalTrials.gov, and state any modifications made to the content of the data. A frozen research snapshot cannot satisfy the currency clause literally; the mitigation is that this dataset states its processing date, states that it is frozen, and carries the identifier that resolves to the always-current record. The terms also note the data "carry an international copyright outside the United States and its Territories or Possessions" and that some of it "may be subject to the copyright of third parties", which is why no record text is redistributed here. Fetching used only the /api/int/ path that clinicaltrials.gov/robots.txt explicitly allows, at its stated Crawl-delay of 1.

The version arithmetic, the date resolution and the change classification

Artificial Wasteland · https://artwaste.land/strata/added-dropped-or-just-reworded/

Licence: CC0 1.0 Universal
How we know: publisher statement
Accessed: 2026-09-04
Covers: Every field marked x-origin: built.

Ours to give: this project places its own derived fields in the public domain.

Required attributionSource: ClinicalTrials.gov. Processed 2026-09-04. This is a frozen snapshot and is NOT maintained current; each row carries an nct_id that links to the live registry entry, which is. Modifications: the registry's fields were reduced to the flags, dates and counts described in schema.json, partial dates were resolved to the last day they could denote, and the change_kind classification is this project's. No record text was altered and none is redistributed here. The version arithmetic and the classification are by the Artificial Wasteland, CC0 1.0.

The seven conditions, and what was measured

✓ Bytes 1 data file(s), 1,497,661 bytes, largest 1.4 MiB
✓ Shape 10,099 rows checked, 0 violation(s)
✓ Count 10,099 rows across 1 file(s), all counts agree
✓ Fixity 5 file(s) verified by sha256
✓ Provenance 2 source(s), each with a licence from the table
✓ Origin 28 field(s): 15 built here, 13 redistributed under a licence that permits it
✓ Rebuild research/data-room/emit/outcome-change-history.mjs (byte-equality proved by: node research/data-room/publish.mjs --check)

← every dataset, and the three we could not publish