FDA PFDD Guidance 3 and ClinicalTrials.gov 2026: PRO vs ClinRO Tokens, Patient-Reported Outcome Registration, and eCOA Implementation

TL;DR

Takeaway: Fit-for-purpose is not a ClinicalTrials.gov field. Literal COA taxonomy tokens appear in 272 of 579,073 studies that have an outcome row (0.047%). PRO-language appears in 16.7%. FDA PFDD Guidance 3 is the selection file. The public outcome table is a naming surface.

Three computed facts fix the scale of that gap.

On the 1 August 2026 AACT copy of ClinicalTrials.gov, the protocol-specified outcome table holds 3,716,648 rows. 579,073 studies have at least one such row. Search that text for the literal tokens COA, ClinRO, ObsRO, PerfO, ePRO or eCOA and you get 589 rows in 272 studies — 0.047% of studies with outcomes. Broaden the ClinRO pattern to “clinician-reported,” “investigator-assessed,” “investigator-reported” and “physician-reported,” and ClinRO-pattern studies are still 0.347% (2,011). ObsRO-pattern studies are 0.224% (1,299). An eCOA / electronic-diary pattern reaches 0.157% (912). ClinicalTrials.gov / AACT — EClinCloud analysis, accessed August 2026 [1][2].

PRO, quality-of-life and named-questionnaire language is a different frequency class. A patient-reported / ePRO / PROMIS / EQ-5D / SF / EORTC / QoL / diary pattern hits 184,540 rows in 96,440 studies (16.7%). Most of those PRO hits sit on secondary rows (135,296), not primary (37,390). Measure text is blank on 9 rows. The registry is not missing names. It is missing the object Guidance 3 asks you to build: type, context of use, and evidence that the measure is fit for that use [1][2][3].

FDA’s final Patient-Focused Drug Development: Selecting, Developing, or Modifying Fit-for-Purpose Clinical Outcome Assessments (Guidance 3) published on 22 October 2025. It tells sponsors how to select, develop or modify a COA for a stated concept, population and context of use [1]. ClinicalTrials.gov’s protocol-registration data elements store an outcome title, a description and a time frame, typed as primary, secondary or other. They do not store COA type, instrument version, administration mode, language, rater identity or fit-for-purpose evidence [3]. A blank ClinRO token is not a bad endpoint. A named EQ-5D row is not a validation file.

The decision this changes is the one an endpoint or eCOA lead makes before locking the build. If you key an instrument library to public outcome titles, you will confuse naming with selection. If you score registry quality by the presence of “ClinRO” or “eCOA,” you will false-fail almost the entire historical file — including studies registered a decade before Guidance 3 existed. Put type, instrument and version, mode, language, rater and visit window in the protocol and in the eCOA / scale-management specification. Do not wait for ClinicalTrials.gov to grow a field it was never designed to hold.

We already explained the four COA types, and what “fit-for-purpose” asks of a sponsor, in a shorter Insight [4]. This paper does not retell that taxonomy as the thesis. It asks a different question: after you have read Guidance 3, can you see the selection in the public record?

What Guidance 3 requires that a public row cannot hold

Takeaway: Guidance 3 is a selection standard for a concept, a population and a context of use. ClinicalTrials.gov stores the name of an outcome. Those are different objects. The four COA types belong in one short box; the rest of this paper is about the public file.

Guidance 3 is the third of four methodological PFDD documents. It finalizes a June 2022 draft. FDA’s own series page positions it as the file for selecting, modifying, developing and evaluating clinical outcome assessments that measure outcomes important to patients [1][5]. The 2009 guidance Patient-Reported Outcome Measures: Use in Medical Product Development to Support Labeling Claims remains the older PRO-labeling object. It tells you how FDA reviews a PRO instrument used to support a claim. It does not turn the ClinicalTrials.gov outcome row into that review [6].

The four types, in one box, then we leave them.

A COA is fit-for-purpose when the evidence supports this instrument, in this population, for this role in the endpoint strategy. That is a scientific file: concept of interest, content validity, reliability, ability to detect change, administration, language, scoring. We walked that argument in From regulatory intent to practice: choosing a fit-for-purpose clinical outcome assessment [4]. This paper will not retell it.

What the Insight cannot tell you — and what the registry will not show you — is whether a given NCT ever made that selection. The protocol-registration schema has no COA-type code. It has no instrument-version field. It has no mode-of-administration enumerator (paper, provisioned device, web, interview). It has no language field on the outcome row. It has no rater field. Title, description and time frame can name a scale. They cannot hold the evidence package Guidance 3 describes [3].

That is the split this article is built on. Guidance 3 is the selection file. ClinicalTrials.gov is the naming surface. The next sections measure the surface.

The outcome table is a naming surface

Takeaway: 3,716,648 protocol-specified outcome rows sit on 579,073 studies. 62.8% of those rows are secondary, 31.6% primary, 5.5% other. Measure text is blank on 9 rows. The field is filled with names. It is not filled with COA type, version, mode or evidence.

AACT is CTTI’s public relational copy of every ClinicalTrials.gov protocol and results data element, refreshed from the registry and also published as static downloads [7][2]. The table that holds protocol-specified outcomes is design_outcomes. It is not the results-module outcomes table. Posted results can add, drop or re-label measures after the fact. This analysis stays on the design table on purpose: it is the public naming surface a clinical-science lead actually scrapes when they ask “what did this study say it would measure?”

The 1 August 2026 static copy holds 3,716,648 design-outcome rows. Joined to the ClinicalTrials.gov study file current as of 25 July 2026 (595,630 NCT identifiers), 579,073 studies have at least one design-outcome row. That is the study-level denominator used everywhere below. The 16,557 studies in the study file with no design-outcome row are a different gap — registration without a public outcome list — and they are not mixed into the 0.047% taxonomy share [2][8].

Row type on that table:

ClinicalTrials.gov / AACT — EClinCloud analysis, accessed August 2026 [2].

Two facts in that table are easy to misread.

First, secondary rows outnumber primary rows 2.0 to 1. A study-level “this trial has a PRO” flag, built by scanning any outcome text, will mostly be a flag on a secondary or “other” line. That matters for eCOA build: a secondary PRO still needs an instrument, a mode, a language pack and a visit window. It does not automatically belong in the primary estimand. Treating every named QoL row as a primary endpoint is a design error the registry will not prevent, because the registry’s type field is primary / secondary / other, not “this is the thing we powered on and this is a supportive PRO.”

Second, blank measure text is not the problem. Nine rows in 3.72 million lack a measure string. The public file is not an empty-field story. It is a filled-with-the-wrong-object story. Sponsors type a name because the Protocol Registration system requires a title. NLM’s protocol data-element definitions say a primary outcome measure is the planned measurement of greatest importance, usually the one used in the power calculation, and that each outcome carries a title, a description and a time frame. Time frame is limited to 254 characters and must be specific to the metric, generally the duration over which each participant is assessed, not the overall study duration [3]. That is a milestone. It is not an eCOA visit window, a diary recall period, or a device-provisioning interval.

PRS quality-control review goes one step further on naming and still does not reach Guidance 3. The published QC criteria ask that the outcome-measure title describe the specific measurement — a scale, a physiological parameter, a questionnaire — and the metric for how data will be aggregated, for example mean change from baseline. Title and description should say what will be measured, not the study’s aim [9]. A well-written public row can therefore read “Mean change from baseline in EQ-5D-5L index at week 12.” That row can pass QC. It still does not record that the instrument is a PRO, that the 5L version rather than the 3L was licensed, that administration is a provisioned tablet, that the language pack is US English plus simplified Chinese, or that content validity was established in the target population.

Applicable clinical trials have a legal reason to put some outcome text in the protocol module. FDAAA 801 and 42 CFR Part 11, as ClinicalTrials.gov explains them, require registration of specified data elements, including primary and secondary outcome measures, for trials that fall under the statute [10][8]. The duty is to name the planned measurement. The duty is not to deposit a COA evidence package. Reading statutory completeness as scientific completeness is the error.

One more grain trap, then we leave other registries alone. In a companion cut of 12,123 EU CTIS public trial summaries, 2,215 were blank on primary/endpoint text [11]. That is a different file, a different field, and a different completeness story. Do not mix 18.3% CTIS endpoint-blank into the AACT study-level shares above. ClinicalTrials.gov’s design-outcome measure field is almost never blank. CTIS’s public endpoint text is sometimes blank. Both statements can be true. They are not one rate.

The results module is a third object. ClinicalTrials.gov copies protocol outcome information into the results section the first time results are created, and it allows post-hoc measures to be reported there [12]. We did not analyse posted results. A results table that names a PRO is still not a validation file. It is a later naming surface.

What you can defend from this table: studies register outcome names, they register them as primary, secondary or other, and they almost always fill the measure string. What you cannot defend: that the name is a COA type, that the type was selected under Guidance 3, or that an eCOA system was used to capture it.

How rarely the taxonomy appears

Takeaway: ClinRO-pattern studies are 0.347% of the outcome-file denominator. ObsRO 0.224%. PerfO 1.932%. eCOA / e-diary 0.157%. Literal taxonomy tokens 0.047%. You cannot audit COA type from ClinicalTrials.gov.

The patterns are not mysterious, and they are not generous. Each was run against the concatenated measure title plus description on every design-outcome row. A study counts as a hit if any of its rows match. Row hits and study hits are both reported because a three-row PRO battery should not be mistaken for three trials [2].

ClinicalTrials.gov / AACT — EClinCloud analysis, accessed August 2026 [2].

Read the first five rows as the taxonomy conversation. Read the last two as the naming conversation. They are not the same census.

Literal tokens are the honest “does the registry speak Guidance 3?” test. COA, ClinRO, ObsRO, PerfO, ePRO and eCOA are the words the selection file uses. They appear on 589 of 3,716,648 rows — 0.016% of rows — and in 272 studies. That is not a rounding error you can wait out. It is a schema fact. Registrants are not asked for those words, so they almost never type them.

ClinRO language is still rare after you broaden it. “Investigator-assessed” and “physician-reported” are the phrases a protocol author actually uses. They still tag only 2,011 studies, 0.347%. If your mental model is “oncology is full of investigator-assessed response, so ClinRO must be everywhere in the registry,” the file disagrees. Many ClinRO concepts are registered as “overall response rate,” “ECOG performance status,” “NIHSS,” or a disease-specific severity score, with no rater vocabulary at all. The concept can be a ClinRO. The row does not say so.

ObsRO is rarer. Caregiver-, parent- and proxy-reported language tags 1,299 studies, 0.224%. Pediatric, rare-disease and cognitive-impairment programmes that depend on an observer will not be found by grepping “ObsRO.” They will be found, if at all, by reading the protocol.

PerfO is the noisy exception, and the noise is in the pattern. 11,185 studies (1.932%) hit a performance-outcome / 6-minute-walk / TUG / walk-test expression. That is an order of magnitude above ClinRO, and it is not evidence that sponsors prefer PerfO to ClinRO. The pattern includes “walk test.” Six-minute walk and Timed Up and Go are real PerfO instruments. “Walk test” is also ordinary clinical English. Treat 1.932% as an upper bound on named task language, not as a PerfO prevalence.

eCOA is not hidden in the file under a synonym you forgot. The eCOA pattern already includes electronic diary, e-diary, electronic clinical outcome and electronic patient-reported — not only the four-letter acronym. It still reaches only 912 studies, 0.157%. Electronic capture of a PRO can happen without those words. Paper capture of a PRO can happen with a perfect EQ-5D title. The public row does not distinguish them.

None of these absences is a finding against the trial. Guidance 3 is dated 22 October 2025 [1]. The AACT file is a history. A 2015 registration cannot be scolded for omitting a 2025 token. A 2026 registration can omit the token and still have a rigorous COA package in the protocol, the statistical analysis plan and the eCOA specification, because the registry has nowhere to put that package. Keyword absence is not evidence of a missing or invalid endpoint. Keyword presence is not evidence of a validated one.

The operational implication is narrower, and it is the one that changes a build. You cannot derive COA type from a ClinicalTrials.gov scrape. You cannot rank sponsors by how often they type “ClinRO.” You cannot estimate eCOA adoption from the 912-study e-diary pattern. If a library, a dashboard or a due-diligence memo is keyed to those tokens, it is keyed to a field that does not exist.

PRO/QoL language versus instrument names

Takeaway: 16.7% of studies carry PRO/QoL language. 17.0% carry a named-instrument string. Those are not two measurements of “PRO trials,” and the instrument bucket is not a PRO-instrument census. ECOG, NYHA, VAS and NRS sit inside it. PRO hits are mostly secondary: 135,296 versus 37,390 primary.

PRO-language is common enough to tempt the wrong conclusion. 96,440 studies, 16.7% of the outcome-file denominator, match patient-reported, self-reported, ePRO, PROM, PROMIS, EQ-5D, SF-36, SF-12, EORTC, FACT, QLQ-C30, quality of life, HRQoL, diary card or symptom diary [2]. That is two orders of magnitude above the literal taxonomy. It is also not “16.7% of trials use a fit-for-purpose PRO.”

Three compressions are doing the work.

A QoL word is not a COA type. “Quality of life” in a secondary line can be a single VAS, a full EORTC QLQ-C30, a locally developed checklist, or a sentence with no instrument at all. The PRO pattern treats those as one hit class because that is what the text supports. Guidance 3 does not.

Primary versus secondary is the build split the 16.7% hides. Of 184,540 PRO-pattern rows, 135,296 are secondary, 37,390 are primary, and 11,854 are other.

ClinicalTrials.gov / AACT — EClinCloud analysis, accessed August 2026 [2].

As a share of all rows of that type, PRO language is still a minority: 37,390 / 1,174,906 primary rows (3.2%) and 135,296 / 2,335,625 secondary rows (5.8%). Even among secondary outcomes, “patient-reported” is not the default vocabulary. When it does appear, it appears as extra work for the eCOA team, not as proof that eCOA is optional. A secondary PRO still has to be licensed, translated, trained, windowed and captured. It does not have to be the thing the sample size was built on. Collapsing those jobs into “16% of trials use PROs, so eCOA is a nice-to-have” is the before-picture this file breaks.

The instrument bucket is adjacent, overlapping, and inflated by clinician scales. 196,817 rows in 98,575 studies (17.0%) match a list of common instrument strings: EQ-5D, PROMIS, SF-36, SF-12, EORTC, QLQ-C30, FACT-, HADS, PHQ-9, GAD-7, MMSE, MoCA, Karnofsky, ECOG, NYHA, VAS, NRS, Brief Pain Inventory, St George’s, SGRQ, HAMD, Hamilton depression [2]. 17.0% sits next to 16.7% and looks like a confirmation. It is not.

The two patterns share EQ-5D, PROMIS, SF, EORTC and FACT. A study can hit both with one “EQ-5D” title. They are not independent censuses, and they should not be added. More important: the instrument regex includes ECOG, NYHA, VAS and NRS. ECOG performance status and NYHA class are clinician-rated scales. VAS and NRS are used as patient-reported and as clinician-recorded scores. Calling 17.0% “PRO instruments” would launder those clinician scales into a patient-reported share. This article will not do that. The 17.0% is a named-common-instrument bucket with a disclosed clinician-scale inflation. Regex also under-counts instruments that appear only under a local or Chinese name, so the bucket is simultaneously too big (ECOG/NYHA/VAS/NRS) and too small (unnamed or non-English scales). It is a naming census, not an instrument inventory.

An EQ-5D row illustrates the whole trap. PRS QC is happy to see the scale named [9]. The 2009 PRO guidance is happy to review EQ-5D when the claim, the context of use and the evidence line up [6]. Guidance 3 will still ask whether this version, in this population, for this endpoint role, is fit-for-purpose [1]. The public row can satisfy QC and still answer none of those questions. Named is not validated. Named is not versioned. Named is not eCOA.

The action is therefore a classification step that the registry will not perform for you. For each named outcome in the protocol, write down whether it is a PRO, ClinRO, ObsRO, PerfO or not a COA at all. Then attach instrument, version, mode, language, rater and window. Do that in the study build. Do not infer it from a 16.7% headline, and do not skip it because the ClinicalTrials.gov title already says “EQ-5D.”

NIH and literature are different conversations

Takeaway: Europe PMC talks PRO (56,931 title/abstract hits in a 101,685-record eClinical union) and barely talks eCOA (36) or ClinRO (294). NIH’s 31-phrase eClinical corpus holds 4,876 unique projects; the exact phrase “electronic clinical outcome assessment” retrieves 2. “Electronic data capture” retrieves 1,023. “Wearable sensor” retrieves 2,049. Those are query volumes, not adoption, and not a third copy of the registry.

Once the registry has been shown to be a naming surface, the temptation is to look somewhere “richer” — papers, grants — and read hit counts as the missing adoption series. The third and fourth evidence lanes exist to stop that move.

Europe PMC is a PRO conversation with a thin eCOA tail. In an eClinical title/abstract union of 101,685 records, dated 1 August 2026:

Europe PMC — EClinCloud analysis, accessed August 2026 [13][2].

56,931 PRO hits in a corpus that was already scoped to eClinical text is not “56% of biomedical literature is about PROs.” It is “inside this eClinical union, ‘patient-reported outcome’ is common vocabulary.” 36 eCOA hits in the same union is the matching negative. Papers discuss PROs. They rarely use the electronic-capture acronym. 294 ClinRO hits against 56,931 PRO hits is the same split the registry showed, in a different medium: the scientific literature has a PRO language. It does not have a balanced four-type language. 160 “fit-for-purpose” hits are a phrase count, not 160 Guidance 3 dossiers.

Search-set frequency is not adoption. A review article that names PROs once counts the same as a methods paper that implements ePRO. The union is not a random sample of PubMed. It is an eClinical filter. Use it as a contrast to the registry, not as a market-share table.

NIH RePORTER, on a 31-phrase eClinical query set, is a boundary file. 4,876 unique project applications were retrieved. That is not all NIH awards. It is the union of 31 quoted phrases chosen to find eClinical operations language. Inside that union, query totals include:

NIH RePORTER — EClinCloud analysis, accessed August 2026 [14][2].

Two projects against the exact phrase “electronic clinical outcome assessment” is a query boundary, not a finding that NIH ignores eCOA. Investigators write “patient-reported outcome,” “PROM,” “questionnaire,” “symptom diary,” “mobile app.” They do not have to type the five-word phrase this extract searched. The hyphenated and un-hyphenated ePRO queries each retrieved 17; those two result sets can overlap, and they are not added here into a fake 34.

What the table does support is a comparison inside the same 31-phrase net. “Electronic data capture” is ordinary grant English in this corpus (1,023). “Wearable sensor” is large (2,049) because the net is full of sensor methods that are not clinical outcome assessments. That 2,049 is boundary noise if someone briefs it as eCOA funding. A wearable that streams step count may feed a PerfO or a digital biomarker. It is not, by phrase, an eCOA implementation. “Rater training” at 46 is the closest operational neighbour to ClinRO quality in the phrase list, and it is still a small bar.

Do not rank “eCOA adoption” by putting registry tokens, Europe PMC hits and NIH phrase totals on one axis. The registry measures what registrants typed into outcome titles. Europe PMC measures what authors typed into titles and abstracts of an eClinical union. NIH measures what applicants typed into 31 quoted queries. Three conversations. One shared lesson: PRO is the word the world already has. ClinRO, eCOA and fit-for-purpose are not registry fields, and they are not grant-phrase defaults either.

FDA’s own objects have dates. Use the dates. A title inventory of FDA guidances is not a trial census, but it does fix which selection file was in force when. COA- and PFDD-related titles in that inventory include:

• Patient-Reported Outcome Measures: Use in Medical Product Development to Support Labeling Claims — final 9 December 2009 [6]
• Principles for Selecting, Developing, Modifying, and Adapting Patient-Reported Outcome Instruments for Use in Medical Device Evaluation — final 26 January 2022 [15]
• Digital Health Technologies for Remote Data Acquisition in Clinical Investigations — final 22 December 2023 [16]
• Core Patient-Reported Outcomes in Cancer Clinical Trials — final 17 October 2024 [17]
• Patient-Focused Drug Development: Selecting, Developing, or Modifying Fit-for-Purpose Clinical Outcome Assessments (Guidance 3) — final 22 October 2025 [1]
• Patient-Focused Drug Development: Incorporating Clinical Outcome Assessments Into Endpoints for Regulatory Decision-Making (Guidance 4) — draft 5 April 2023 [18][5]

The 2009 PRO guidance is sixteen years older than Guidance 3. DHT remote-data-acquisition is the December 2023 file for sensors, apps and other digital health technologies used to acquire data remotely — a fit-for-purpose discussion about a device-and-mode object, not a synonym for eCOA [16]. Guidance 4, still draft as of this writing, is the endpoints-and-analysis companion: how a selected COA becomes an endpoint FDA can decide on [18]. Core PROs in cancer is a disease-area collection recommendation, not a registry schema [17]. The device PRO principles say, in FDA’s own words, that they are not a replacement for the PFDD series [15].

CDER’s Clinical Outcome Assessment Compendium and the COA Drug Development Tool qualification program are resource catalogs: labeled uses, qualified tools, contexts of use [19][20][21]. A public table of 163 clinical-tool rows, 139 of them COA-ish by string, is not 139 trials and not 139 eCOA studies. It is a catalog. Do not add it to the 272-study token count.

The dates also close a false indictment. Most of the 579,073 studies with outcome rows were registered when Guidance 3 did not exist. They could have used 2009 PRO-guidance vocabulary. They still had no ClinicalTrials.gov field in which to deposit a fit-for-purpose package. Scolding the file for missing 2025 tokens is a category error. Using the 2025 selection file, now that it is final, and putting the resulting specification in the eCOA build, is the current job.

What to put in the eCOA specification because ClinicalTrials.gov will not

Takeaway: The public row will not carry COA type, instrument version, administration mode, language, rater or a real visit window. Write those six fields into the protocol and the eCOA / scale-management specification before screens are built. That is an implementation file. It is not a claim that a named product makes an endpoint succeed.

The gap above is the specification. Each field below is something Guidance 3 expects you to have decided and something the registry has nowhere to keep.

Classify each protocol outcome against that table during study build. A laboratory value does not belong in the eCOA library. A clinician-rated ECOG belongs in a ClinRO configuration, not in a patient app, even if the instrument regex tagged it. A secondary EQ-5D still needs a version, a mode and a language pack. A primary PRO needs those plus a clear link to the estimand. The registry will not make those calls. The specification will.

Two merges this file does not support.

eCOA is not EDC. EDC holds the casebook. eCOA holds the outcome assessment as administered to a patient, an observer or a clinician — on a device, a browser or paper transcribed under control. NIH’s own phrase volumes in the eClinical corpus already treat “electronic data capture” as a large, ordinary term (1,023) and “electronic clinical outcome assessment” as a rare one (2) [14]. Do not collapse them because both start with “electronic.”

eCOA is not independent imaging review. IRC is a different endpoint object: blinded, centralized reading of scans. We wrote that argument separately [22]. A COA specification that quietly absorbs imaging review, or an imaging charter that quietly absorbs ePRO, will fail both jobs.

The implementation layer sits after the gap, not instead of it. Once type, instrument and version, mode, language, rater and window are written down, they have to be configured, licensed, translated, trained, provisioned and monitored. EClinCloud eCOA is built to capture ePRO, eClinRO, eObsRO, ePerfO and eDiary against that kind of specification; scale management, linguistic validation and rater training sit next to the configuration [23][24]. Naming a vendor does not make a COA fit-for-purpose. Skipping the specification and scraping ClinicalTrials.gov for “EQ-5D” does not either. Connected data and computerized-system governance matter for the capture layer, as ICH E6(R3) has already made the operating-model point [25]. They still do not replace Guidance 3’s selection file.

No endpoint-success claim follows from any number in this paper. A complete eCOA specification can be executed badly. An incomplete specification can sit behind a clean NCT page. The public file cannot tell those studies apart. That is the finding.

Frequently asked questions

Takeaway: A missing ClinRO token is not a failed endpoint. A named EQ-5D is not validation. Guidance 3 is October 2025. NIH “wearable sensor” counts are boundary noise. eCOA is a mode; PRO is a rater type.

If a study has no ClinRO token on ClinicalTrials.gov, is the endpoint invalid?

No. Literal ClinRO/ObsRO/PerfO/COA/ePRO/eCOA tokens appear in 272 of 579,073 studies with an outcome row (0.047%). The broader ClinRO pattern still tags only 0.347% [2]. ClinicalTrials.gov has no COA-type data element [3]. Absence of the word is the expected state of the file, including for studies that used a rigorous clinician-reported measure and for studies registered years before Guidance 3. Invalid endpoints are a scientific and regulatory judgement. They are not a grep.

The outcome title says EQ-5D. Is that a validated PRO?

It is a name. PRS QC wants a specific measurement in the title [9]. The 2009 PRO guidance and Guidance 3 both treat validation as context-of-use evidence, not as a brand string [6][1]. The public row does not record version (3L versus 5L), module, mode, language, scoring, or whether the instrument is fit for this population. Named is not validated.

Why are you not criticising 2015 trials for omitting “eCOA” and “fit-for-purpose”?

Guidance 3 final is 22 October 2025 [1]. Most of the AACT history predates it. The 2009 PRO guidance is the older object, and even it never created a ClinicalTrials.gov field for the evidence package [6][3]. This paper measures what the public record can hold. It does not grade historical registrations against a 2025 vocabulary.

NIH retrieved 2,049 projects for “wearable sensor.” Is that eCOA funding?

No. The NIH lane is 4,876 unique projects from 31 quoted eClinical phrases, not the NIH portfolio [14]. “Wearable sensor” is a wide methods phrase. A sensor study may have nothing to do with a clinical outcome assessment. In the same net, “electronic clinical outcome assessment” retrieved 2 projects and “electronic data capture” retrieved 1,023. Use those as query-boundary contrasts, not as an adoption ranking and not as a wearable-equals-eCOA brief.

Is eCOA the same thing as a PRO?

No. PRO is a rater type: the patient reports, unfiltered by a clinician [4][1]. eCOA is an electronic mode of capturing a COA, which may be ePRO, eClinRO, eObsRO, ePerfO or an electronic diary. A paper PRO is still a PRO. An eClinRO is not a PRO. The registry’s 16.7% PRO-language share and 0.157% eCOA / e-diary share are different patterns. Do not brief them as one number.

Can I estimate how often trials use eCOA from ClinicalTrials.gov?

Not from this file. The eCOA / electronic-diary pattern hits 912 studies (0.157%) [2]. Electronic capture can happen without those words. Paper capture can happen with a perfect instrument name. Mode is not a protocol-registration data element [3]. Treat 0.157% as a token frequency, not as an implementation census.

Is the 17% “instrument” share a count of PRO instruments?

No. The instrument regex includes ECOG, NYHA, VAS and NRS — clinician scales as well as instruments often used as PROs — and it overlaps the PRO-language pattern on strings such as EQ-5D and EORTC [2]. 17.0% is a named-common-instrument bucket with disclosed inflation. It is not “17% of trials use a PRO instrument.”

Should we add COA type to the public ClinicalTrials.gov row?

That is a policy question for NLM and FDA, not a finding this extract can settle. What the extract shows is that today’s schema does not hold type, version, mode, language, rater or fit-for-purpose evidence [3]. Until that changes, the operational file is the protocol plus the eCOA / scale-management specification, not a registry scrape.

Methodology and limitations

Takeaway: Four independent lanes, one ClinicalTrials.gov grain. Regex both under-counts and over-counts. Guidance 3 is dated 2025. CTIS is a one-line contrast at a different grain.

• Lane 1 — ClinicalTrials.gov via AACT design_outcomes. Static AACT copy dated 1 August 2026, 3,716,648 protocol-specified outcome rows, 579,073 studies with at least one row. Study-level join to the ClinicalTrials.gov study file current as of 25 July 2026 (595,630 NCT identifiers). Patterns were applied to concatenated measure title plus description. Outcome type used the design table’s primary / secondary / other flag. ClinicalTrials.gov / AACT — EClinCloud analysis, accessed August 2026 [2][7][3]. Nested ClinicalTrials.gov JSON and the AACT results-module outcomes table were inspected as duplicate or later naming surfaces and were not used as a second registry lane.
• Lane 2 — FDA PFDD / COA guidance objects. Official FDA.gov landings and PDFs for Guidance 3 (final 22 October 2025), the 2009 PRO labeling guidance, DHT remote data acquisition (final 22 December 2023), Guidance 4 (draft 5 April 2023), core PROs in cancer (final 17 October 2024), device PRO principles (final 26 January 2022), the COA Compendium, DCOA and the COA qualification program [1][6][16][18][17][19][20][21][15]. An FDA.gov title inventory of guidances and a clinical-tools table (163 rows, 139 COA-ish by string) were used only to locate those public objects. They are not a trial census and are not cited as one.
• Lane 3 — Europe PMC eClinical union. 101,685 records, snapshot 1 August 2026. Title/abstract regex counts as tabulated. Search-set frequency is not adoption [13].
• Lane 4 — NIH RePORTER eClinical phrase corpus. 31 quoted queries; 4,876 unique project applications; snapshot 23 August 2026. Query totals are retrieval counts, not unique-project sums across phrases, and not the NIH portfolio [14].
• CTIS contrast (not mixed into AACT shares). 2,215 / 12,123 public trial summaries blank on primary/endpoint text, from a companion August 2026 compute. Different grain from AACT study-level shares [11].
• Regex limitations. Instrument strings under-count local and Chinese names. They over-count ECOG, NYHA, VAS and NRS inside the “instrument” bucket. The PerfO pattern includes “walk test” and is an upper bound. PRO-language on a secondary line is not an eCOA deployment. Patterns are not nested; shares must not be divided across sets as if they were subsets.
• Timing. Guidance 3 is October 2025. Historical registrations cannot be scored against 2025 tokens.
• What we did not claim. Missing tokens do not mean invalid endpoints. Named instruments do not mean validated instruments. Literature and NIH hit counts do not mean market share. Company product metrics and endpoint-success rates are out of scope.

Conclusion

Takeaway: Guidance 3 is the selection file. ClinicalTrials.gov is a naming surface. Put type, version, mode, language, rater and window in the eCOA specification — the public row will not hold them.

Four takeaways survive the file.

Fit-for-purpose is not a registry field. Guidance 3, final 22 October 2025, is the selection standard [1]. ClinicalTrials.gov stores a title, a description and a time frame [3]. Literal taxonomy tokens appear in 272 / 579,073 studies (0.047%) [2].

The outcome table is a naming surface, and it is full. 3,716,648 rows; primary 1,174,906; secondary 2,335,625; other 206,117; measure blank 9. PRO-language in 16.7% of studies, mostly on secondary rows (135,296 versus 37,390 primary), is a naming frequency. The 17.0% instrument bucket includes clinician scales. Neither number is a validation census [2].

Papers, grants and guidances are other conversations. Europe PMC’s eClinical union talks PRO (56,931) and not eCOA (36) or ClinRO (294). NIH’s 31-phrase net retrieves 2 projects for “electronic clinical outcome assessment,” 1,023 for “electronic data capture,” and 2,049 for “wearable sensor” — a boundary, not a league table [13][14]. DHT remote acquisition is a December 2023 mode-and-device file. Guidance 4 remains draft.

Write the specification the registry cannot store. Type, instrument and version, mode, language, rater, visit window. Then configure them in eCOA and scale management. Do not scrape ClinicalTrials.gov for a ClinRO token. Do not treat a named EQ-5D as the evidence package. Do not merge eCOA with EDC or with independent imaging review. No public row, and no vendor name, makes an endpoint succeed.

The Insight remains the place to learn the four types [4]. This paper is the place to stop expecting the public record to show whether you used them.

Sources

1. U.S. Food and Drug Administration, Patient-Focused Drug Development: Selecting, Developing, or Modifying Fit-for-Purpose Clinical Outcome Assessments (Guidance 3), final guidance, October 2025 (issue date 22 October 2025; page current as of 23 October 2025). PDF: https://www.fda.gov/media/159500/download. Docket FDA-2022-D-1385.

2. ClinicalTrials.gov / AACT — EClinCloud analysis, accessed August 2026. Protocol-specified outcome text from the 1 August 2026 AACT design_outcomes copy (3,716,648 rows; 579,073 studies with any outcome row), joined to the ClinicalTrials.gov study file current as of 25 July 2026. Pattern definitions as tabulated in the article. Not a census of endpoint quality, eCOA adoption, or fit-for-purpose evidence.

3. ClinicalTrials.gov, Protocol Registration Data Element Definitions for Interventional and Observational Studies — Outcome Measure Title, Description and Time Frame; primary, secondary and other outcome measures. National Library of Medicine, accessed August 2026.

4. EClinCloud, From regulatory intent to practice: choosing a fit-for-purpose clinical outcome assessment, 30 October 2025 — four COA types and the fit-for-purpose selection argument. This Deep Research article does not retell that Insight as its thesis.

5. U.S. Food and Drug Administration, FDA Patient-Focused Drug Development Guidance Series — Guidance 3 (COA selection) and Guidance 4 (COA-based endpoints); page current as of 23 October 2025.

6. U.S. Food and Drug Administration, Patient-Reported Outcome Measures: Use in Medical Product Development to Support Labeling Claims, guidance for industry, December 2009 (issue date 9 December 2009). PDF: https://www.fda.gov/media/77832/download. Docket FDA-2006-D-0362.

7. Clinical Trials Transformation Initiative, AACT Database — publicly available relational copy of ClinicalTrials.gov protocol and results data elements, with static downloads. Accessed August 2026.

8. ClinicalTrials.gov, Home — U.S. National Library of Medicine public registry of clinical studies. Accessed August 2026.

9. ClinicalTrials.gov, Protocol Registration Quality Control Review Criteria — outcome-measure title should name the specific measurement (scale, parameter or questionnaire) and the aggregation metric. Accessed August 2026.

10. ClinicalTrials.gov, Clinical Trial Reporting Requirements — FDAAA 801 and 42 CFR Part 11 registration and results-reporting duties, including outcome measures for applicable clinical trials. Accessed August 2026.

11. EU Clinical Trials Information System (CTIS) public summaries — EClinCloud analysis, accessed August 2026: 2,215 of 12,123 public trial summaries blank on primary/endpoint text. Different grain from AACT study-level shares; not mixed into those denominators.

12. ClinicalTrials.gov, Results Data Element Definitions for Interventional and Observational Studies — protocol outcome information copied into the results module; post-hoc measures may also be reported. Accessed August 2026. This article analyses protocol-specified design outcomes, not posted results.

13. Europe PMC — EClinCloud analysis of an eClinical title/abstract union (101,685 records, 1 August 2026 snapshot), accessed August 2026. Search-set frequency is not adoption.

14. NIH, RePORTER — EClinCloud analysis of a 31-phrase eClinical query corpus (4,876 unique project applications, 23 August 2026 snapshot), accessed August 2026. Phrase retrieval is not the NIH portfolio and not vendor adoption.

15. U.S. Food and Drug Administration, Principles for Selecting, Developing, Modifying, and Adapting Patient-Reported Outcome Instruments for Use in Medical Device Evaluation, 26 January 2022. The guidance states that it is not meant to replace the PFDD series. Docket FDA-2020-D-1564.

16. U.S. Food and Drug Administration, Digital Health Technologies for Remote Data Acquisition in Clinical Investigations, final guidance, December 2023 (issue date 22 December 2023). PDF: https://www.fda.gov/media/155022/download. Docket FDA-2021-D-1128.

17. U.S. Food and Drug Administration, Core Patient-Reported Outcomes in Cancer Clinical Trials, guidance for industry, October 2024 (issue date 17 October 2024). Docket FDA-2020-D-2303.

18. U.S. Food and Drug Administration, Patient-Focused Drug Development: Incorporating Clinical Outcome Assessments Into Endpoints for Regulatory Decision-Making (Guidance 4), draft guidance, April 2023 (issue date 5 April 2023). Series context: [5].

19. U.S. Food and Drug Administration, Clinical Outcome Assessment Compendium — catalog of labeled and qualified COAs by review division; not a trial census. Accessed August 2026.

20. U.S. Food and Drug Administration, Division of Clinical Outcome Assessment (DCOA). Accessed August 2026.

21. U.S. Food and Drug Administration, Clinical Outcome Assessment (COA) Qualification Program — qualification is a context-of-use conclusion, not a registry field. Accessed August 2026.

22. EClinCloud, Why decisive imaging endpoints need an independent second look, 24 June 2026 — IRC / central imaging review is a different endpoint object from eCOA.

23. EClinCloud, Electronic Clinical Outcome Assessment (eCOA) — implementation layer for ePRO, eClinRO, eObsRO, ePerfO and eDiary after the specification exists. No endpoint-success claim.

24. EClinCloud, Services — Scale Management — instrument/scale management, with linguistic validation and rater training as companion services. Distinct from EDC and from independent imaging review.

25. EClinCloud, ICH E6(R3): Clinical trials enter the era of risk-driven digital governance, 10 March 2026 — computerized-system and data-governance operating model; not a COA-selection substitute.