Design your trial on real patient data

Ask before you start designing. Apply your criteria and see whether they hold in real records.

Backed by

• You cannot tell whether any patients survive your criteria until the design is already underway.

  Build a cohort from your criteria

  Ask with a drug, a condition and a window, and you see whether patients matching it exist in the records. Brand names, generic names and English spellings are matched together in the source text, so nobody drops out over spelling.

  Patients on Keytruda at stage III or later

  키트루다펨브롤리주맙Keytrudapembrolizumab

• Every hospital writes the lab name differently, so one measure gets counted as several.

  Spelling variants unified automatically

  AST alone is written sixteen ways across hospitals. They have to be one item before anything is comparable, and the product does that folding for you.

  AST(GOT)GOT (AST)AST(SGOT)SGOTALT(GPT)GPT (ALT)

  ASTALT

• Stratifying on an axis that has no standard code means someone has to read the charts.

  Stratify on axes only the text holds

  Endoscopic grade, biomarker, line of therapy, recurrence — none of it survives coding, but all of it is written in the notes. You can split on those axes, and when a cell is small it is merged rather than reported on its own.

  • Stage I
  • Stage II
  • Stage III
  • Stage IV

  Early I–II · Advanced III–IV

• Vendors answer that data exists when it does not, and you find out after signing.

  If it is not there, we say so

  Nothing outside the corpus gets invented. The answer states what was matched and how, and a question we cannot answer says why, right there.

  Is the reason for discontinuation recorded?

  No narrative stating a discontinuation reason exists in the corpus.

Common questions

What if the patients we need are not in it?

We can collect more. Give us the criteria and we will scope how long it takes. Because patients request their own records, building a new cohort is something our structure allows.

Do we need ethics review first?

Not to check a patient count. What we hand over is an aggregate, not research data. The procedures for conducting the study itself remain yours to follow.

How do we know the number is right?

The output states what was matched and how. That method measured 100% recall with zero false positives.

Can we take the patient data and analyse it ourselves?

No. Patient-level data does not leave, contract or not. What leaves is the aggregate result and de-identified excerpts from the text.