Methodology
Where the data comes from, how we clean it, and exactly how every metric and score is calculated, in plain language.
Includes ClinicalTrials.gov updates up to 9 October 2026
Trialmetrics uses only public sources. We don't buy data, scrape private sites or use AI to estimate numbers.
The largest public registry of clinical studies, run by the US National Library of Medicine. All trial, site, sponsor and investigator data on Trialmetrics comes from here, refreshed every week.
We use the Orphanet nomenclature of rare diseases (INSERM) to identify rare disease trials, under the CC BY 4.0 licence. Orphanet does not endorse Trialmetrics.
The population health figures on site profiles come from public health agencies. For US sites we use CDC PLACES estimates at ZIP code level, compared with the US national average. For sites outside the US we use country-level figures from the WHO Global Health Observatory, compared with the global average. These are modelled population estimates for the area around a site, not counts of the site's own patients. Neither agency endorses Trialmetrics.
In three places, all to put information into words: plain-language summaries of trial descriptions and results, the written site summary on site profiles, and the written summary on condition analytics. The site and analytics summaries only describe figures we have already calculated. AI is never used to calculate metrics, scores or rankings, and every figure it mentions can be checked against the numbers on the page.
Raw registry data is messy. Each week, the refresh runs these steps in order.
Every week we load the latest public data from ClinicalTrials.gov: trials, sponsors, locations, investigators, conditions and posted results.
The same site is often listed under several names, such as “Mt. Sinai” and “Mount Sinai Hospital”. We clean facility names and group records that share the same institution and location, so each site's full trial history sits in one profile.
Country names are matched to one standard name, so searching for England or UK finds United Kingdom.
Each trial's conditions are mapped to therapeutic areas such as cardiovascular or oncology using the standard medical vocabulary (MeSH) the registry already uses.
A trial counts as rare disease when one of its conditions matches a disease name in the Orphanet nomenclature of rare diseases.
Finally, we recalculate every site metric and site score from the refreshed records, using the rules below.
Every metric on a site profile is counted directly from the site's registry records.
Every site profile shows a track record ranking: how reliably the site sees its trials through, compared with other sites. Here's how it's worked out.
Count completed trials
We look at every trial at the site that has ended, and count how many completed.
Only count stops that could be the site's fault
We read the reason each stopped trial gives. Slow recruitment, low enrolment or staffing problems count against the site. Sponsor business decisions, funding, safety or efficacy findings and COVID don't. Stops with no clear reason count half.
Be fair to sites with few trials
A site's rate is blended with the typical rate until it has enough history of its own, so 2 out of 2 doesn't outrank 180 out of 200.
Rank against other sites
We show where the site sits among other sites with enough history, such as Top 10%, Top 25% or Above average. Generic sponsor names like "Investigational Site" are not ranked.
Show how sure we are
High confidence means 15 or more ended trials behind the ranking, Medium means 5 to 14. With fewer than 5 we say there isn't enough history to rank the site.
We ranked about 700 sites using only their history up to 2019, then checked how the trials they started in 2019 and 2020 actually turned out. We tried several approaches, including trial volume, repeat sponsors, recent activity, results reporting and a model adjusting for each trial's phase, sponsor and size. The fair completion record was the measure that held up best, so it's the only thing that drives the ranking. We kept the simplest method that worked rather than a complicated formula.
The other measures are still useful, so each profile shows them alongside the ranking as context:
Most sites complete the large majority of their trials, so the differences between them are often small. Use the ranking as a starting point for a shortlist, not a final verdict.
Spotted something that looks wrong? Email support@trialmetrics.com and we'll look into it.