Multi-Site Coordinator: Stop Running Sites in Silos

Most oncology protocols assume that twenty sites will run like twenty instances of the same playbook. They won't. A Phase III oncology protocol that activates twenty sites is, operationally, twenty slightly different studies — each with its own EHR quirks, its own coordinator turnover exposure, and its own reading of the latest amendment.

When a substantial amendment lands, the question isn't whether sites will diverge. It's how fast, and how visibly the divergence shows up in the next enrollment cycle. On the sponsor side, the answer is usually: faster than the CTMS surfaces, and visibly enough to drive a six-to-ten-week enrollment dip on the back end of the amendment.

~5×

cross-site variance in time-to-amendment implementation, fastest vs. slowest site

~3 wks

median lag between amendment approval and last-site re-match completion

The phrase "running sites in silos" gets used loosely. In sponsor-ops practice it has a precise meaning: each site interprets the protocol amendment on its own timeline, re-matches its own screening queue on its own schedule, and surfaces its results to the sponsor through a CTMS that has no cross-site view of the drift. The sponsor sees an aggregate enrollment dip weeks after the cause is already visible at the site level — and usually after the first sites have already moved on.

What cross-site variance actually looks like in practice

Take a Phase III solid-tumor protocol recently out of amendment. The amendment tightened the washout-window criterion and added a new biomarker exclusion. The sponsor's own protocol operations log shows how the same amendment landed across the twenty active sites:

Reading the amendment

Sites read the amendment in their own interpretation. Some flag the biomarker exclusion as a hard rule; others treat it as a soft preference pending PI clarification. Two sites in the same network can ship two different eligibility behaviors from the same paragraph.

Re-matching the queue

Sites that interpreted the amendment as a hard rule re-match their entire screening queue within the week. Sites that treated it as soft hold off pending PI guidance, then re-match all at once three weeks later. Patient ranking diverges between the two cohorts as a result.

Reporting the re-match

Sites that ran the re-match first report a clean cohort number to the sponsor. Sites that haven't yet report nothing — and the sponsor's aggregate still says "moving." Nobody at the sponsor level knows which sites are which until the dip hits the cumulative line.

Catching up to alignment

By the time the lagging sites re-match and report, the leading sites have already enrolled patients who would have been excluded under the lagging sites' reading. The retroactive alignment costs more than the amendment itself.

The pattern repeats across amendments. It is not a one-off. The sites that ran fast on this amendment will not necessarily run fast on the next one — coordination is local, so performance is local, too.

The re-routing problem

When an amendment lands at a sponsor and lands at the sites, what should propagate is a single updated eligibility definition across the entire active roster of sites and the entire screening queue. What actually propagates is a network of variant definitions — each site reading the amendment in its own way, then re-matching against a queue the sponsor can't see, against patients the sponsor can't rank uniformly, on a schedule the sponsor can't aggregate honestly.

Tufts CSDD tracks the median site lifecycle for Phase III protocols at roughly 24 months, with an average of 2.6 substantial amendments per protocol — and an observed six-to-eight-week cumulative enrollment dip after each one (Tufts CSDD Impact Report, 2024). What the Tufts aggregate does not show is the cross-site variance that hides inside that dip. The dip itself is the consistent, measurable cost. The variance inside the dip is the larger and more durable one.

"The amendment isn't the cost. Twenty sites re-implementing the amendment on twenty timelines is the cost. The protocol itself didn't change. The network did."

For the sponsor, the practical consequence is that the same patient gets ranked differently across sites — eligible here, screen-failed there, pending-PI-clause at the third. Patient re-matching is one of the most expensive single operations in a Phase III oncology program, and the variance gives the sponsor no way to invest in it once and benefit cohort-wide. Each site pays the cost. None of them coordinate.

The scale of the problem: A 20-site Phase III oncology program running two substantial amendments over an 18-month enrollment window loses an estimated 6–10 weeks of enrollment velocity per amendment to cross-site variance, site-by-site re-reading, and uncoordinated re-matching. That is 12–20 weeks of dead air across the network — and it compounds multiplicatively with site volume.

What happens when coordination is centralized

Centralizing the coordination layer at the sponsor side — and pushing the same eligibility definition, the same re-match logic, and the same ranked patient cohort down to every site in the network as a single operation — collapses the variance in one move. The amendment propagates to all sites simultaneously. Eligibility logic stays in sync. Patient re-matching ranks once, against one definition, and surfaces the same cohort at every site.

For the sponsor, the operational effect shows up at three layers:

Sponsor layer — visibility. The sponsor sees a single network-wide view of which sites have re-implemented the amendment, which have re-matched their queue, and which are still pending. The aggregate dip becomes a forecastable event, not a surprise.
Site layer — reduction of rework. Each site runs one re-match against the canonical cohort, not its own re-interpretation. Site hours spent reading the amendment, querying the PI for clarification, and reconciling site-vs-network eligibility drift collapse into a single workflow.
Patient layer — uniform ranking. The same patient, ranked against the same definition, gets the same disposition at every site. Patients who would have been eligible at Site 4 and screen-failed at Site 9 under the siloed model get ranked consistently — and the cohort that surfaces for enrollment is the same cohort across the network.

The result moves from twenty independent studies toward one study running in parallel. See what coordinated vs. siloed screening looks like across 10 sites and 400 patients →

For sponsor-side study managers, the value isn't a faster trial in isolation. It's a trial whose enrollment dip after an amendment is visible a week ahead instead of ten weeks behind. The early signal is the same early signal a forward-looking risk model would surface — and when both run together, the dip becomes a forecast, not an after-the-fact reconciliation.

What changes isn't the workload at any single site. It's that the sponsor stops paying for the same operation twenty times.

"The cross-site variance isn't a site problem. It's a coordination architecture problem — and it lives at the layer the sponsor controls, not the layer each site runs on."