Why GU trials stall at screening, and the first three things to change

At an NCI-designated comprehensive cancer center, 21.6% of patients enroll in a treatment trial. At an academic comprehensive cancer program, the figure is 5.4%. Same disease, same protocols, often the same catchment. That four-fold gap is the one worth explaining, and it is not a gap in science.

Most conversations about trial accrual open with the national average, which is now 7.1% for cancer treatment trials across all program types.¹ It is a better number than the 3% to 5% still quoted in most slide decks, and it is worth correcting the record. But the national average is not actionable. The spread between comparable institutions is.

I have spent fifteen years in clinical research, nine of them in oncology and genitourinary trials, running Phase I–IV portfolios and the coordinator teams behind them. Over those years I have watched a great many GU studies open with sound science, a motivated PI, and a catchment full of eligible patients, and then accrue at half of target. When that happens, the post-mortem almost always blames the protocol or the population. It is almost always neither.

It is screening. Specifically, it is the stretch between identifying a patient who could enroll and obtaining a signature from a patient who did. That stretch is where GU studies lose people, and it is measurable, and it is fixable.

“Low accrual” is four different failures wearing one label

Eighteen percent of trials in the NCI’s National Clinical Trials Network closed with low accrual, or were still accruing below half of target three years or more after opening.² Below 50% of target is the line that triggers formal review and possible termination, so this is not an abstract concern. It is the threshold at which a study starts dying.

The trouble with the phrase “low accrual” is that it names a symptom and hides the cause. A study that never identifies eligible patients and a study that identifies plenty but cannot get them consented both report the same number at the same review. They need opposite interventions.

So before changing anything, count the transitions separately: how many patients were identified, how many cleared prescreen, how many were actually approached, how many consented, how many enrolled. Most sites can produce the first and last figures on demand and have to go digging for the middle three. The middle three are where the answer lives.

In GU specifically, three losses account for the overwhelming majority of what I see.

Where GU studies actually lose patients

Loss 1 — Eligibility written tighter than the science requires

Across fifteen randomized studies auditing screen failure, 51% of screened patients failed screening, and the single largest reason was not meeting inclusion criteria, at 54.9% of all failures.³ Patients declining treatment came a distant second at 22.2%.

Read that carefully. It means the most common way a trial loses a patient is that someone already identified them as a plausible candidate, invested coordinator hours in screening them, and then discovered a criterion they could not meet. Every one of those is a fully-loaded screening cost with no enrollment attached.

This is well-trodden ground at the policy level. ASCO and Friends of Cancer Research have now issued two rounds of recommendations: the first addressing minimum age, HIV status, brain metastases, organ dysfunction, and prior or concurrent malignancies; the second addressing washout periods, concomitant medications, prior therapies, laboratory reference ranges and test intervals, and performance status.⁴

In GU trials, the criteria that quietly do the most damage are renal function thresholds and prior-therapy washouts. A creatinine clearance floor inherited from a protocol template excludes a meaningful share of the renal cell population by definition — these are frequently post-nephrectomy patients. A 28-day washout written without reference to the half-life of the prior agent excludes patients whose disease will not wait 28 days.

Loss 2 — The patient is never asked, or is asked in language they do not use

This is the loss that does not appear in any screening log, because a patient who is never approached never becomes a screen failure. They simply are not there.

In prostate cancer specifically, the numbers are stark. Black men have the highest incidence and mortality of any racial group in the United States, yet enroll in industry-sponsored prostate trials at a rate 67% below what incidence would predict.⁵ When researchers asked why, the leading answer was not mistrust, and it was not fear of randomization. It was that nobody asked them: 55.1% of non-participants cited never having been invited, with a further 13.5% citing a lack of information about risks and benefits.⁵

There is a second, subtler version of this failure, and it is one I have worked on directly. In a study of men newly diagnosed with prostate cancer, my co-authors and I measured comprehension of the standard clinical vocabulary used to discuss genitourinary function — the words that appear in every consent form and every treatment discussion. Comprehension was poor. More usefully, we found that tailoring the conversation with colloquial alternatives was preferred by the majority of patients regardless of their measured health literacy.⁶

That last clause is the operationally important one. This is not a matter of identifying low-literacy patients and simplifying for them. The clinical vocabulary was working less well than assumed for nearly everyone. A patient who does not follow the description of what a trial involves cannot meaningfully consent to it, and will often decline rather than admit confusion — which is recorded as a patient refusal, not as a communication failure.

Loss 3 — The site is reachable on paper only

In metastatic prostate cancer, 50.2% of trials would require a patient to drive more than sixty minutes to reach a participating site.⁷ For a study with weekly visits during a treatment cycle, that is not an inconvenience. It is an exclusion criterion that does not appear in the protocol.

This interacts badly with Loss 2. The populations most underrepresented in GU trials are frequently the same populations least able to absorb a two-hour round trip and a lost day of wages for a visit that is not therapeutic. The catchment map says the patient is reachable. The visit schedule says otherwise.

The first three things to change

These are sequenced deliberately. Each one makes the next cheaper.

How to tell it worked

Accrual is a lagging indicator; by the time it moves, two quarters have gone. Instrument the transitions instead, and watch four figures monthly:

• Identification-to-approach rate. Of patients flagged as potentially eligible, what share were actually approached? This is the direct measure of Loss 2 and typically the first to move.
• Screen failure rate, decomposed by criterion. Not the headline percentage — the breakdown. A falling rate concentrated in one criterion tells you the waiver worked.
• Approach-to-consent rate. If this is low while identification-to-approach is healthy, the problem is the conversation or the visit burden, not the funnel above it.
• Time from identification to consent. In GU oncology this is frequently the binding constraint; patients whose treatment decision cannot wait will start standard of care and become ineligible.

One caution on benchmarking. A figure of roughly 80% circulates as a national screen-failure rate. It is a planning assumption, not a measurement, and measured rates vary widely by phase and design — 21.7% to 26.4% in early-phase work at three cancer centers, around 51% across a set of randomized studies.³ Benchmark against your own trailing twelve months and against studies of comparable design. Anything else produces false comfort or false alarm.

None of the three changes above requires new technology, and none requires a larger catchment. They require knowing which transition is failing, which is a question most portfolios cannot currently answer — and answering it is usually a matter of weeks, not quarters.