The One-Word Test for Abandoned Clinical Trials

Two Phase 3 trials say they are recruiting, and both are years past their own completion date. One record is maintained. The other has been silent since 2014. A single word on the page tells you which is which.

August 16, 2026 · Data from ClinicalTrials.gov

Two Trials, Both Listed as Recruiting

NCT04951622 (registry record) is a Janssen Phase 3 study in myasthenia gravis. Its deadline for finishing primary data collection passed almost three years ago, but the record is well maintained: 63 versions, posted results, the latest change in July 2026.

NCT01185132 (registry record) is a Phase 3 breast cancer study from Rocky Mountain Cancer Centers. Its deadline passed in July 2018. The record was last edited in June 2014.

It has said "Recruiting" for twelve years without a single edit.

The same status line, seen from the version history
NCT04951622 · Janssen · "Recruiting" 63 record versions · last edited July 27, 2026
registered Jul 2021 still being edited ▶ completion date Nov 2023 · ACTUAL
Completion date Nov 2023, tagged ACTUAL. Still being edited.
NCT01185132 · Rocky Mountain · "Recruiting" 6 record versions · last edited June 26, 2014
registered Aug 2010 NO EDITS · 12 YEARS completion date Jul 2018 · ESTIMATED
Completion date Jul 2018, tagged ESTIMATED. No edits in 12 years.
record being edited    no edits at all    |the completion date the sponsor set, with its registry tag

On ClinicalTrials.gov the two records carry the same status line, and nothing in the status itself says which trial is alive and which is abandoned.

They have plenty of company. Nearly three in ten trials currently listed as recruiting are past their own deadline: 13,517 of 47,419, or 28.5 percent (interventional trials, ClinicalTrials.gov, August 16, 2026).

The deadline in question is the primary completion date: the day the sponsor said they would finish collecting data for the trial's main question. These trials are still enrolling patients after that day came and went.

This breaks no rule. A completion date is an estimate, and sponsors only owe the registry one update a year. But it leaves the record saying two things that cannot both be true, and it leaves you guessing which trials are late and which are dead.

One Word Tells Them Apart

Every completion date on ClinicalTrials.gov carries a one-word tag: Estimated or Actual.

That tag separates the maintained records from the abandoned ones:

Recruiting trials with a past ESTIMATED completion date: untouched since the date passed?
97 of every 100 are untouched since the date passed 12,403 of 12,789 trials
The 3 open squares stand for the 386 edited records:
262 lapsed within 12 months
122 in 12 to 24 months
2 older than 24 months
Each square is 1 of every 100 recruiting trials whose estimated completion date has passed. Hatched = untouched since the date passed; open = edited after it. ClinicalTrials.gov, recruiting interventional trials, August 16, 2026.
Control: the 728 recruiting trials whose past date is tagged Actual include 0 untouched records. Zero here is definitional: a date becomes Actual only through an edit made after the date.

The Actual group is a control, not a second discovery: a past date becomes Actual only when someone edits the record after the date, so it must count zero untouched records if the tag means what it says, and it does (0 of 728). The finding is the Estimated bar.

A past estimated date on a recruiting record means, 97 times out of 100, that nobody has touched the record since the date passed. The 386 exceptions are nearly all recent lapses: two thirds of those dates passed within the last year, 99 percent within the last two, which is what the once-a-year update duty noted above would predict. If the date is estimated and well behind you, you are reading the last thing anyone ever wrote there.

Researchers have documented stale registry records for a decade: 24% of open-recruitment trials already listed a past completion date in 2014 (Pfiffner et al., PLOS ONE), and a 2017 audit found 31% of trials with a wrong or badly outdated status (Jones et al., BMJ Open). What those studies did not offer is a way to tell, from the record alone, which stale trials are dead. The date tag closes most of that gap, and it is on every study page already.

When a Deadline Moves, It Usually Moves Later

Deadlines still in the future deserve the same skepticism. We watched every completion-date change our pipeline caught between March 27 and August 7, 2026: about 79,000 monitored trials for most of that window, until a one-time backfill in late July extended coverage to every active trial on the registry, 150,000+ records in all. Every rate below also holds inside the constant cohort of 79,288 trials monitored for the entire window (details in the methodology note). When a date moved, it moved later three times out of four, by a typical five months:

Pushed later

74%
of all date changes
Median push: +153 days
vs

Pulled earlier

26%
of all date changes
The minority case
If you are modeling a competitor's readout, do not treat the registry date as a midpoint. It moves later far more often than earlier, and it can move more than once.

The largest pushes we caught were on big Phase 3 industry programs, each still recruiting:

SponsorTrialCompletion date movedDelay
Loxo Oncology (Lilly) NCT05254743
Pirtobrutinib vs ibrutinib, CLL
Jun 2025 → Oct 2027+843 d
AbbVie NCT06012240
Upadacitinib, pediatric
Jan 2028 → Apr 2030+821 d
AstraZeneca NCT06129864
Volrustomig (MEDI5752)
Apr 2028 → Apr 2030+729 d
Seagen (Pfizer) NCT05253651
Tucatinib combo, colorectal
Apr 2026 → Dec 2027+637 d
Anthos Therapeutics NCT05712200
Abelacimab, atrial fibrillation
Aug 2026 → Dec 2027+486 d

Each verified against the live registry record. A date change is a documented sponsor action, not evidence of a problem with the drug.

We expected Phase 3 industry trials to slip more often than everything else. They do not. Once we controlled for how long each trial had been under observation, they slip at about the same rate as the rest of the registry; the methodology note has the details.

Enrollment Targets Get Cut in Half

The second signal is quieter than a moved date. When a sponsor cuts an enrollment target, the median cut removes roughly half the planned patients.

Sponsors do not trim enrollment; they halve it. A team that cannot recruit does not shave 10% off and try again. That makes a downward enrollment revision one of the most informative events on a record, and one of the easiest to miss: a single number changing in a field few people read.

The three warning signs (a pushed date, a cut target, a changed endpoint) rarely appear together. Our Trials at Risk scan currently counts 4,706 active trials showing at least one signal in the past 90 days, 310 showing two, and 22 showing all three; how the signals are computed and ranked is in the launch post.

It Is All in the Version History

All of this is public. The registry keeps every version of every record, but the history sits behind a tab, one trial at a time, and the study page always shows the latest version. A date that has moved three times looks identical to a date that has never moved. In July we published a correction to our launch post for wrongly claiming the registry discards old versions; this analysis is built on that same version history, and every figure in it can be re-derived from the registry's own records.

Meanwhile, the records go quiet. Among trials that say they are recruiting today, the median record has not been edited in 192 days, and three in ten have gone more than a year. "Recruiting" means the sponsor said so once, on a date you have to dig for.

How to Use This

See which of the trials you track are slipping

DataLookout monitors ClinicalTrials.gov every morning and flags the three signals in this post on the sponsors and diseases you follow. Every trial page shows the record's last-update date, its amendment history, and a Record status note when a record is untouched since its own past completion date. Free forever for one sponsor and one disease, no credit card.

Start monitoring free

Trials at Risk, the ranked per-trial slippage view behind this analysis, is included with Starter at $49/mo for 3 sponsors and 3 diseases. Pro $149/mo adds unlimited watchlists and CSV export.

Data and methodology. Two distinct instruments, each measuring something different.

(1) Registry-wide counts. One rule governs every past-due figure in this post: a completion date counts as past only when its entire stated period is over, so a day-precision date is past after that day and a month-precision date only after the month's last day. Under that rule, of the 47,419 recruiting interventional trials queried from the ClinicalTrials.gov v2 API on August 16, 2026, 13,517 (28.5%) are past their primary completion date; among active-not-recruiting trials the share is 58.3% (9,944 of 17,050). The partition is clean: exactly one recruiting record lacks a primary completion date, so the denominator is not inflated by records missing dates. The Actual/Estimated split (728 + 12,789, summing exactly to 13,517) uses the registry's own date-type field. The 97.0% figure (12,403 of 12,789) compares each record's last-update-posted date against its own primary completion date; a record counts as untouched only if its last update is on or before the earliest day its completion date could mean, so month-precision ambiguity counts against the finding and the figure is a floor. We verified the split two ways on August 9: 30 random draws from the estimated past-due population, each re-fetched individually (30 of 30 confirmed untouched since the date passed), and an inverse check drawing 30 recently-updated past-due recruiting trials (29 carried Actual dates, one had moved its estimated date into the future, none carried a past estimated date). The zero on the Actual side is a control, not a finding: a past date becomes Actual only through an edit made after the date, so that column tests whether the tag behaves as documented, and it does (0 of 728). Of the 386 estimated-date exceptions, 262 (68%) have dates that passed within the last twelve months and 384 (99%) within the last twenty-four. Record-staleness figures (median 192 days since update; 51.6% over six months, 30.4% over one year, 7.1% over two) cover the 47,419 trials listed as recruiting on August 16, fetched with ages measured as of August 16.

(2) Change counts. The date-change direction (74% later, 4,787 of 6,444 changes; median push 153 days; 17.0% of pushes longer than a year) and the enrollment finding (1,247 downward revisions, median cut 47% of target) come from DataLookout's own daily comparison of every tracked record (interventional and observational), March 27 to August 7, 2026. These measure movement we observed in that window, not lifetime slippage. The signal-stack figures (4,706 / 310 / 22) reflect the rolling 90-day Trials at Risk scan as of August 16, 2026.

The corpus grew mid-window, and we checked what that does to the numbers. In late July a one-time backfill roughly doubled our monitored universe, from about 79,000 trials to full coverage of every active trial on the registry (150,000+ records, including trials that finished while we watched). The added half was, by construction, the half with no recent record activity, and it entered with a suppressed baseline that produced no spurious events. The counts above therefore understate registry-wide movement. Rates are the bigger risk, so we recomputed every percentage inside the constant cohort of 79,288 trials monitored since at least June 18: the 74% pushed-later share holds exactly, the 47% median enrollment cut holds, and the median push is 140 days there against 153 overall. One claim did not survive that check. At first glance, recruiting Phase 3 industry trials had dates pushed at a higher rate than the full monitored set (5.6% of 1,707 versus 3.6% of 65,780 in the window), but industry trials were overrepresented in the half of the registry we watched longest; inside the constant cohort the rates are 7.4% versus 6.9%, computed over recruiting trials only, so we do not claim late-stage trials slip more. These figures are also not comparable to our June 7 launch post (2,443 flagged, 254 enrollment cuts): smaller corpus, shorter window, and several change-detection fixes since.

On sponsor obligations. Under 42 CFR 11.64, registration information generally must be updated once every 12 months; recruitment-status changes require updates within 30 days, but a primary completion date requires a 30-day correction only once the trial has actually completed. A past-due estimate is therefore consistent with the rules, which is part of why so many exist.

Known limitation. The extreme tail of completion-date pushes contains sponsor data-entry errors; the single largest push we recorded moved a date to the year 2050. All push statistics in this post are medians and percentiles for that reason. The named examples were individually verified against the live registry record (the delay table and the two opening trials, including their version counts of 63 and 6, on August 16, 2026). The timeline diagram shows each record's first-posted date, last-edit date, and stated primary completion date, all from the live registry. Registry-wide counts are current as of August 16, 2026 and will drift.

External sources cited: Jones CW, Safferman MR, Adams AC, Platts-Mills TF. Discrepancies between ClinicalTrials.gov recruitment status and actual trial status: a cross-sectional analysis. BMJ Open 2017;7(10):e017719 (BMJ Open); Pfiffner PB, Oh J, Miller TA, Mandl KD. ClinicalTrials.gov as a Data Source for Semi-Automated Point-Of-Care Trial Eligibility Screening. PLOS ONE 2014;9(10):e111055 (PLOS ONE).

Prior work: our Trials at Risk launch post describes how the three slippage signals are computed and ranked.
Previous
We Can Now Show You Which Trials Are at Risk