Why One in Three Phase 2 Oncology Trials Stop Early — And What Protocol Designers Can Learn
Most people assume clinical trials fail because the drug doesn't work. The data tells a different story.
We analyzed 22,704 Phase 2 oncology interventional trials registered on ClinicalTrials.gov that have reached a definitive status — completed, terminated, or withdrawn. Of those, 5,027 were terminated early. When we examined the documented reasons for termination, the leading cause wasn't toxicity. It wasn't lack of efficacy. It was slow accrual.
36.1% of documented trial terminations in Phase 2 oncology happened because teams couldn't enroll patients fast enough.
That's not primarily a drug problem. It often reflects assumptions embedded in protocol design — particularly around enrollment feasibility, eligibility criteria stringency, and competition for eligible patients.
The Termination Landscape
Across our corpus, 22.1% of Phase 2 oncology trials with definitive outcomes were terminated early. Among the 5,941 trials that documented a reason for stopping (not all terminated trials report a reason), the breakdown challenges common assumptions about why trials fail:
Slow accrual: 36.1%. The single largest category. These trials had functioning sites, approved protocols, and available drug supply — but the patient population either didn't exist in the numbers assumed, faced competing trials, or had eligibility criteria too narrow to support the enrollment plan.
Consider NCT02967133 — a Phase 2 study of nivolumab with nab-paclitaxel in NSCLC. It enrolled 7 patients before terminating due to lack of enrollment. Or NCT03257722, a pembrolizumab combination study in lung cancer that enrolled 4 patients before stopping for poor accrual. These aren't obscure compounds or implausible hypotheses — they're checkpoint inhibitor combinations in one of oncology's most active areas. The drugs weren't the primary constraint. Enrollment assumptions appear to have played a central role.
Sponsor decision: 20.0%. Strategic and business decisions — portfolio reprioritization, program discontinuation, mergers, funding changes. These aren't protocol design failures, but they represent one in five terminations.
Lack of efficacy: 6.0%. The reason most people assume dominates. It doesn't. Fewer than one in sixteen documented terminations cited efficacy failure as the primary cause.
Toxicity and safety: 4.9%. Similarly small. Safety-driven terminations, while critical when they occur, account for a fraction of the total.
Which Indications Struggle Most
Termination rates aren't uniform across oncology. Some indications consistently lose more trials than others:
| Indication | Termination Rate | Trials |
|---|---|---|
| Small cell lung cancer | 28.9% | 180 |
| Non-small cell lung cancer | 27.1% | 1,189 |
| Melanoma | 26.5% | 868 |
| Renal cell carcinoma | 24.3% | 460 |
| Bladder / urothelial | 24.1% | 319 |
| Breast cancer | 23.7% | 2,324 |
Compare that with the lower end:
| Indication | Termination Rate | Trials |
|---|---|---|
| Gastric cancer | 15.8% | 374 |
| Mesothelioma | 18.2% | 110 |
| Sarcoma | 18.9% | 492 |
| Colorectal cancer | 19.9% | 1,201 |
NSCLC and SCLC trials terminate at nearly twice the rate of gastric cancer trials. If slow accrual is the dominant termination driver, these indications likely face the most intense competition for eligible patients — a factor teams can assess before finalizing a protocol.
The Sponsor Gap Nobody Talks About
One of the more unexpected patterns: NIH-sponsored trials outperform industry on completion.
| Sponsor Type | Completed | Terminated | Results Posted |
|---|---|---|---|
| NIH | 78.3% | 17.7% | 63.7% |
| Network | 78.8% | 14.6% | 34.0% |
| Academic (OTHER) | 69.5% | 22.0% | 40.7% |
| Industry | 70.0% | 24.6% | 49.1% |
Industry-sponsored trials terminate at 24.6% — nearly 7 percentage points higher than NIH. The gap likely reflects different decision calculus: industry sponsors terminate early when commercial viability shifts, while NIH-funded trials are more insulated from portfolio-level business decisions. But the results posting gap is striking: NIH posts results in 63.7% of trials versus 49.1% for industry. Whatever the cause, the data suggests that sponsor type is a meaningful predictor of whether a trial will produce publicly available evidence.
What This Means for Protocol Design
This analysis doesn't predict whether your trial will succeed. But it surfaces patterns worth considering before a protocol is finalized:
Stress-test your enrollment assumptions against current trial competition and realistic screen failure rates before protocol lock. If one in three Phase 2 oncology terminations happen because of slow accrual, then enrollment feasibility isn't a logistics question to solve after protocol lock — it's a design decision that deserves the same rigor as endpoint selection. How many competing trials are recruiting similar patients right now? What is the realistic screen failure rate for your eligibility criteria? What does the historical enrollment trajectory look like for trials in your indication?
Know your indication's base rate. If you're designing an NSCLC Phase 2, you're entering an indication where 27.1% of trials with definitive outcomes were terminated. That doesn't mean yours will be — but it means your protocol rationale should address why your enrollment plan will succeed where a quarter of predecessors didn't.
Consider what happens after completion. Completing a trial isn't the finish line — posting results is. The gap between clinical (49.4%) and surrogate (44.7%) endpoint trials in results posting rates is worth factoring into endpoint selection. A completed trial that never posts results produces no public evidence.
Methodology
This analysis is based on ClinicalTrials.gov registry data accessed via the v2 API in April 2026. The corpus includes 39,818 Phase 2 oncology interventional trials, of which 22,704 have reached a definitive status (completed, terminated, or withdrawn).
Endpoint classification uses keyword matching against primary outcome measure text, achieving a 70.2% classification rate. "Why Stopped" categorization uses keyword matching against the free-text field populated by study sponsors and is approximate — 28.5% of documented stop reasons fell into an "other" category that resists simple classification. We manually reviewed a sample of 20 randomly selected classified trials and confirmed that slow accrual signals were directionally consistent, with the primary classification gap being undercount of sponsor-driven decisions in the "other" category. Classification keywords were expanded based on this review.
This is registry data only. It does not include FDA review documents, published outcomes, or protocol amendment history. These additional data sources are part of a planned expansion.
All findings describe historical patterns and frequencies. They do not make causal claims or predict individual trial outcomes.
What We're Building
This analysis is the first output from ClinStacks Protocol Risk Scan — a decision-validation system for clinical trial design. We're building a tool that lets medical directors and biostatisticians validate their protocol design choices against historical precedent before finalization.
If you're designing a Phase 2 oncology protocol, we can run this analysis on your specific design — comparing your endpoint choice, enrollment assumptions, and indication against similar historical trials. We'll show you what teams in comparable situations encountered and how they responded.