SiteciteMethod
Every number has a receipt.
The sources, the order, the formula and the published tests, on one page.
Sources
- ClinicalTrials.gov (public API v2): trials, their phases, statuses, dates, enrollment, facilities and the investigators named on them.
- PubMed (NCBI E-utilities): papers matched to an investigator by name, affiliation and condition, with the match confidence shown (High or Medium; lower matches are left out).
Each sample page states the date its data was pulled.
The order you see
Sites are listed the way a careful manual registry search lists them: by the number of comparable trials the site is listed on, most first; ties by latest completion, then name. Spelling variants of the same facility are merged first. The Sitecite score is a second column.
A trial is comparable when it is a completed or terminated drug trial in the same condition and phase family, ended within the lookback window (12 years). A trial is competing when it is recruiting, active or not yet recruiting in the same condition.
The score
A number from 0 to 100, from four parts at default weights:
| Part | Weight | What it measures |
|---|---|---|
| Experience | 35 | Comparable trials listed, against the 90th percentile of sites in the pool |
| Completion | 25 | Share of listed trials completed rather than terminated, less a penalty for terminations whose recorded reason is slow accrual |
| Enrollment | 25 | Participants per listed site on its trials, against the 75th percentile of the pool |
| Recency | 15 | Full marks if the site's latest completed trial ended within 3 years, none beyond 12 |
The published test
Before building further we tested the score against a plain trial count. For 30 completed Phase 2-3 trials in 10 indications, we built each shortlist only from trials that finished before the held-out trial started, then counted how many of that trial's own listed sites appeared in the top 10, 20 and 50. Mean hits per trial:
| Cutoff | Random pick | Trial count (registry order) | Sitecite score | Score / count |
|---|---|---|---|---|
| top 10 | 0.03 | 1.70 | 1.23 | 0.73 |
| top 20 | 0.30 | 2.77 | 2.07 | 0.75 |
| top 50 | 0.97 | 4.17 | 3.33 | 0.80 |
Result: ordering by trial count found more of a trial's later sites than the score did (the score did worse than the count on 37% of the 30 trials). That is why every list here is in the registry's order and the score sits in a second column, as an optional sort. Both beat a random pick by a wide margin.
What the registry does not contain
- Listed is not enrolled. The registry lists sites, not how many patients each site enrolled. A site “listed on a trial” was activated for it.
- History is a partial guide. Only 30% of the held-out trials' sites had appeared on an earlier comparable trial. A site with no history is invisible to any ranking built from the registry.
- Names vary. Merging spellings is automatic and can miss or over-merge a facility; the merged names are shown.
- Papers are matched, not confirmed. Each paper shows its match confidence.
A cited data summary, not a recommendation to select or exclude any site.
Check it yourself
Every NCT and PMID on this site is a link. Open any row's trials on ClinicalTrials.gov and compare.
Check your CRO's list the same way
Leave your indication and the CRO's site list. We reply by email with each name checked against the registry.