Clinical Development · Feasibility · External Controls
Design the trial against the real population — before the protocol is locked.
Feasibility, inclusion/exclusion simulation, external-control cohorts, translational cuts — run as a query across federated real-world data, not as an eight-month RFP.
TRIAL FEASIBILITY
Sites, patients and I/E impact in hours, not months
PROTOCOL DESIGN
Simulate criteria before the protocol is locked
EXTERNAL CONTROLS
Defensible synthetic comparator cohorts
TRANSLATIONAL
Genotype–phenotype cuts on consented cohorts
01 · The shift
Most trials are designed before anyone has counted the patients.
THE OLD CADENCE
- Protocol drafts based on KOL intuition and published epidemiology.
- Feasibility vendor takes 3–6 months per country, per indication.
- Criteria are tightened to reduce risk — recruitment stalls anyway.
- External controls need a bespoke data acquisition for every submission.
WITH UNISON
- Author eligibility as OMOP-concept logic — not PDF prose.
- Count patients and characterise sites the same day.
- Tighten or loosen I/E criteria live — see the yield shift before you lock.
- Reuse the same cohort logic to build external controls and translational cuts.
02 · What Clinical Development can run
Five workflows, one query surface.
FEASIBILITY
Patient counts & site characterisation before RFP
Fan a single query across connected biobanks and hospital networks. Get eligible patient counts, geography, treatment landscape and comorbidity patterns — without moving data.
OUTPUT
Eligible-count table · geography · comorbidity profileI/E SIMULATION
Move criteria, watch the yield curve
Age band, prior therapy, biomarker threshold, washout window — every change re-runs the count. Find the point where the protocol is both enrollable and defensible.
OUTPUT
Criteria-sensitivity curve · bottleneck attributionEXTERNAL CONTROLS
Synthetic comparator cohorts, pre-specified
Build matched external-control arms for single-arm and open-label trials. Pre-specified, replayable, aligned to regulator expectations — same logic for every submission.
OUTPUT
PS-matched cohort · baseline table · methods artefactTRANSLATIONAL
Genotype–phenotype queries on consented cohorts
Characterise subpopulations — responders, non-responders, variant carriers — across federated biobanks. Same query, every cohort, aggregate-only.
OUTPUT
Subgroup tables · phenotype enrichment · cohort overlapPROTOCOL AMENDMENTS
Amendment impact before the clock starts
Before filing an amendment, re-run the cohort. See how many of your enrolled and planned-enrolled patients the new criteria would have kept. Bring evidence, not guesses.
OUTPUT
Impact delta · retention curve · replay artefact03 · From question to count
One cohort logic. Every downstream artefact.
01
Draft eligibility
Express inclusion and exclusion as OMOP-concept logic. Criteria become executable, not prose.
02
Count
Federate across connected datasets. Eligible patients, site distribution, treatment landscape — same day.
03
Tune
Slide the criteria. See the yield curve. Identify the bottleneck — age band, biomarker, washout.
04
Extend
Reuse the same cohort to build external controls, translational cuts and amendment simulations.
05
Defend
Every count backed by a replayable UQL artefact. Regulator or DMC can re-run, not just read.
A MORNING IN FEASIBILITY
"How does dropping the prior-therapy exclusion change the yield?"
A clinical lead is pressure-testing a Phase III protocol. Tightening the exclusion cuts the eligible pool in half; loosening it raises confounding risk. Question gets answered in the meeting, not the next quarter.
Same-day
counts across 4 federated datasets
Live tuning
9 criteria variations run in parallel
Aggregate-only
no patient-level movement
Defensible
replayable UQL artefact per run
# unison · feasibility workspace
> "Drop exclusion: prior anti-IL-17. Show yield delta."
→ Cohort: phase-3-psa-v7 · 4 datasets · aggregate-only
Current protocol n = 3,214
Without anti-IL-17 exclusion n = 5,887 (+83%)
Newly eligible breakdown:
· 1 prior biologic line n = 1,840
· 2+ prior biologic lines n = 833
Geographic distribution (top 3):
· US network 62%
· EU5 combined 27%
· UK national data 11%
# uql://query/feas-3c14 · replayable
04 · Built for regulated trial conduct
Aligned to how regulators and IRBs expect trial evidence to arrive.
Pre-specified, not post-hoc
Cohort logic is authored and versioned before execution. The analysis that runs is the one you registered.
Reproducible by construction
Every count, every comparator, every subgroup is a replayable UQL artefact. A regulator re-runs it on demand.
Custodian-controlled
Data never leaves the custodian. Only aggregate results and pre-approved artefacts move. Sovereignty preserved.
Standards & fit:· OMOP CDM-native· Cyber Essentials Plus· CFR 21 Part 11-ready· GCP-aligned· ICH E9(R1) estimands-aware· EHDS-aligned
05 · What changes for the function
From "lock the protocol and hope" to "tune the protocol and know."
FEASIBILITY CYCLE
months → the same day
Counts, geography and I/E sensitivity on demand — before the protocol is locked.
EXTERNAL-CONTROL WORK
bespoke build → reusable cohort
The same cohort logic that answers feasibility builds the external control arm.
AMENDMENT RISK
guess → measure
Re-run the cohort with the amended criteria before filing. Bring evidence, not a hypothesis.
Scope a feasibility pilot in 2 weeks