Terminated Trials: Stated Reasons, Classification Overlap, and Phase-Specific Enrollment

ClinicalTrials.gov contains 34,123 records with terminated status. A stated reason appears on 30,598 records, while 3,525 contain no reason text. A precedence-based keyword classification assigns 9,193 reasons to recruitment, 3,636 to funding or business, 1,380 to safety, 1,089 to futility or efficacy, and 985 to Covid-19. Another 14,315 remain other or unclassified. Keyword categories overlap before precedence is applied. Phase 2 interventional studies carry an 11.1 percent terminated share, compared with 8.3 percent in Phase 3. Within every main phase, terminated studies report lower median enrollment than completed studies.

1. Cohort and status rules

The cohort includes all study records carrying terminated status at extraction. Terminated is a registry status and does not include withdrawn, suspended, or unknown records. The reason field is sponsor supplied and optional. One record is one registered study, not one drug program or one participant. Phase-specific percentages use all interventional records in each phase as the denominator. Enrollment comparisons include positive numeric values only.

2. Reason availability precedes classification

Stated reason text appears on 89.7 percent of terminated records. The remaining 10.3 percent contain no reason. Classification performance must use the 30,598 records with text, while any statement about all terminated studies must retain the 3,525 missing reasons. Missing reason text is not evidence for an unclassified operational cause.

3. Recruitment is the largest classified category

The classification checks recruitment first, followed by funding or business, safety, Covid-19, and futility or efficacy. Recruitment receives 9,193 records under that precedence. Funding or business receives 3,636, safety receives 1,380, futility or efficacy receives 1,089, and Covid-19 receives 985. Other or unclassified text remains the largest category with 14,315 records, or 46.8 percent of stated reasons.

4. Keyword concepts overlap before precedence

Some reason statements contain several concepts. Recruitment and Covid-19 terms co-occur on 641 records. Recruitment and funding terms co-occur on 563. Safety and futility terms co-occur on 435, while recruitment and safety co-occur on 274. A single-category chart therefore depends on the declared precedence. Reordering the checks would move records between categories without changing the source text.

5. Phase counts and termination shares answer different questions

Phase 2 has 7,229 terminated records, the largest count among declared phases. It also has an 11.1 percent terminated share among 65,203 Phase 2 interventional records. Combined Phase 1/Phase 2 records have an 11.4 percent share. Phase 3 records have an 8.3 percent share, Phase 1 records have 8.1 percent, and Phase 4 records have 7.6 percent. Raw termination counts must not substitute for phase-specific rates.

6. Enrollment differs within every main phase

Terminated records report lower median enrollment than completed records within each main phase. Phase 1 medians are 15 and 30. Phase 2 medians are 17 and 60. Phase 3 medians are 87 and 245. Phase 4 medians are 22 and 82. The comparison reduces phase composition bias but still mixes indications, designs, sponsors, and planned versus actual enrollment values.

7. Supported interpretation

The registry supports three conclusions within the classification rules. Recruitment terms form the largest assigned reason category, while nearly half of stated reasons remain other or unclassified. Category counts depend on precedence because concepts overlap. Phase 2 has both the largest declared-phase termination count and a higher terminated share than Phase 3. Lower enrollment among terminated records remains after phase stratification, but the data do not establish causation or prevent planned-enrollment differences.

8. Limitations

Sponsors can revise status and reason text after initial registration. Keyword rules can miss synonyms and assign ambiguous phrases. No manual adjudication was applied. Enrollment can be anticipated or actual, and missing or zero values are excluded from the phase comparison. A registered termination does not establish failure of a molecule, endpoint, or scientific hypothesis. Counts can change as records are added or updated.