Case study · Trial Matching
Duke and iCubed built a full oncology trial in 7.5 days.
Duke Clinical Research Institute & iCubed
Dr. Christoph Hornik
MD, MPH Associate Director of i-Cubed, Duke Clinical Research Center
“Health Universe brings the platform discipline needed to generate, test, and improve AI agents for clinical trials.”
Setup
To stand up a full oncology trial, down from 6–9 months
Speed
Faster end to end — a 93% reduction in setup time
Precision
Fewer false-positive strong matches — ~15 fewer dead-end reviews per patient
The challenge
Months to build a trial, and matching that never reads the chart.
For Duke and iCubed, standing up a trial took six to nine months by hand, and matching is manual, slow, and buried in false positives — because keyword matching scores patients eligible without ever reading the chart.
Fewer than 5% of adult cancer patients enroll in a trial, even when they may be eligible.
What they needed
01
To stand up a trial protocol far faster than six-to-nine months of manual work
02
Matching that actually reads the full patient record, not just keywords
03
A dramatic reduction in false-positive “strong matches” that waste coordinator time
04
Transparent, inspectable reasoning on every match, with the final call left to the team
The difference
Why they chose Health Universe
Building a trial and finding its patients were two slow, disconnected jobs, usually solved by two tools months apart. Only Health Universe did both on one platform — and showed its work at every step.
Generates the whole trial, not fragments
From a synopsis, 12+ coordinated agents produce the full protocol and the regulatory submissions needed to launch.
Reads the chart, not keywords
Each patient's full record is evaluated against the criteria, catching age and logic errors baseline tools miss.
Shows and ranks its reasoning
Every match carries its clinical logic, and coordinators make the final eligibility call.
In practice
The Approach
A coordinated team of 12+ agents turned a synopsis into the full protocol plus IRB and ClinicalTrials.gov submissions, end to end, with audit trails the research team reviewed and approved. The same platform then read each patient's full record against the criteria and ranked the matches by clinical reasoning.
Before Health Universe
After Health Universe
Trial setup time
6–9 months by hand
7.5 days — 30–40× faster, a 93% reduction
Protocol creation
Manual, fragmented
12+ agents produce a 75-page protocol from a synopsis
Regulatory submissions
Manual
IRB + ClinicalTrials.gov automated end to end
Matching basis
Keyword scoring, no chart read
Full-record reading with inspectable reasoning
False positives
High — reviews buried in dead ends
37% fewer false-positive strong matches
Trial setup time
Before
6–9 months by hand
After
7.5 days — 30–40× faster, a 93% reduction
Protocol creation
Before
Manual, fragmented
After
12+ agents produce a 75-page protocol from a synopsis
Regulatory submissions
Before
Manual
After
IRB + ClinicalTrials.gov automated end to end
Matching basis
Before
Keyword scoring, no chart read
After
Full-record reading with inspectable reasoning
False positives
Before
High — reviews buried in dead ends
After
37% fewer false-positive strong matches
Proof
The Results
Trial setup went from six-to-nine months to seven and a half days.
Evaluated against a real oncology trial using actual trial standards, Project Loom delivered 10× faster document processing, +21% higher matching precision at identical discriminative accuracy (ROC AUC) — and about 15 fewer dead-end reviews per patient.
Setup
To stand up a full oncology trial, down from 6–9 months
Speed
Faster end to end — a 93% reduction in setup time
Precision
Fewer false-positive strong matches — ~15 fewer dead-end reviews per patient
Trial Matching
“Working with Health Universe on Project Loom allowed us to develop AI agents that move beyond isolated tasks to support the full, end-to-end clinical trial workflow.”
Michael Cohen-Wolkowiez
Executive Director of i-Cubed · Duke Clinical Research Center
See these outcomes reproduced on your own charts, live with our team.