Variacle combines the data you already hold with the data your study will collect — completed trials, registries, real-world cohorts — into a comparison regulators can follow. Fewer patients in the control arm, no randomization where it is not feasible or ethical, and one team from design to analysis to the regulatory submission.
Assumed bias δ = 0.20beyond the tipping point+0.6 (-0.7–+1.9)
-101234
← Favours controlFavours treatment →
The conclusion holds until the external comparison is biased by δ ≈ 0.16 on the restricted-mean-survival-time scale — a magnitude your clinicians can judge.
Who we work with
One method, three ways in.
Sponsors pay for the evidence, CROs deliver it, investigators design the studies that need it. Each gets a different deliverable from the same rigorous core.
External comparators are already in approved dossiers.
What separated accepted comparators from rejected ones was process: the comparator, time zero and the analysis plan fixed in advance. That is the part we own.
ICH E9(R1) — estimands and sensitivity analysis: the shared language we write in.
FDA draft guidance (2023) on externally controlled trials: comparability, time zero, outcome assessment, pre-specification.
EMA started work in 2026 on a reflection paper on external controls.
European Health Data Space (Regulation (EU) 2025/327): secondary use of health data in secure environments from 2029.
MedicineAgencyExternal evidence in the dossier
Skyclarys (omaveloxolone)
Friedreich's ataxia
EMA · FDA
Open-label extension compared with a propensity-matched natural-history cohort, as supporting evidence
Wainzua (eplontersen)
Hereditary transthyretin amyloidosis
EMA
Placebo group carried over from an earlier trial in the same programme
Elrexfio (elranatamab)
Multiple myeloma
EMA
Single-arm study compared with a real-world external cohort using weighting, doubly robust estimation and E-values
Orencia (abatacept)
Prevention of acute graft-versus-host disease
FDA
Registry data as pivotal evidence of effectiveness
Public precedent, not our work. Sources: EMA clinical data publication; FDA real-world evidence examples.
The problem
Nine in ten trials fail. The evidence they produced is rarely used again.
Every completed or failed trial leaves behind randomized control patients, measured to protocol and then archived — maintained at a cost and almost never reused. New studies recruit fresh controls for questions existing data could partly answer, or stall because a control arm is not feasible at all.
Regulators now accept external comparators when randomization is impractical. What they do not accept is a comparison assembled after the fact. The value is in doing it properly, and early.
90%
of drug candidates that enter clinical trials never reach approval
Sun et al., Acta Pharm Sin B, 2022
$41117
median cost per patient in pivotal trials
Moore et al., JAMA Intern Med, 2018
How it works
Four steps, each with a fixed scope and a decision point.
You decide at the end of every step whether to continue. Nothing is analysed before the plan is written and dated.
STEP 01
Scientific advice
Before the protocol is written: a short synopsis of the design, the data and the method, ready for scientific advice with the EMA, a national agency or the FDA. You learn whether the regulator will follow before you commit.
DeliverableSynopsis and briefing materials
STEP 02
Feasibility and design
We audit the data you hold or can access, and the data your study will collect: comparability, endpoints, time zero, missing data and sources of bias. If an external control is not viable for your question, this step says so in writing.
DeliverableReport with risks and red flags, proposed design and a simulation of the final analysis
STEP 03
Protocol and submission
The statistical sections of the protocol and the analysis plan. If you have no regulatory team, the full Part I and Part II dossier, submitted through CTIS on your behalf, with the regulators' questions answered on time.
DeliverableProtocol statistics, SAP and, where needed, the CTIS application
STEP 04
Pre-specified analysis
The analysis runs as planned, alongside your study, with a sensitivity analysis for every identifying assumption and the documentation trail the dossier needs.
DeliverableStatistical report and regulatory package
Optional modules: sourcing and access requests for external data, and medical writing with a partner.
What it looks like
Your own placebo arms, put back to work.
Placebo arms from a sponsor's own randomized trials are the strongest external control there is: randomized, measured to protocol, already paid for. Most external comparators in published dossiers come from health records or registries; prior randomized arms need lighter assumptions.
External historical control for a single-arm study
The comparator is built from the placebo arms of a sponsor's own completed randomized trials and transported to the population of the new study. Evidence already paid for, put back to work.
What you receive
Estimands written to ICH E9(R1), with every identifying assumption stated one by one
A sensitivity analysis for each assumption, relaxed one at a time, with its tipping point and E-values
Independent double programming of every derived quantity
A reproducible package: versioned code and checksummed inputs, so every reported number can be regenerated
Why identification
More data does not fix a biased comparator.
Pool more historical patients and the confidence interval narrows — around whatever bias the comparison carries. We start from identification: which assumptions make the external control answer your trial's question, and how far each one would have to fail before the conclusion changes.
The unit that varies is the study, not the patient. In one well-known analysis1, 1279 historical patients from 19 studies carried about as much information as 50 concurrent controls. We quote effective sample sizes, not headcounts.
“Identification avoids a confidence interval that becomes smaller and smaller around something that is wrong.”
Naive pooling Identified, with between-study variance True effect
-1.0-0.8-0.6-0.4-0.200.2
10040016006400
Historical control patients
Illustrative simulation, not client data. Naive pooling becomes precise around a biased value. An identified estimator stays centred on the truth, and its interval stops shrinking at a floor set by the number of studies, not the number of patients.
Patient-level data is the most sensitive thing you own. Our default is that it never leaves your control.
Data stays where it is
Where patient-level data cannot move, the analysis goes to it: a versioned package that runs inside your environment, or your data partner's, and returns aggregates only.
Reproducible by construction
Versioned code, checksummed inputs and independent double programming. Every number in a report can be regenerated and traced to its source.
Documented quality system
SOPs for deliverable review and approval, change control, data protection and business continuity, aligned with ISO 9001 principles and with ICH E6(R3) where it applies to our role.
Contracts that fit your MSA
GDPR data processing agreements and documented technical and organisational measures that slot into your vendor framework and supplier questionnaire.
Questions
What sponsors ask us first.
Will a regulator accept an external control?
When randomization is not feasible or not ethical, regulators have accepted external comparators as pivotal or supportive evidence. Acceptance depends on process: the comparator, time zero and the analysis plan fixed in advance, and sensitivity analyses that show how robust the conclusion is. That is why we start with scientific advice, so you know where the agency stands before you commit.
Our biostatistics team already does this. Why add you?
We do not replace them. Your team keeps the primary analysis; we add the identification argument, the register of assumptions and the sensitivity package — the parts that are hardest to certify from inside. Sharper, not substituted.
Can you handle the regulatory submission in CTIS?
Yes, for sponsors without a regulatory team — including hospitals that sponsor investigator-initiated trials and small biotechs. We prepare Part I (protocol, statistical methods, comparator) and Part II (informed consent, sites, insurance, financial arrangements, data protection), submit through CTIS with a role delegated by the sponsor, and answer the regulators' requests for information within the 12-day deadline. Legal responsibility stays with the sponsor, as the EU Clinical Trials Regulation requires.
Can we add an external control to a study that is already running?
Not as confirmatory evidence: an external control has to be pre-specified. For a running or closed study we can run a retrospective or exploratory analysis. It shows what your data could support next time, without touching the current submission.
Do you provide the data?
We are a methods company, not a data vendor. We work with data you own or can access — completed trials, registries, real-world cohorts. When external data is needed, we find suitable sources and handle the access requests.
What happens if the study is negative?
Method and result are kept apart. The estimand, the assumptions and the tipping points are fixed and dated before outcomes are seen, and the risks are flagged in writing at the start. A negative result is then a finding about the treatment, not an artefact of the comparator.
How is it priced?
A fixed fee per step, agreed before each step starts and sized as a fraction of what the alternative study would cost. You decide at the end of each step whether to continue. The first conversation is free.
Supported by
Find out what your data can support.
A 30-minute call: whether an external control is viable for your question, what data could serve, and what regulators have accepted in similar settings.
