Month 7 of a Phase II pancreatic cancer trial. Your enrollment dashboard is green. Cumulative screens are tracking to target. Screen-fail rate is within normal range. You're on track.
Month 8, two CRCs leave Site 12 — one for a competing trial, one for an offer she couldn't refuse. Site 12 was carrying 23% of projected enrollment. The dashboard is still green.
Month 9, enrollment is off by 34 patients and the sponsor wants answers. The answers were all in the data. Nobody was reading it the right way.
"The dashboard was reporting what had already happened. By the time you see a shortfall, you're too late to prevent it."
This is the fundamental limitation of every enrollment platform on the market: they show you history. Not hazard. As we covered in the first post of this series, enrollment failure is structural — and structural problems have leading signals, if you know where to look.
What Every Platform Is Showing You (And Why It's Already Too Late)
Open any CTMS enrollment dashboard today. Here's what you're reading:
- Cumulative enrollment vs. target — tells you how many you've enrolled, not how many you'll enroll
- Screen-fail rate — tells you patients didn't qualify, after they've already been screened
- Time-to-consent — tells you how long consent takes, not whether consent is declining
- Site activation timeline — tells you which sites started, not whether the ones that started are about to stall
These are the metrics that have always existed. They're not wrong — they're just late. By the time cumulative enrollment shows a deficit, the window to prevent it is measured in weeks, not months. You've already missed the opportunity to open backup sites, reallocate budget, or intervene at the site level.
The industry has normalized arriving at month 9 with a problem. What it hasn't built — until now — is a system that tells you at month 4.
What Predictive Risk Scoring Actually Means
Risk scoring is not a dashboard. It's a model — trained on site-level behavioral signals that precede enrollment failure, not indicators of enrollment failure that's already occurred.
A properly constructed enrollment risk model ingests:
- CRC capacity signals — turnover events, concurrent trial load, time-in-role trends
- EMR query volume — screens being run against the patient database, week-over-week
- Screen-to-consent latency trend — not the average, but the direction of the trend
- Catchment competition — other recruiting trials in the same indication within the site's catchment area
- Protocol amendment frequency — every amendment changes the eligible pool; the direction matters
It outputs a forward-looking probability of missing enrollment targets at the trial level and at the site level, updated weekly. Not a dashboard — a forecast.
The Three Signals Every Dashboard Ignores
Here's where the model earns its name. These three signals are visible in site operational data 60–120 days before enrollment collapses. No standard CTMS shows them.
CRC Capacity Decay
CRC turnover is a leading indicator, not a lagging one. When a coordinator leaves Site 8, the site doesn't immediately show lower enrollment — it shows it 60–90 days later, after the open patient slots expire without replacement. But the departure signal is present on day one. A site running at 80% CRC capacity versus baseline is on a trajectory to miss target, even if current enrollment is nominal. Concurrent trial load compounds this: a CRC running two trials is not twice as slow, she's 3–4x slower on each one.
Catchment Competition
If Site 6 is in a 50-mile radius of another recruiting trial in the same indication, the eligible patient pool is partially drained — before either trial has screened their first patient. This signal is not in your CTMS. It's not in ClinicalTrials.gov flags. It's in a cross-trial overlay of the local recruitment landscape that requires data fusion across trials, sites, and geographies. As we covered in the Multi-Site Coordination Tax post, internal cannibalization within a sponsor's own portfolio is also a leading signal — and it's entirely invisible without a portfolio-level view.
Protocol Amendment Drift
Every substantial protocol amendment tightens the eligible pool. Not immediately — the effect is lagged — but predictably. A change in inclusion criteria that removes a biomarker threshold doesn't just affect new screens; it retroactively disqualifies patients who were already in the screening queue. Sites that don't have automated re-matching re-screen manually over the following 6–8 weeks. The enrollment velocity drop during that window is a known, quantifiable cost. Most platforms don't track it. A risk model should.
What Changes When You Score Risk Forward
Here's the number that matters: month 4 reallocation versus month 9 explanation.
A sponsor running a 12-site Phase II oncology trial identifies two sites at month 4 via the risk model: Site 4 (moderate risk, CRC capacity at 72%) and Site 9 (high risk, competing trial opened in catchment 6 weeks prior). The model puts both in "Watch" status with 60% probability of missing their individual targets.
What changes at month 4:
- Outreach budget shifts 30% from Sites 4 and 9 toward three lower-risk sites
- Backup site identification begins — two candidates already pre-qualified
- CRC capacity at Sites 4 and 9 flagged for proactive sponsor outreach — site leadership knows the issue before it becomes a crisis
- Amendment impact assessment run against the new protocol version — eligible pool re-calculated before manual re-screening begins
At month 9, the sponsor sees: enrollment on target, sites performing as expected. Without the model, they see the same dashboard — still green — and the collapse happens in month 10. The trial misses by 34 patients. The explanation to the investment committee costs more than the model would have.
Why Nobody's Done It Well (And What Makes It Hard)
Predictive enrollment risk scoring is not a new idea. Vendors have been claiming to do it for years. Here's why most of those claims don't hold up under scrutiny:
Multi-site data integration is genuinely hard
The model needs site-level behavioral signals — CRC turnover, EMR query volume, screen-to-consent latency — across all active sites in a trial. Most sponsors don't have a unified operational data layer that spans their site network. Data is siloed at the site level, partially in EDC systems, partially in CTMS, partially in spreadsheets that CRAs email on Fridays. A model trained on incomplete or lagged data produces unreliable scores. Building a data pipeline that ingests site operational signals in near-real time is a non-trivial infrastructure problem — it's also the reason the model works when the infrastructure is right.
Calibration across indications is non-trivial
A CRC turnover signal means different things in rare disease versus oncology versus cardiovascular. Enrollment velocity benchmarks differ by indication, by phase, by site type. A model that uses a single enrollment curve across all trials will be systematically wrong on at least half of them. Building indication-specific calibration layers — with meaningful sample sizes — requires a data flywheel: the more trials the model runs across, the better the calibration, the better the scores. Early adopters are training the model on their own portfolio.
The trust problem
Sponsors will not act on a risk score they can't interrogate. "Why is this site at 73% risk?" needs to be answerable in plain language. A black-box model that outputs a number with no signal attribution will be ignored by clinical ops leaders who've been burned by optimistic dashboards before. Syncra's risk model surfaces the contributing signals — the CRC departure, the catchment competition factor, the protocol amendment impact — so sponsors can interrogate the score, challenge it if needed, and act on it with confidence. A model you can't explain is a model nobody trusts.
Trials Closing on Time as the Default
The goal isn't a smarter dashboard. It's a different relationship with time.
Right now, the industry operates in a reactive loop: enrollment problem emerges, dashboard shows it, sponsor intervenes late, delay is absorbed or explained away. The loop perpetuates because the data arrives too late to change the outcome.
A predictive risk model changes the loop: site-level signals are ingested weekly, risk scores are updated, sponsors act at month 4 instead of month 9, trial closes on time, the loop stops. Not because the science got better or the sites got more motivated — because the information arrived when it was still useful.
That's what Syncra is building. Not a dashboard. A different relationship with time.
See your trial's risk profile in 60 seconds
The Predictive Enrollment Risk Scorer runs on your trial parameters — indication, phase, site count, enrollment window. Enter yours and get a score with contributing signal breakdown.
Try the Risk Scorer → Request a pilotFree Tool — NCT → PDF in 30 seconds
Get a personalized risk report for your actual trial
Enter your NCT number and work email. We pull the data from ClinicalTrials.gov, score it against Tufts CSDD benchmarks, and email you a 1-page PDF you can forward to your PI or ops lead.
Get my risk report →