CTIS vs ClinicalTrials.gov vs China CTR: What a Failed WHO ICTRP Join Means in 2026

TL;DR

Takeaway: ClinicalTrials.gov, CTIS, China Drug Trials and WHO ICTRP are four public files with four identity contracts. A failed NCT ↔ CT number ↔ CTR ↔ TrialID join is usually the key, not proof that the study is unregistered.

When a sponsor or CRO lead is told a multinational study “is registered,” the next question is which file that sentence names. ClinicalTrials.gov issues an NCT to a study record [1][2]. The EU Clinical Trials Information System issues a CT number to a public trial under Regulation (EU) No 536/2014 [3][4]. China’s National Medical Products Administration posts drug-trial plans on chinadrugtrials.org.cn under a CTR number [5]. The Chinese Clinical Trial Registry issues a ChiCTR id on a different portal [6]. WHO’s International Clinical Trials Registry Platform collects TrialIDs from contributing registers; it is a search bridge, not a fifth equivalent register [7][8].

Three computed facts, from public extracts dated 25 July through 1 August 2026, change what a missing row means.

ICTRP is not a complete ClinicalTrials.gov mirror. The 25 July 2026 ClinicalTrials.gov year-CSV snapshot holds 595,630 unique NCT numbers. This ICTRP union holds 202,993 NCT-shaped TrialIDs, of which 202,671 overlap those NCT numbers. 392,959 ClinicalTrials.gov rows have no NCT TrialID in the union. ClinicalTrials.gov and WHO ICTRP — EClinCloud analysis, accessed August 2026. Before this file: “If it is not in ICTRP, it is not registered.” After: search the destination register. ICTRP absence is not ClinicalTrials.gov absence.

ICTRP CTIS TrialIDs do not join until the CTIS prefix is stripped. Public CTIS identity looks like 2022-500007-52-00. ICTRP writes the same object as CTIS2022-500003-41-00. Exact string overlap with 12,123 public CT numbers is 0. After stripping the prefix, 7,602 ICTRP ids remain, 7,517 of them overlap the CTIS public file, and 4,606 public CTIS trials are still absent from ICTRP. CTIS and WHO ICTRP — EClinCloud analysis, accessed August 2026. Before: “CTIS and ICTRP disagree, so one is wrong.” After: the join key is namespaced. Normalize the prefix. Do not treat the residual 4,606 as illegal.

China is two portals with zero overlapping identifiers in these files. The 30 July 2026 China Drug Trials list holds 35,996 rows and 35,989 unique CTR numbers; seven CTR values appear twice as distinct portal records. The list has no NCT field and no detail bodies. The ChiCTR extract used here is a 2026 listing only: 13,317 ChiCTR ids. Overlap with the CTR list is 0. ICTRP carries 0 CTR values as TrialIDs and 83 CTR strings in Secondary_ID, all of which match the China Drug Trials list. China Drug Trials, ChiCTR and WHO ICTRP — EClinCloud analysis, accessed August 2026. Before: “Check the China registry.” After: name which China file.

ICMJE still treats registration in a WHO primary register, or in ClinicalTrials.gov as a data provider to ICTRP, as the journal condition [9]. WHO itself says you cannot register a trial with WHO, and that country-level legal requirements differ from that journal rule [8]. A study can satisfy a journal’s “registered once” test and still be missing from the destination portal an ethics committee or inspector will open. The operational decision is which identity field to lock before calling the study missing. Getting it wrong delays a start-up that is already registered under another contract, or ships a study that is absent from the file that actually matters.

This paper is an identity workpaper for that decision. It is not a trial-volume ranking, not a how-to-register explainer, and not a GCP-process tour. Adjacent EClinCloud Insights cover ICH E6(R3) and a World Cup analogy for independent review. Those pieces answer different questions, on purpose.

What “registered” meant in the file you opened

Takeaway: “Registered” in a downloaded file means that file’s identity field is filled on a row of that file’s grain. 595,630 NCT rows, 12,123 CT numbers, 35,989 CTR numbers and 488,051 ICTRP TrialIDs are not four sizes of one list.

The first start-up error is ranking registers by CSV length. ClinicalTrials.gov is a US National Library of Medicine study registry. One public row is one NCT [1][10]. CTIS is the public face of the EU Clinical Trials Regulation for interventional trials of medicinal products in the EU/EEA; one public row is one CT number [3][4][11][12]. chinadrugtrials.org.cn is the NMPA/CDE drug-trial registration and disclosure platform; one list row is a CTR number, and seven CTR values in this snapshot are duplicated as distinct records [5][13][14]. ChiCTR is a WHO primary registry run as a research-registration portal, not the NMPA drug-trial list [6][15]. ICTRP is a platform that ingests contributing registers. You cannot register a trial with WHO [16]. The WHO Trial Registration Data Set is a 24-item minimum for a record to be considered fully registered in that network, not a promise that every national portal is mirrored [17][18].

This snapshot’s grain, from public extracts computed 23 August 2026:

ClinicalTrials.gov, AACT/CTTI, CTIS, China Drug Trials, ChiCTR and WHO ICTRP — EClinCloud analysis, accessed August 2026. Official landings: ClinicalTrials.gov and its API notes [1][2]; AACT [19][20]; CTIS and the Regulation [3][4][11]; China Drug Trials and the NMPA drug-registration rule [5][13]; ChiCTR and WHO’s ChiCTR profile [6][15]; ICTRP search and WHO ICTRP pages [7][8][16].

Read the table as a question filter.

• “Does this NCT exist on ClinicalTrials.gov?” — the year CSV, 595,630 unique NCT numbers, can answer.
• “Does this study have a CT number on the public CTIS portal?” — the 12,123-trial public extract can answer, after you use ctNumber, not an NCT harvested from a title.
• “Is this drug trial on chinadrugtrials.org.cn?” — the CTR list can answer. It cannot tell you the NCT.
• “Is this study on ChiCTR?” — a different portal, a different id, and in this paper a 2026 listing rather than a full ChiCTR dump.
• “Is this study in ICTRP?” — a TrialID question, after you normalize prefixes. It is not a substitute for the four questions above.

If the question and the grain do not match, stop. A bigger download will not fix a unit error. The next four sections name each object.

The NCT object (ClinicalTrials.gov)

Takeaway: A ClinicalTrials.gov row is an NCT object. Other-register identifiers are optional secondary strings. In the 1 August 2026 AACT identifier table, 4,151 studies carry a CTIS-class id, 287 carry a CTR, and 117 carry a ChiCTR id — sparse crosswalks, not a multinational census.

The Protocol Registration Data Element Definitions state the grain in one sentence: the NCT number is a unique identification code given to each clinical study record, format NCT plus eight digits [10]. Until an NCT is assigned, the study is not registered on ClinicalTrials.gov. Secondary IDs are defined as identifiers other than the organization’s unique protocol ID or the NCT, including numbers assigned by other public registries [10]. That sentence is permission to store a CT number or a CTR on the NCT record. It is not a requirement that the other register’s public file will join.

The 25 July 2026 year-CSV snapshot is 595,630 rows and 595,630 unique NCT numbers. Study type: INTERVENTIONAL 454,529; OBSERVATIONAL 139,061; EXPANDED_ACCESS 1,059; BLANK 981. Sponsor class: OTHER 425,599; INDUSTRY 131,151; OTHER_GOV 15,780; NIH 11,559; NETWORK 4,990; FED 4,914; plus thin tails. Status: COMPLETED 325,239; UNKNOWN 94,294; RECRUITING 65,408; TERMINATED 34,011; NOT_YET_RECRUITING 29,086; ACTIVE_NOT_RECRUITING 21,968; WITHDRAWN 16,588; ENROLLING_BY_INVITATION 5,248; SUSPENDED 1,748; WITHHELD 981; plus smaller buckets. Phase: NA 232,760; BLANK 141,220; then PHASE2 64,983, PHASE1 48,269, PHASE3 41,964, PHASE4 35,564, and combination strings. Enrollment is blank on 7,125 rows. Results are posted on 79,339. Start-year counts in this extract: 2024 40,185; 2025 39,277; 2026 27,590 (2026 is partial through 25 July). ClinicalTrials.gov — EClinCloud analysis, accessed August 2026 [1][2].

Those mixes describe this file. They are not a global interventional-trial census, not an industry-market share, and not a quality score. UNKNOWN 94,294 is a ClinicalTrials.gov status value, not a finding that 94,294 studies vanished. Phase NA 232,760 plus BLANK 141,220 is 373,980 rows whose phase cell is not a numbered phase — expected for observational and device-heavy records, not a hidden Phase 3 reservoir. The extract has no country column. Any country claim built from this CSV is an invention.

Secondary identifiers live in a different table. AACT, the Clinical Trials Transformation Initiative’s relational rebuild of ClinicalTrials.gov, publishes an identifier table keyed to NCT [19][20]. The 1 August 2026 snapshot holds 774,244 id rows. Source mix: organization study id 596,622; secondary id 174,282; NCT alias 3,340. Typed identifier type is BLANK on 628,862 rows, OTHER on 52,261, NIH on 41,023, REGISTRY on 19,325, EudraCT number on 18,328, other grant on 11,836, CTIS on 2,099. Most identifier strings are not classified as a foreign register id. AACT/CTTI — EClinCloud analysis, accessed August 2026.

Classifying values by identifier pattern plus the published type field produces the crosswalk a start-up team actually has on the NCT side:

Extracted CT-number strings from AACT: 4,132, of which 3,708 overlap the 12,123-row CTIS public extract. 424 CT-shaped strings on ClinicalTrials.gov are not in that CTIS public file. Extracted CTR values: 292, of which 289 match the China Drug Trials list. AACT/CTTI, CTIS and China Drug Trials — EClinCloud analysis, accessed August 2026.

Read those joins as coverage of a secondary field, not as coverage of the other register.

4,151 studies with a CTIS-class id, against 595,630 NCT rows, is 0.70%. A missing CT number on an NCT record is the default. It is not evidence the study lacks a CT number in CTIS. The EU file can hold a public trial whose NCT was never typed into PRS.

22,416 studies with a EudraCT id is the older EU object. The EU Clinical Trials Register remains the public search for trials approved under the Clinical Trials Directive and for some paediatric and third-country files; trials under or transitioned to Regulation 536/2014 are pointed to CTIS [21][22]. A EudraCT number on an NCT record is not a CT number. Treating them as interchangeable is the identity error this paper is written to stop.

287 studies with a CTR and 117 with a ChiCTR id are even thinner. They prove that some NCT records carry a China identifier. They do not prove that China Drug Trials or ChiCTR is a subset of ClinicalTrials.gov. The China Drug Trials list, as shown below, has no NCT column to reverse the join.

The 424 CT-shaped AACT values absent from the CTIS public extract are a remainder, not a non-compliance finding. Possible explanations, none of which this snapshot can choose among: withdrawn or not-yet-public CTIS records, mistyped strings, numbers outside this 30 July 2026 public cut, or timing between the 25 July ClinicalTrials.gov CSV and the 1 August AACT zip. AACT and the year CSV differ by one week. Small NCT drift is possible. Do not brief “424 illegal EU trials.”

The NCT object, compressed: lock NCT to answer a ClinicalTrials.gov question. Treat other-register strings as optional clues. If the destination file is CTIS or chinadrugtrials.org.cn, open that file. The NCT record will not do the join for you.

The CT number object (CTIS)

Takeaway: A public CTIS row is a CT number. Status in this extract is a numeric code, not COMPLETED or RECRUITING. The additional-registry array is nonempty on 826 / 12,123 trials and holds an NCT pattern in name or number on 41. Titles are not an identity field.

CTIS is the portal and database for the Clinical Trials Regulation. The Commission’s Regulation page is the legal landing; the public search at euclinicaltrials.eu is the file a start-up lead can actually open; EMA documents the public portal and the 18 June 2024 transparency rules [4][3][12][23]. The public website states the scope in plain language: interventional trials of investigational medicinal products. Observational studies and medical-device investigations are out of scope of this portal [3]. A missing CTIS row for an observational NCT is a scope miss, not a registration miss.

The 30 July 2026 public extract holds 12,123 summary records and 12,123 matching detail records. A catalog line that reads “24,246 rows” is summary plus details. It is not 24,246 trials. Identity: ctNumber is filled on all 12,123. Example grain: 2022-500007-52-00. CTIS — EClinCloud analysis, accessed August 2026 [3].

Status is a numeric code. Top values: 4 = 3,859; 8 = 3,579; 5 = 2,171; then 2 = 1,379; 3 = 799; 11 = 207; 6 = 110; plus a thin tail (9 = 11; 7 = 3; 12 = 2; 1 = 2; 10 = 1). This paper does not map those integers onto ClinicalTrials.gov COMPLETED, RECRUITING or UNKNOWN. No official public code table is cited here, so no English label is invented. If a tracker writes “COMPLETED” next to code 4, that label came from somewhere else. It did not come from this extract.

Country is the column ClinicalTrials.gov’s year CSV lacked. Trial countries are filled on all 12,123 trials. Countries per trial: minimum 1, median 1, p75 4, p90 8, p95 11, p99 16, maximum 25, mean 3.18. Half of public CTIS trials in this file are single-country trials. The p90 of 8 is the multinational tail, not the median. Country-token frequency (a trial can contribute more than one token): Spain 4,915; France 4,562; Germany 3,989; Italy 3,739; Poland 2,748; Netherlands 2,621; Belgium 2,397; then Denmark, Czechia, Hungary, Sweden, Austria, Greece, Portugal, Romania. Those are mentions, not exclusive national censuses, and not enrollment.

Phase strings in this file are already English and already CTIS’s own labels, so they can be quoted as file values: Therapeutic exploratory (Phase II) 3,565; Therapeutic confirmatory (Phase III) 3,351; Therapeutic use (Phase IV) 1,239; Human Pharmacology (Phase I)–Other 1,092; First administration to humans 822; plus integrated Phase I/II, II/III and III/IV buckets. Sponsor-type is messy because the field concatenates multiple parties: Pharmaceutical company 6,589 as a single token; Hospital/Clinic/Other health care facility 3,592; then educational, laboratory and patient-organisation labels, including duplicated concatenations. Do not collapse concatenations into a clean “industry vs academic” pie.

Primary or endpoint text is blank on 2,215 of 12,123. That is a fill limitation of the public extract. It is not proof the trial has no endpoint. Enrolled is blank on 0 rows in this projection — a different completeness profile from ClinicalTrials.gov’s 7,125 blank enrollment cells. Different files, different fields. Neither number is an enrollment-quality score.

The identity trap is the additional-registry array and the identifiers JSON.

Details: additional-registry is nonempty on 826 / 12,123 (6.8%). Named ClinicalTrials.gov / Clinicaltrials.gov additional-registry rows, combining two spellings in the name field, total 19. An NCT pattern in the additional-registry name or number appears on 41 trials. An NCT pattern anywhere in the public identifiers object, including titles, appears on 4,828. Titles are not an identity field. Harvesting NCT0… from a public title and calling it a join key will over-count. WHO Universal Trial Number is filled on 1,750. The EudraCT transitioned flag is set on 4,652. CTIS — EClinCloud analysis, accessed August 2026.

Top additional-registry names in the nonempty 826: IND 350; EudraCT 74; DRKS 25; OMON 13; Clinicaltrials.gov 12; then a scatter of EUDRACT, jRCT, ANZCTR, IND number, EU Clinical Trials Register, ClinicalTrials.gov 7. CTIS’s extra-register field, when used at all, is more often an IND or a EudraCT string than an NCT. A blank additional-registry array is the normal public CTIS row.

The CT number object, compressed: lock ctNumber to answer a CTIS question. Do not decode status integers without an official table. Do not treat a missing NCT on a CTIS detail as proof the study lacks an NCT. Do not join on titles. If the destination ethics committee will open euclinicaltrials.eu, the CT number is the field. The NCT, if it exists, belongs in study-build metadata as a second id.

The CTR object — and the ChiCTR object people mix with it

Takeaway: chinadrugtrials.org.cn and chictr.org.cn are two China portals. This snapshot’s CTR list and ChiCTR 2026 listing share 0 identifiers. Seven CTR numbers appear twice as distinct list records. There is no NCT column on the drug-trial list.

NMPA’s Provisions for Drug Registration require the sponsor, before carrying out a drug clinical trial, to submit the protocol and other information on the drug clinical trial registry and information disclosure platform, and to keep that information updated [13]. CDE maintains that platform. NIAID ClinRegs points the same obligation at NMPA Notice No. 9 of 2020 and at the 2020 Drug Registration Regulation [14]. The public search list is chinadrugtrials.org.cn [5]. A CTR number is the identity of that object.

ChiCTR is a different object. It is a WHO primary registry, with a WHO registry-network profile and an English-language portal, accepting interventional and other study types for research registration [15][6][24]. WHO’s own trial-registration page still says it is necessary to register once in a primary registry to meet WHO transparency and ICMJE-facing publication tests, while warning that regulatory requirements differ by country [16][9]. That “once” is a journal-network sentence. It is not a sentence that collapses CTR into ChiCTR.

The 30 July 2026 China Drug Trials list is 35,996 rows, 35,989 unique registration numbers, 7 CTR values each appearing twice as distinct portal records. Every unique number uses the CTR prefix. Filled on every row in this projection: indication, public title, registration number, trial status. Drug name is filled on 35,995 / 35,996. NCT field: none. Detail bodies: 0. China Drug Trials — EClinCloud analysis, accessed August 2026 [5].

Status is Chinese, and it is complete:

Those eight labels sum to 35,996. Do not translate 已完成 into ClinicalTrials.gov COMPLETED and then join the two files on the translated word. Do not translate 进行中 招募中 into RECRUITING and then brief a global recruiting census. The vocabulary is this portal’s.

The seven duplicate CTR numbers, each with two distinct records, are:

CTR20130074, CTR20130320, CTR20130603, CTR20180468, CTR20190909, CTR20230117, CTR20242926.

A duplicate CTR in this list is two portal records sharing one registration number. It is not automatic proof of two trials, of a data-quality failure that a start-up team must “fix” before first-patient-in, or of a missing registration. The operational move is to open both records, compare titles and dates, and record that the number is not unique in the public list. Do not pick one row at random and call the other a ghost.

Because there are no detail bodies in this extract, investigator names, site lists, inclusion criteria and protocol text are file-absent, not study-absent. A start-up tracker that expects a China “full posting” from this list will always look empty. The list can answer “is this CTR on the public search?” It cannot answer “who is the PI in Beijing?”

ChiCTR in this paper is a contrast grain, not a census. The 1 August 2026 listing is 2026-only: 13,317 rows and 13,317 unique ids. Study type in that listing: 干预性研究 7,576; 观察性研究 5,285; 诊断试验 449; plus a handful of treatment, etiology, prognosis, epidemiology and prevention labels. Overlap of those 13,317 ids with the 35,989 CTR numbers: 0. ChiCTR — EClinCloud analysis, accessed August 2026 [6][24].

Zero overlap is the finding that kills the phrase “the China registry.” A drug-trial CTR will not appear as a ChiCTR id in these files. A 2026 ChiCTR listing id will not appear as a CTR. Searching ChiCTR for a CTR string, or chinadrugtrials.org.cn for a ChiCTR string, is the wrong key. The protocol identity choice — NMPA drug-trial registration versus WHO-primary research registration — has to be made on purpose, then stored as two fields if both apply.

ICTRP does not rescue the mix. CTR as TrialID: 0. CTR in Secondary_ID: 83, all 83 matching the China Drug Trials list. ChiCTR as an ICTRP Source_Register is large — 44,062 rows in current XML — and overlaps this 2026 listing on 2,042 ids. That 2,042 / 13,317 ratio is what a 2026-only listing predicts, not a ChiCTR outage. WHO ICTRP has more ChiCTR identifiers than this listing. Using the listing as “all of ChiCTR” would under-count on purpose.

The CTR object, compressed: lock CTR to answer a chinadrugtrials.org.cn question. Lock ChiCTR id to answer a ChiCTR question. Name which file in the start-up checklist. Store both when the protocol actually has both. Do not infer one from the other. Do not infer NCT from either: the drug-trial list does not carry it.

ICTRP is a bridge that namespaces and truncates

Takeaway: ICTRP TrialID is a namespaced, provenance-split identifier. Exact CTIS overlap is 0; after stripping CTIS, 7,517 / 12,123 public CTIS trials join. 392,959 ClinicalTrials.gov NCT numbers are absent. The older CSV’s blank Source_Register will under-count ClinicalTrials.gov if you filter on that field.

WHO’s ICTRP page and the Search Portal are the public doors [8][7]. The network page lists primary registries; the trial-registration page states you cannot register with WHO and that one primary-registry record can satisfy WHO/ICMJE transparency tests [18][16][9]. ICMJE’s FAQs still send registry-acceptability questions to WHO rather than to ICMJE itself [25]. None of those sentences says ICTRP is a complete copy of ClinicalTrials.gov, CTIS or chinadrugtrials.org.cn.

The 1 August 2026 ICTRP union holds 488,051 unique TrialIDs. Provenance: 267,740 verified current XML plus 220,311 older capped CSV. Coverage note in the snapshot: the current XML component is complete for the portal’s current-trial total; the older CSV is not a complete historical portal. Source_Register is BLANK on all 220,311 capped-CSV rows. WHO ICTRP — EClinCloud analysis, accessed August 2026.

That blank is a join landmine. If a programmer filters Source_Register = ClinicalTrials.gov, the filter sees 125,093 rows — current XML only — and reports 124,825 overlapping the ClinicalTrials.gov NCT set, with 470,805 ClinicalTrials.gov NCT numbers “missing.” The NCT-shaped TrialID count is 202,993, overlapping 202,671 ClinicalTrials.gov rows, with 392,959 ClinicalTrials.gov NCT numbers still absent and 322 ICTRP-only NCT-shaped ids. Using Source_Register under-counts ClinicalTrials.gov inside ICTRP. Using TrialID still leaves two-thirds of this ClinicalTrials.gov snapshot outside the union (392,959 / 595,630).

ICTRP is not “wrong” for that miss. It is a bridge with a capped historical supplement and a current-XML core. A 2014 NCT that never landed in the current XML and that sits outside the capped CSV will not join. A 2026 NCT that contributing registers have not yet pushed will not join. Neither case is a ClinicalTrials.gov deletion.

CTIS is the prefix problem.

ICTRP Source_Register = Clinical Trials Information System holds 3,280 rows. Exact TrialID ∩ public ctNumber: 0. ICTRP does not store 2022-500003-41-00. It stores CTIS2022-500003-41-00. After stripping the CTIS prefix, 7,602 ids remain; 7,517 overlap the 12,123 public CT numbers; 4,606 public CTIS trials are still not in ICTRP. The 7,602 − 7,517 = 85 stripped ids that do not hit this public CTIS extract are the same class of remainder as the 424 AACT CT-shaped strings: timing, non-public, mistype, or outside this cut — not a finding of illegal registration.

Secondary_ID is a third, weaker key. NCT-like strings in Secondary_ID: 2,463. CTIS-like: 1,364. CTR: 83, all matching China Drug Trials, with 0 CTR as TrialID. Secondary_ID can recover a handful of cross-register clues. It cannot replace TrialID, and it cannot replace the destination file.

Other Source_Register counts, current XML only, for orientation rather than ranking: ChiCTR 44,062; IRCT 16,766; JPRN 15,752; CTRI 11,554; EU Clinical Trials Register 10,166; NL-OMON 9,217; German Clinical Trials Register 6,779; ANZCTR 5,880; ISRCTN 4,180. “EU Clinical Trials Register” 10,166 is the EudraCT-era contributing register, not CTIS. Joining that field to a CT number will fail for the same reason a EudraCT secondary id is not a CT number.

Countries are blank on 24,314 ICTRP rows. That is another fill limit, not a geography of unregistered trials.

The ICTRP object, compressed: use it as a discovery bridge after you normalize prefixes and after you stop treating Source_Register as complete. Do not use it as the global census. Do not use a failed ICTRP lookup as the start-up fail. Open ClinicalTrials.gov for NCT, CTIS for CT number, chinadrugtrials.org.cn for CTR, ChiCTR for ChiCTR id. ICTRP’s 488,051 TrialIDs are a real public file. They are a different contract.

A four-step identity sequence

Takeaway: Name the destination file, lock the identity field that file actually fills, normalize prefix and case, and only then treat the remainder as a gap to investigate. ICTRP’s 392,959 missing NCT rows, CTIS’s CTIS prefix, and China’s 0 overlap are the reasons the order matters.

A workable sequence, derived from the joins above rather than from a generic registration SOP:

1. Name the destination file in one sentence. “Will the EU ethics committee open euclinicaltrials.eu?” is a CTIS question [3]. “Will the US PRS record be the public face?” is a ClinicalTrials.gov question [1]. “Must this drug trial appear on chinadrugtrials.org.cn before launch?” is a CTR question [5][13]. “Is the publication path a WHO primary ChiCTR record?” is a ChiCTR question [15][9]. “I searched ICTRP and found nothing” is not yet a destination. Write the portal name before you write “missing.”

2. Lock the identity field that file fills. NCT on ClinicalTrials.gov. ctNumber on CTIS. CTR on chinadrugtrials.org.cn. ChiCTR id on chictr.org.cn. TrialID on ICTRP, knowing it is namespaced. Do not lock on NCT for the China Drug Trials list — the column does not exist. Do not lock on NCT for CTIS additional-registry — it is filled as an NCT pattern on 41 of 12,123 trials. Do not lock on a title string. Do not lock on a translated status word.

3. Normalize prefix, spelling and case before you join. Strip CTIS from ICTRP TrialIDs before comparing to ctNumber; exact overlap is 0, prefix-stripped overlap is 7,517. Accept both ClinicalTrials.gov and Clinicaltrials.gov in CTIS additional-registry names (those two spellings are how the 19 named rows appear). Treat CTR as CTR plus digits, not as a ChiCTR id. Do not case-fold a ChiCTR id into a CTR. Do not treat a EudraCT number as a CT number: AACT still carries EudraCT on 22,416 studies, and the EU Clinical Trials Register still exists as a separate public object [21][22].

4. Only then treat the remainder as a gap to investigate — not as a crime. Remainder classes already measured here: 392,959 ClinicalTrials.gov NCT numbers not in this ICTRP union; 4,606 public CTIS trials not in ICTRP after prefix strip; 424 CT-shaped AACT strings not in the CTIS public extract; 85 prefix-stripped ICTRP CTIS ids not in that extract; 7 CTR numbers with two portal records; ChiCTR 2026 listing overlap with ICTRP ChiCTR 2,042 / 13,317 because the listing is 2026-only. Each remainder has at least one non-illegal explanation: capped history, namespace, snapshot date, non-public record, mistype, sister portal, listing scope. Investigate with the destination file and the protocol’s actual register list. Do not close a start-up gate on a failed ICTRP exact match.

Store every register id the protocol actually holds as its own field. A single cell labelled “registry ID” will hold an NCT one week and a CT number the next, and the join will fail in both directions. EDC study-build metadata and CTMS study records are the operational places those fields live; eTMF holds the registration documents that correspond to them. Those are three products, not one merged “platform registry.” EClinCloud does not join ClinicalTrials.gov, CTIS, China Drug Trials, ChiCTR or ICTRP into a global index [26][27]. The systems that run the study still have to store each id.

Frequently asked questions

Takeaway: Most “unregistered” tickets in a multinational start-up are failed joins: ICTRP used as a census, CTIS prefix left on, CTR searched in ChiCTR, or a status integer read as COMPLETED.

The study has an NCT on ClinicalTrials.gov but I cannot find it in ICTRP. Is it unregistered?

No. This ClinicalTrials.gov snapshot has 595,630 unique NCT numbers. This ICTRP union has NCT-shaped TrialIDs overlapping 202,671 of them. 392,959 NCT numbers are absent from the union. ICTRP is a bridge with 220,311 older rows whose Source_Register is blank and whose history is capped. Search ClinicalTrials.gov for an NCT question [1][7][8]. WHO ICTRP — EClinCloud analysis, accessed August 2026.

CTIS and ICTRP show a 0% exact match. Which file is wrong?

Neither, on this evidence. ICTRP TrialIDs for CTIS rows are prefixed CTIS (example CTIS2022-500003-41-00) against public ctNumber values such as 2022-500007-52-00. Exact overlap is 0. After stripping the prefix, 7,517 / 12,123 public CTIS trials join. 4,606 public CTIS trials remain absent from ICTRP; do not treat that residual as illegal non-registration. CTIS and WHO ICTRP — EClinCloud analysis, accessed August 2026 [3].

I was told to “check the China registry.” Which one?

Name the portal. chinadrugtrials.org.cn is the NMPA/CDE drug-trial platform; identity is CTR; this list is 35,996 rows / 35,989 unique CTR numbers, with no NCT field [5][13]. ChiCTR is a WHO primary research registry; this extract is a 2026 listing of 13,317 ids [6][15]. Overlap of CTR list with that listing is 0. ICTRP has 0 CTR TrialIDs and 83 CTR Secondary_IDs, all matching the drug-trial list. “China registry” without CTR or ChiCTR is already the wrong key.

Seven CTR numbers appear twice. Which record is the real trial?

This snapshot cannot say. The seven values — CTR20130074, CTR20130320, CTR20130603, CTR20180468, CTR20190909, CTR20230117, CTR20242926 — each appear as two distinct records on the public list. Open both, compare titles and dates, and record that uniqueness failed in the file. Do not collapse them in a tracker without looking. China Drug Trials — EClinCloud analysis, accessed August 2026 [5].

CTIS status is a number. Can I map code 4 to COMPLETED?

Not from this paper. Top codes are 4 = 3,859, 8 = 3,579, 5 = 2,171. No official public code table is cited here, so no English label is attached. Mapping those integers onto ClinicalTrials.gov COMPLETED or RECRUITING invents a crosswalk. If you need a label, cite the CTIS/EMA table you used. CTIS — EClinCloud analysis, accessed August 2026 [3][23].

AACT shows a CTIS id on the NCT record, but I cannot find that CT number in the public CTIS extract. Is the EU trial missing?

Not on the strength of 424 unmatched CT-shaped strings. AACT extracted 4,132 CT-number values and overlapped the public CTIS file on 3,708. The 424 remainder can be non-public, mistyped, withdrawn, or outside the 30 July 2026 public cut. AACT is dated 1 August 2026; the ClinicalTrials.gov CSV is dated 25 July 2026. Timing is a live explanation. Investigate the CT number on euclinicaltrials.eu before calling the study unregistered in the EU [19][3].

CTIS details mention NCT in the public identifiers object. Can I join on that?

Not if the hit came from a title. An NCT pattern appears anywhere in the identifiers JSON on 4,828 trials, and in additional-registry name or number on 41. Additional-registry is empty on 11,297 / 12,123. Titles are not an identity field. Lock ctNumber. Store NCT separately if the protocol has one.

If ICMJE only needs one primary-registry record, why store four ids?

Because journal policy and destination-file obligations are different jobs. ICMJE accepts a WHO primary register or ClinicalTrials.gov [9][25]. WHO says one primary-registry record can meet its transparency test, and also says national legal requirements differ [16]. A multinational protocol can have an NCT, a CT number and a CTR at once. AACT shows how rarely those other ids are typed back onto the NCT record (4,151 CTIS-class, 287 CTR). Store each id the protocol actually holds.

Methodology and limitations

Takeaway: Counts are records in dated public extracts, not enrollment, not market size, and not a finding of non-compliance. Filters have edges; they are reported rather than smoothed.

• ClinicalTrials.gov: 25 July 2026 year-CSV snapshot, 595,630 unique NCT rows. No country column in this extract. Status, phase, study type, sponsor class, enrollment-blank and results-posted as tabulated. ClinicalTrials.gov — EClinCloud analysis, accessed August 2026 [1][2].
• AACT / CTTI: 1 August 2026 identifier table, 774,244 id rows. Other-register classes by pattern plus the published identifier-type field. AACT and the year CSV differ by one week; small NCT drift is possible [19][20].
• CTIS: 30 July 2026 public summaries and details, 12,123 trials. Catalog “24,246” is summary+details. Status integers are not decoded. Additional-registry and NCT-pattern counts as tabulated. CTIS — EClinCloud analysis, accessed August 2026 [3][12].
• China Drug Trials: 30 July 2026 public list, 35,996 rows, 35,989 unique CTR, seven duplicated numbers, no NCT field, no detail bodies. China Drug Trials — EClinCloud analysis, accessed August 2026 [5].
• ChiCTR: 1 August 2026 listing, 2026-only, 13,317 ids. Used as a contrast grain, not a ChiCTR census. ICTRP current-XML ChiCTR count is 44,062 [6][15].
• WHO ICTRP: 1 August 2026 union, 488,051 unique TrialIDs: 267,740 current XML + 220,311 older capped CSV with Source_Register blank. Prefix-stripped CTIS join as tabulated. WHO ICTRP — EClinCloud analysis, accessed August 2026 [7][8].
• Primary pages: ClinicalTrials.gov data-element definitions; Regulation (EU) No 536/2014 and CTIS public search; NMPA drug-registration provisions and ClinRegs; ChiCTR and WHO primary-registry profile; ICMJE registration policy; WHO ICTRP trial-registration and data-set pages [10][11][13][14][9][16][17].
• Limitations: ICTRP older CSV under-counts any analysis that needs Source_Register. CTIS status codes have no English labels in this paper. China list has no details. ChiCTR extract is 2026-only. 424 unmatched AACT CT-shaped strings and 4,606 CTIS public trials absent from ICTRP are remainders, not non-compliance findings. Counts are not rates, forecasts or enrollment quality. Missing field is not unregistered.

Conclusion

Takeaway: Lock NCT, CT number, CTR and ChiCTR id as separate fields before treating a missing row as a start-up failure. ICTRP is a bridge. EDC and CTMS are where those ids should live — not a single “registry ID” cell.

Four takeaways survive the joins.

Four files, four contracts. ClinicalTrials.gov is an NCT object (595,630). CTIS is a CT-number object (12,123 public trials, not 24,246). China Drug Trials is a CTR list (35,989 unique numbers, seven duplicated as records, no NCT). ChiCTR is a different China portal (2026 listing 13,317, overlap with CTR 0). ICTRP is a TrialID bridge (488,051), not a super-register.

A failed join is usually the key. 392,959 ClinicalTrials.gov NCT numbers are absent from this ICTRP union. Exact CTIS overlap is 0 until CTIS is stripped, after which 7,517 join and 4,606 public CTIS trials still do not. CTR as ICTRP TrialID is 0. Searching the wrong portal, or the right portal with the wrong string, produces a precise answer to a different question.

Secondary ids are sparse. AACT records a CTIS-class id on 4,151 studies, a EudraCT id on 22,416, a CTR on 287, a ChiCTR id on 117. CTIS additional-registry is nonempty on 826 trials and NCT-patterned on 41. The NCT record will not carry the EU and China identities for you. Titles will not either.

Store each register id as its own field. Name the destination, lock the field, normalize the prefix, then investigate remainders. EDC study-build metadata and CTMS study records are the place those fields belong — NCT, CT number, CTR, ChiCTR id, ICTRP TrialID — rather than one cell called “registry ID.” eTMF holds the corresponding documents. EClinCloud does not claim to join these public files or to host a global trial index [26][27].

If the question you are actually asking is which public identifier to lock before treating a row as missing, start there. Product pages for EDC, CTMS and eTMF are on eclincloud.com [26][27].

Sources

1. U.S. National Library of Medicine, ClinicalTrials.gov — public study registry; NCT identity. ClinicalTrials.gov — EClinCloud analysis of the 25 July 2026 year-CSV snapshot (595,630 unique NCT), accessed August 2026.

2. ClinicalTrials.gov, About the API — official notes on machine-readable ClinicalTrials.gov content used to land the public file.

3. European Medicines Agency / CTIS, Search for clinical trials — public CTIS portal. CTIS — EClinCloud analysis of the 30 July 2026 public extract (12,123 trials), accessed August 2026.

4. European Commission, Clinical trials — Regulation (EU) No 536/2014 — CTIS as the portal and database for the Regulation; transition from the Clinical Trials Directive; transparency rules.

5. National Medical Products Administration / Center for Drug Evaluation, Drug Clinical Trial Registration and Information Disclosure Platform — search list — public CTR list. China Drug Trials — EClinCloud analysis of the 30 July 2026 list (35,996 rows; 35,989 unique CTR), accessed August 2026.

6. Chinese Clinical Trial Registry, ChiCTR — WHO primary registry portal. ChiCTR — EClinCloud analysis of the 1 August 2026 listing (13,317 ids, 2026-only), accessed August 2026.

7. World Health Organization, ICTRP Search Portal — public search across contributing registers. WHO ICTRP — EClinCloud analysis of the 1 August 2026 union of current XML and older CSV (488,051 unique TrialIDs), accessed August 2026.

8. World Health Organization, International Clinical Trials Registry Platform — ICTRP overview; ICTRP is not itself a registry that accepts direct registrations.

9. International Committee of Medical Journal Editors, Clinical Trials — Registration — registration in a WHO ICTRP primary register or in ClinicalTrials.gov as a condition of consideration for publication.

10. ClinicalTrials.gov, Protocol Registration Data Element Definitions for Interventional and Observational Studies — NCT number format; Secondary ID including other public-registry identifiers.

11. European Union, Regulation (EU) No 536/2014 — Clinical Trials Regulation; EU portal and EU database.

12. European Medicines Agency, Clinical Trials Information System — CTIS public portal; revised transparency rules applicable 18 June 2024.

13. National Medical Products Administration, Provisions for Drug Registration — Article 33: sponsor registration of the drug clinical trial protocol and related information on the registry and information-disclosure platform before the trial is carried out.

14. National Institute of Allergy and Infectious Diseases, ClinRegs — China — English compilation pointing at the Drug Registration Regulation and NMPA Notice No. 9 of 2020 (Drug Clinical Trial Registration and Information Disclosure Management Standards).

15. World Health Organization, Chinese Clinical Trial Registry (ChiCTR) — primary registry profile — ChiCTR as a WHO ICTRP primary registry, distinct from the NMPA drug-trial platform.

16. World Health Organization, Trial registration — you cannot register a trial with WHO; one primary-registry record can meet WHO transparency tests; national regulatory requirements differ.

17. World Health Organization, WHO Trial Registration Data Set — 24-item minimum data set for a trial to be considered fully registered in the network.

18. World Health Organization, WHO Registry Network — primary registries that supply ICTRP.

19. Clinical Trials Transformation Initiative, AACT Database — Aggregate Analysis of ClinicalTrials.gov. AACT/CTTI — EClinCloud analysis of the 1 August 2026 identifier table (774,244 id rows), accessed August 2026.

20. Clinical Trials Transformation Initiative, AACT data dictionary — NCT-keyed protocol identifiers (organization study id, secondary id, NCT alias).

21. European Union Clinical Trials Register, Search — EudraCT-era public register; points Regulation 536/2014 trials to CTIS.

22. European Medicines Agency, EudraCT public website — as of 31 January 2025, ongoing EU/EEA trials are under Regulation 536/2014 / CTIS; EudraCT remaining uses are limited.

23. European Medicines Agency / CTIS, Guidance and Q&As — public-portal scope and revised transparency rules.

24. Chinese Clinical Trial Registry, ChiCTR English home — English-language entry to the WHO primary registry, used here as the contrast portal to chinadrugtrials.org.cn.

25. International Committee of Medical Journal Editors, Clinical Trials Registration — FAQs — WHO primary registries are ICMJE-acceptable; ICMJE is no longer the entity that reviews registries for acceptability.

26. EClinCloud, Home — EDC, CTMS, eTMF and related clinical-research products; contact landing. EClinCloud does not join the public registers analysed in this paper.

27. EClinCloud, Products — distinct product lines including EDC and CTMS, the operational systems in which each register identifier should be stored as its own field.