Harnessing the Potential of Synthetic Control Arms
A synthetic control arm (SCA) — more precisely, an external control arm — replaces a trial’s placebo or standard-of-care group with a comparator built from real-world and historical data, instead of newly enrolled control patients. It lets a study estimate treatment effect without randomizing patients to a control they’d rather avoid — valuable when a placebo is unethical, enrollment is slow, or the condition is rare.
The idea isn’t new. Oncology has long used external controls where withholding treatment would be unethical, drawing the comparator from patients on standard of care. What’s changed is the data infrastructure that makes it rigorous enough for a regulator to consider.
Synthetic, external, or historical control — what’s the difference?
The terms get used loosely; the distinctions matter at submission.
| Term | What it means |
|---|---|
| Historical control | Comparator drawn from patients treated in the past, often from prior trials |
| External control | The regulator-preferred umbrella term — any control outside the current randomized trial, from real-world or historical sources |
| Synthetic control arm | An external control assembled and standardized from multiple real-world data sources to mirror the treatment arm’s characteristics |
FDA and EMA frame this space around externally controlled trials, and FDA has issued guidance on their design and conduct — with a consistent message: external controls are acceptable in the right circumstances, but comparability between the treatment and external groups is the burden you have to meet.
How does a synthetic control arm work?
It comes together in stages, and each stage is a data problem before it’s a statistical one.
1. Collect and standardize the data. Health data now comes from electronic health records, claims, patient-generated data from wearables and home devices, registry studies, and historical trial data. On their own they’re incompatible — different formats, vocabularies, and quality. Standardized and combined into one database, they become usable comparator material. This is the step that fails most often, and it fails silently.
2. Build a synthetic registry. A normalized database of standardized patient records becomes a registry you can query. In a trial investigating an HIV therapy, for instance, an investigator could search a registry of 10,000 patients on lab values and demographics to identify those who meet the trial’s inclusion criteria — assembling a matched comparator without enrolling a single new control patient. (Illustrative example.)
3. Match and analyze. The external cohort is matched to the treatment arm on the characteristics that drive outcomes, and the treatment effect is estimated against it. The quality of that matching — not the size of the dataset — is what determines whether the evidence holds up.
What are the benefits — and the limits?
The benefits are real. A well-built external control can reduce or remove the need to recruit control patients. In a trial needing a 500-patient treatment arm, an SCA model means locating 500 participants for the treatment arm instead of recruiting 1,000 (500 treatment + 500 control) — cutting cost, shortening timelines, and, most importantly, getting therapies to patients faster. BCG has documented these efficiency and cost advantages across applications. [confirm: BCG 2021 “Synthetic Control Arms Changing Clinical Trials”]
The limits are just as real, and worth naming. External controls are not a default substitute for randomization. Regulatory acceptance is case-by-case. The central risk is confounding — if the external cohort differs from the treatment arm in ways the data doesn’t capture, the comparison misleads. That’s why data provenance, standardization, and pre-specified matching aren’t housekeeping; they’re the whole game.
What does a synthetic control arm actually require?
Everything above depends on one thing: the ability to pull disparate real-world data into a single, standardized, governed, submission-ready database. A platform like REDCap Cloud — built to assimilate real-world data from diverse sources and normalize it into synthetic registries — is what turns the concept into evidence a regulator will read. Without that foundation, a synthetic control arm is an idea; with it, it’s a defensible comparator.
A synthetic control arm replaces a trial’s placebo or standard-of-care group with a comparator built from standardized real-world and historical data, rather than newly enrolled control patients. It is a type of external control used when randomizing to a control group is impractical or unethical.
External control is the broader, regulator-preferred term for any comparator outside the current randomized trial. A synthetic control arm is a specific kind of external control, assembled and standardized from multiple real-world data sources to mirror the treatment arm.
Regulators accept external controls in appropriate circumstances — commonly rare diseases, oncology, or where a placebo is unethical — but acceptance is case-by-case and hinges on demonstrating comparability between the treatment and external groups. FDA has issued guidance on externally controlled trials.
When a placebo would be unethical, when the condition is rare and enrollment is slow, or when recruiting a full control group is prohibitively costly — and when you have real-world data of sufficient quality to build a comparable external cohort.
Standardized data from sources such as EHRs, claims, registries, wearables, and historical trials — normalized into a single database so patients can be matched to the treatment arm on outcome-driving characteristics.
