Clinical Development

Oncology trial protocol design: lessons from failed endpoints

There is a number we don't say out loud often enough in our industry: roughly half of every clinical trial we launch never reaches its primary endpoint.

Oncology trial protocol design: lessons from failed endpoints

Not because the science was wrong, not because the molecule failed, but because the protocol — the document we wrote to ask the question — quietly undermined the answer. When we sit with that figure, the abstract becomes uncomfortably concrete. Behind every failed endpoint is a patient who consented to biopsies, blood draws, travel, and waiting rooms. Behind every amendment is a cohort whose lived experience was reshaped by a protocol we hadn't finished thinking through. Designing oncology protocols well is, at its core, a question of how much we are willing to ask of the people we serve.

In our work as clinical strategy partners, we have seen the same handful of design errors repeat themselves across sponsors, geographies, and therapeutic areas. They are not exotic mistakes. They are quiet ones, buried in the back pages of feasibility worksheets and eligibility checklists, and they are remarkably consistent. Below we walk through what the evidence — and the patient experience — keeps telling us.

The Strategic Cost of Protocol Complexity: Why High Endpoint Counts Derail Trials

Every additional endpoint we add to a protocol carries a hidden tax. It is paid not in dollars but in patient burden, in site fatigue, and in the slow erosion of a study's timeline. The data on this is unusually clear: protocols carrying between eleven and fifteen endpoints see dropout rates in the 25–35% range and typically require around 2.3 study amendments to stay viable. Push the design past sixteen endpoints, and dropout climbs to between 38% and 52%, with cumulative delays that exceed twelve months.

Those numbers are not statistics. They are people. A 40% dropout rate in an adjuvant breast cancer trial, for instance, doesn't just dilute statistical power — it means four out of every ten women who said yes to participate eventually said no to the next visit, the next scan, the next cycle. Somewhere in that decision was a moment when the protocol stopped matching what their daily life could sustain.

We do not enroll subjects. We enroll caregivers, commuters, working parents, and people in pain. Every endpoint we ask of them is a negotiation with the rest of their week.

The temptation to add endpoints is well intentioned. Sponsors want to extract maximum scientific value from every consented participant; regulatory teams want robust safety characterization; biomarker leads want exploratory signals. But the protocol is not a wish list. It is a contract with a finite human attention span. When we treat it as the former, we discover — late and expensively — that we have built the latter against ourselves.

A practical threshold we have come to rely on: if a protocol cannot be summarized on a single laminated card that a site coordinator can read in under three minutes, it is probably overbuilt. Clarity at the bedside is not a stylistic preference. It is an operational requirement.

Endpoint Discrepancies: Bridging the Gap Between Registries and Clinical Protocols

One of the more uncomfortable findings from the systematic review literature is just how often the protocol we submit to a regulator and the protocol we register publicly diverge. In a 2018 review of 71 randomized controlled oncology trials, 63% showed discrepancies in their secondary endpoints between the clinical trial protocol and the registry listing. Primary endpoint discrepancies were less common — about 6% — but they are the ones that hurt most when discovered.

Why does this matter for patient care? Because the registered endpoint is, for many patients and advocacy groups, the one they will read. It is the promise the study makes to the world. When the protocol quietly shifts its definition of progression-free survival, or changes how it scores response, that promise changes too — without anyone quite admitting it has. Investigators in the field have pointed out that this kind of opacity corrodes the trust relationship between trial sites and the patient communities that refer participants to them.

A 2024 commentary in the oncology press drew a useful line under the problem: protocol amendments are not inherently improper when transparently justified and reviewed by the relevant oversight bodies, but they become corrosive when they happen below the surface of public registries. The fix is procedural rather than scientific. Lock the endpoint definitions before first patient in. Resolve analytical ambiguities in advance. Treat the registry record as a binding instrument, not a draft.

The Surrogate Endpoint Dilemma: Balancing Regulatory Speed with Clinical Validation

Nowhere does the tension between speed and meaning become sharper than in the choice between Overall Survival and its surrogates. OS remains the regulatory gold standard in oncology — objective, unambiguous, and clinically interpretable in the simplest possible terms: did the patient live longer? But OS takes time, and time is the one resource oncology programs have least of. So we reach for Progression-Free Survival and Objective Response Rate, knowing that many of these surrogates have never been formally validated against the outcomes patients actually care about.

We shorten timelines by standing on endpoints we haven't fully validated — and then we ask patients to stand on them with us.

The evidence on this is mixed in ways we should be honest about. Surrogates are not uniformly bad; in some disease settings, particularly where survival is confounded by extensive crossover or long post-progression survival, PFS and ORR may genuinely accelerate access to meaningful therapies. The error is treating surrogate use as a free option. When a surrogate endpoint is unvalidated, the trial is no longer asking "does this work?" but "does this correlate with something we hope works?" Those are different questions, and the second is harder to answer at the bedside.

For sponsors navigating this, the practical discipline is to declare, in advance, how the surrogate result will translate — or fail to translate — into a survival claim. If a positive PFS finding will not support a survival claim without a confirmatory study, that limitation belongs in the protocol and the informed consent document, not in a footnote two years later.

Operational Feasibility: Addressing Recruitment Bottlenecks and Strategic Misalignment

Complexity and endpoint design are upstream of a third failure mode that lives entirely on the operational side. When we look at why trials fail by phase, the picture is instructive: in Phase 1, strategic decisions account for roughly 42.3% of terminations, often meaning sponsors pulled programs based on portfolio reshuffles rather than clinical signals. In Phase 2, poor recruitment rises to the leading cause of failure at about 30.0%, and in Phase 3 it remains stubbornly present at around 22.0% — even at the most expensive stage of development, the front door of the trial is where we lose.

This is the moment where protocol design meets reality most harshly. A protocol that looked elegant on paper can become impossible to recruit when it asks for a biomarker-positive patient who is treatment-naïve, has organ function within a narrow band, lives within driving distance of a specialized center, and is willing to undergo an extra biopsy. Each individual criterion is reasonable. Together, they describe perhaps one patient in a hundred at most.

Feasibility analysis is the part of protocol design we keep under-resourcing, and we know it. Inadequate feasibility assessment, poor site selection, and complex protocols are consistently listed among the preventable drivers of primary endpoint failure. Doing feasibility well means asking sites — not just feasibility vendors — what their actual patient flow looks like. It means piloting eligibility criteria against real clinic schedules. It means budgeting for what happens when enrollment is slower than projected.

Optimizing Patient Eligibility Criteria to Reduce Protocol Amendment Frequency

Eligibility criteria are where the protocol's abstract design becomes a daily workflow. Every exclusion is a conversation a coordinator must have with a referring physician, every inclusion a screening test that costs the patient another afternoon. Industry-wide, the median number of eligibility criteria in oncology phase 2/3 trials runs into the dozens, and many of those criteria exclude patients for reasons that have less to do with safety than with historical caution.

Recent years have seen a quiet but meaningful shift here. Broader eligibility initiatives — piloted across multiple cooperative groups and increasingly adopted by sponsors — have shown that many of the traditional exclusions (mild renal impairment, controlled HIV, prior treated malignancies within a certain window) can be relaxed without compromising either patient safety or interpretability of results. The effect on recruitment and on the representativeness of the trial population is substantial.

The connection to protocol amendments is direct. Protocols with restrictive eligibility tend to recruit slowly, accrue at fewer sites, and end up amended to broaden criteria mid-study — each amendment a signal that the original design did not match the patient population actually available. Roughly half of all protocol amendments are eligibility expansions. We can anticipate most of them if we are honest about who, in practice, will be sitting in the infusion chair.

A useful exercise at the design stage: for every exclusion criterion, write down the specific patient whose participation it would have prevented, and ask whether that prevention served a clinical or a regulatory purpose. If the answer is regulatory, the question becomes whether that regulatory constraint is current. If the answer is historical caution, the question becomes whether the caution still earns its keep.

Bringing It Back to the Bedside

What we are really designing, when we sit down to write a protocol, is a care pathway. Every endpoint we add is a procedure. Every eligibility criterion is a door that closes for someone. Every amendment is a recalibration of a promise we made to patients, sites, and the public registry. The oncology protocols that succeed — that actually deliver meaningful endpoints and translate into bedside benefit — tend to share a quality that is hard to measure but easy to recognize: they were written by people who had thought carefully about what they were asking of others.

The numbers we cited earlier are not inevitable. A 50% primary endpoint failure rate is an industry average, not a law of nature. A 38–52% dropout rate for high-complexity protocols is a pattern, not a destiny. Each of these figures represents thousands of decisions — to add an endpoint, to tighten a criterion, to skip the feasibility call — that, in aggregate, shape whether a promising therapy reaches the people who need it.

What we owe the patients in our trials is not perfection. It is care, applied to the document that governs their participation. Care in the form of validated endpoints. Care in the form of eligibility that reflects who actually sits in front of us. Care in the form of feasibility conversations that happen before, not after, the first site is activated. The protocol is the most consequential patient-facing document we produce, and it deserves the same thoughtfulness we would bring to a treatment decision for someone we loved.

That, more than any regulatory checkbox, is the lesson the failed endpoints keep teaching us.

FAQ

How does the number of endpoints in a clinical trial affect patient dropout rates?
Protocols with 11 to 15 endpoints typically see dropout rates of 25–35%, while those exceeding 16 endpoints experience dropout rates between 38% and 52%.
Why do discrepancies between registered and actual trial endpoints matter?
These discrepancies create opacity that can erode the trust relationship between trial sites and the patient communities that refer participants to them.
What is the primary cause of trial failure in Phase 2 oncology studies?
Poor recruitment is the leading cause of failure in Phase 2 trials, accounting for approximately 30% of terminations.
What percentage of protocol amendments are related to eligibility criteria?
Approximately half of all protocol amendments are driven by the need to expand eligibility criteria.
What is a practical rule of thumb for determining if a protocol is overbuilt?
If a protocol cannot be summarized on a single laminated card that a site coordinator can read in under three minutes, it is likely overbuilt.

Read also