What could slow our studies down, and where?
The risks across every running study: recruitment, the KRIs and QTLs that watch it, and the regional shocks that can hit it from outside.
Sites waiting too long for a first patient
The earliest warning is usually a site's own screening log, not the news. A site is flagged only when a wait this long would be unusual for its expected rate, so a site that is slow by design is not flagged too early. Today, by recruiting study:
- EPI-301: none of its 159 open sites
- EPI-214: 13 out of 120 open sites
- EPI-405: none of its 18 open sites
What a week's warning is worth: each day a medicine reaches patients later costs its sponsor about $500,000 to $800,000 in lost sales. The median across medicines (Smith, DiMasi and Getz 2024; D47). The Tufts white paper’s estimate (Smith, DiMasi and Getz, white paper 2024; D47).
What separates sites that recruit from those that do not?
Every open site in the six studies, 367 in all, is compared with what it was expected to recruit, given its type, track record, country and protocol. Pooling sites narrows the range chance allows, so patterns no single site can show become visible.
The kind of site explains little
Patients randomised as a share of those expected, by site profileAll open sites; the bar is where an ordinary group of that size would land. Profiles run from ×0.61 to ×1.01
Show data
| Group | Sites | Randomised | Expected | Observed ÷ expected | Ordinary range | Judgement |
|---|---|---|---|---|---|---|
| Comprehensive cancer centre | 26 | 23 | 37.8 | ×0.608 | ×0.562 to ×1.529 | Within normal variation |
| Academic medical centre | 28 | 129 | 159.1 | ×0.811 | ×0.627 to ×1.451 | Within normal variation |
| Regional teaching hospital | 98 | 342 | 337.8 | ×1.013 | ×0.766 to ×1.262 | Within normal variation |
| Dedicated research centre | 51 | 232 | 242.0 | ×0.959 | ×0.694 to ×1.357 | Within normal variation |
| Community hospital | 46 | 146 | 145.5 | ×1.004 | ×0.674 to ×1.381 | Within normal variation |
| Community specialist practice | 34 | 73 | 103.6 | ×0.705 | ×0.622 to ×1.455 | Within normal variation |
| Site new to research | 84 | 176 | 217.5 | ×0.809 | ×0.739 to ×1.295 | Within normal variation |
How to read it: each row is a kind of site. ×1 means its sites recruited what sites of their kind are expected to; further left is fewer. The bar is how far an ordinary group of that size strays by chance, so a dot inside its bar is no sign that the kind of site matters.
Source: epiClinical synthetic sites. Expectations already allow for site type and track record, so a profile on ×1 is doing what its kind usually does.
What a site has matters more than what it is
Patients randomised as a share of those expected, sites with each fact (filled) and without (hollow)All open sites; sites with the fact in brackets; the band is where an ordinary group of that size would land
Show data
| Fact | Sites with it | With | Without | Ordinary range |
|---|---|---|---|---|
| Does not check records against the criteria before inviting patients | 163 | ×0.547 | ×1.172 | ×0.807 to ×1.211 |
| Screening scans are done at another hospital | 41 | ×0.609 | ×0.939 | ×0.657 to ×1.405 |
| Half a coordinator or less on this study | 72 | ×0.36 | ×1.027 | ×0.729 to ×1.308 |
| Two or more competing trials open at the site | 77 | ×0.369 | ×1.044 | ×0.735 to ×1.301 |
| Relies on referrals, with no formal referral network | 18 | ×0.346 | ×0.94 | ×0.524 to ×1.611 |
| Two in five patients or more travel over an hour | 43 | ×0.364 | ×0.974 | ×0.645 to ×1.424 |
| Shares its city with another site in the same study | 193 | ×0.874 | ×0.916 | ×0.822 to ×1.193 |
How to read it: each row is something recorded about a site. The filled dot is the sites that have it, the hollow dot the sites that do not; further left is fewer patients than expected. The widest gap: sites that two or more competing trials open at the site recruited ×0.37 of what was expected, against ×1.04 for the rest.
Source: epiClinical synthetic sites and screening logs. The first six facts were built into the synthetic data so that the study pages have causes to find. With real data, this chart is where such links would first show, or fail to. Sharing a city with another site in the same study was not built in: ×0.87 against ×0.92, within normal variation. The evidence says sites compete mainly where the eligible pool is scarce (register S-21).