Skip to content

WHO ICTRP: citing trials that are not on ClinicalTrials.gov

dismech's clinical_trials block grew up around ClinicalTrials.gov. The name slot holds an NCT… identifier, evidence resolves against the ClinicalTrials.gov API, and there are 995 NCT identifiers across the knowledge base. A trial registered anywhere else had nowhere to go.

The WHO International Clinical Trials Registry Platform (ICTRP) is the umbrella over every primary registry — ChiCTR (China), ISRCTN, EUCTR, jRCT/UMIN (Japan), CTRI (India), ANZCTR, IRCT, DRKS, PACTR, and the rest. One prefix (ICTRP:) and one fetch path therefore cover all of them, and ICTRP publishes a normalized record (the 24-element WHO Trial Registration Data Set), so a ChiCTR record and an ISRCTN record cache to the same shape.

Why ICTRP rather than ChiCTR directly

Fetching ChiCTR itself was considered and rejected. chictr.org.cn answers 405 to a plain GET from outside China and publishes no documented API, while the ICTRP record for the same trial is a plain GET away. Supporting one registry at a time would also mean writing a new fetcher and a new prefix for every registry a curator eventually meets. ICTRP is the join point that already exists.

Using it

# Cache a trial record (works for any ICTRP primary registry)
just ictrp-fetch ChiCTR2100045397
just ictrp-fetch ISRCTN67795930 CTRI/2021/05/033585

# Equivalently, through the normal reference-fetch entry point
just fetch-reference ICTRP:ChiCTR2100045397

# Refresh everything already cached, or list it
just ictrp-rebuild
just ictrp-list

Then key the trial on its ICTRP identifier and cite the record:

clinical_trials:
- name: ISRCTN67795930
  phase: PHASE_III
  status: COMPLETED
  description: >-
    GEM3: a multicentre, double-blind, placebo-controlled randomised trial of
    combination methotrexate plus gefitinib versus methotrexate alone.
  evidence:
  - reference: ICTRP:ISRCTN67795930
    supports: SUPPORT
    evidence_source: OTHER
    snippet: "| Register | ISRCTN |"
    explanation: WHO ICTRP registration record establishing the trial's identity.

As with Orphanet and ICEES records, each ## Registration table row is a stable quotable substring, and a quoted snippet may include or omit the leading and trailing pipes. Section prose (eligibility criteria, outcomes, interventions) is quotable too.

evidence_source: OTHER is right for a registry record: it is a registration document, not a report of human, animal, in-vitro, or computational evidence.

What the cached record contains

## Registration (a table: register, main ID, registration and enrolment dates, sponsor, target sample size, recruitment status, study type and design, phase, countries, completion, results availability, source-register URL), ## Titles, then whichever free-text sections the registry populated — health conditions, interventions, key inclusion and exclusion criteria, primary and secondary outcomes, secondary IDs, funding — and a ## Source provenance footer.

Investigator contact details are deliberately dropped. ICTRP records carry names, postal addresses, telephone numbers, and personal email addresses. Those are not evidence, and republishing them into a public git repository is not something a curation cache should do.

No bulk file

Every other structured source pins a bulk download in data/<source>/MANIFEST.yaml. ICTRP does not: its full export is distributed under a separate data-use agreement, so records are fetched per identifier on demand — the same posture as ClinGen's report-page narrative. Consequently just ictrp-rebuild with no --id refreshes what is already cached; a new trial is added by naming it.

Auditing what is still stranded

just ictrp-audit                 # census by registry and placement
just ictrp-audit --format tsv
just ictrp-audit --strict        # exit 1 while any identifier is uncited

Every identifier is classified:

Placement Meaning
CITED cited as ICTRP:<TrialID> with a cache file backing it — snippet-validated
TRIAL_NAME the trial's name, bare or embedded in free text — queryable but unverified
STRANDED present only in prose (description:, notes:, an evidence snippet:) — invisible to any query over trials

The audit is advisory. An identifier inside an evidence snippet: must not be "corrected" in place — that quote belongs to the cited paper, errors included. Add an ICTRP: evidence item alongside it instead.

Why this matters: the ChiCTR case

Progressive_Supranuclear_Palsy carried a trial named ChiCTR-2100045397. No registry holds that identifier — ICTRP returns its "not found" shell for it — and nothing in the validation stack could tell, because name is free text. The identifier was transcribed faithfully from the publication (PMID:36969340), whose abstract prints "Clinicaltrials.gov identifier: ChiCTR-2100045397" — wrong twice over: ChiCTR is not ClinicalTrials.gov, and the canonical form carries no hyphen.

The correct record, ChiCTR2100045397, is now cached and cited. Matching it to the publication also surfaced something worth recording: the registered record describes a single-arm phase 0 study of "movement disorders" with a healthy control group, while the paper reports a randomised placebo-controlled phase 2 trial confined to PSP-RS. Sponsor, investigators, intervention, and dates line up; the design does not. That discrepancy is recorded in the entry's notes: rather than resolved — a registry record you can actually read is what makes such a mismatch visible at all.

Detecting absence

The portal serves its "record not found" page with HTTP 200, so the status code cannot be used. The source detects the absent record by the missing main-ID field and raises with a pointed message:

$ just ictrp-fetch ChiCTR-2100045397
  skipped ChiCTR-2100045397: "ICTRP has no record for trial identifier
  'ChiCTR-2100045397'. Check the identifier against https://trialsearch.who.int
  — a registry ID transcribed from a paper is often malformed (ChiCTR IDs carry
  no hyphen after the registry name)."

Implementation

  • src/dismech/structured_sources/ictrp.py — ICTRPSource, portal parsing, serialization
  • src/dismech/ictrp_audit.py — the KB audit
  • tests/test_ictrp_source.py — offline tests against saved Trial2.aspx fixtures (contact values in the fixture are placeholders, not the live record's)

Trials on the ICTRP search portal are not endorsed by WHO; they are provided as a service. Each cached record repeats that disclaimer and links both the ICTRP record and the source-register record.