Synthetic Control Arms: When Data Stands In For Patients

By Dan Schell, Chief Editor, Clinical Leader

Until a few months ago, I’d never thought much about how you could run a trial without actually recruiting a traditional control group. Then I talked to people who’ve done it, and I got a little obsessed.

The idea is simple in theory: You build your control group using data that already exist, instead of enrolling new patients who get placebo or standard of care. Those data might come from previous trials, disease registries, electronic health records, or insurance claims databases. Then you match that information to the patients in your experimental arm so it’s as close as possible to a real-time comparator.

I quickly learned that, although the terms “external control arm” and “synthetic control arm” are often used interchangeably, they are different. An external control arm uses data from patients who are not enrolled in the same trial as the experimental group but are observed in parallel or drawn from other real-world sources, such as registries or non-interventional studies, often reflecting current standard-of-care treatment. A synthetic control arm is a type of external control created entirely from existing historical data — such as past clinical trials, electronic health records, or claims databases — that is analyzed and matched to mimic a concurrent control group. In short, all synthetic controls are external, but not all external controls are synthetic.

But even from my novice outsider perspective, I recognized that synthetic control arms can be clinical trial game-changers — if — and this is a big if — you design them with intention and precision. That’s one of the main takeaways I got when I spoke with Debra Mitchell, Ph.D., a ClinOps veteran with 20+ years of Big Pharma experience who is now running her own consulting firm, Beofor Research Consulting. Mitchell was involved in multiple synthetic and external control arm projects, and she stresses, “You have to be intentional when choosing to use an external or synthetic control arm. Often, you’re making this decision when the standard of care hasn’t changed, or if it’s not ethical to randomize, or when you’re stuck with too-small rare-disease populations.”  Translation: This is not a “Let’s see what happens” experiment — it’s a deliberate, and sometimes high-stakes, design choice.

Some Examples of When They Work (And When They Don’t)

• Batten Disease (Brineura)
  In 2017, the FDA approved cerliponase alfa (Brineura) for late-infantile Batten disease. The trial had just 22 treated children, compared to a synthetic control group of 42 untreated patients pulled from natural history data. The match was so compelling that regulators didn’t require a traditional randomized arm. For families facing the reality of a rapidly fatal pediatric neurodegenerative disorder, this wasn’t just an innovative design, it was the only ethical way forward.
• Rare Leukemia (Blinatumomab)
  When Amgen pursued approval for Blincyto in a rare type of leukemia, they tapped historical patient records from both U.S. and EU data sources to create a comparator arm. This approach allowed them to present robust efficacy data without the delays of finding and enrolling enough control patients. In a disease where survival times are short, speed matters.
• Non-Small Cell Lung Cancer (Alecensa)
  Not all successes are in the ultra-rare space. Roche used Flatiron Health’s real-world oncology database to create a synthetic control arm for Alecensa (alectinib), used in ALK-positive NSCLC. This allowed the company to meet health technology assessment requirements in 20 European markets faster than if they’d run a separate randomized trial. It’s a good example of how synthetic arms can be a market-access accelerator in more common cancers, when the right data exist.

These cases worked because the available data matched the patient populations in the experimental arm closely enough to satisfy regulators. But the stars don’t always align. If the standard of care has shifted since your data were collected — as happens often in oncology, where new treatments like ADCs and bispecific antibodies appear rapidly — your synthetic control can become obsolete before your trial even ends.

The Messy Parts — Where You Can Misfire

I’ll admit, when I first heard about synthetic controls, I assumed the hardest part would be convincing the FDA they were legit. Mitchell quickly set me straight: The real danger is mismatched or incomplete data.

“If your trial’s data collection schema doesn’t match the external control, that’s not good,” she says. A mistake like that can mean the loss of months or years of work, not to mention millions in sunk costs.

Data quality is another recurring landmine. You can have the perfect match on patient demographics, but if half your critical endpoints are missing from the historical dataset, you’re done. “You have to ask things like how often is the data in a potential dataset collected versus the trial you're comparing it to,” Mitchell explains. “For example, maybe for your trial, you're collecting data every three months for a five-year period. If the dataset you're looking at only collected data every two months for two years, that’s not a match, and it’s not going to work for your trial.” Geography matters, too. If your control data are from a country where standard of care differs from your trial sites, you can introduce unfixable bias.

Dreaded Regulatory Maze — and the “Guidance” Mirage

The FDA published its draft guidance, Considerations for the Design and Conduct of Externally Controlled Trials for Drug and Biological Products, in February 2023. It reads like a survival checklist: specify your protocol, endpoints, and datasets up front; make sure prognostic variables are measurable and comparable; prepare your statistical analysis plan before you start; and ensure the FDA can see patient-level data — not just summaries.

Two years later, the guidance is still a draft. The delay isn’t mysterious. The FDA has been prioritizing other areas, juggling guidance documents across multiple centers, and dealing with the fallout from large-scale staff reductions. Industry insiders say many guidances, including this one, are stuck in a bureaucratic backlog. And because synthetic control arms are still viewed as a niche method, they’re not high on the agency’s immediate to-do list.

This uncertainty forces sponsors to be proactive. Mitchell’s advice — talk to regulators early, explain your plan, and ask what they’ll need to see. That way, you’re not guessing whether your approach will pass muster when it’s too late to change course.

What Actually Makes This Worth It

When synthetic control arms are executed well, they can reduce the number of patients you need to enroll, speed up studies in rare diseases, and avoid ethically questionable scenarios where patients are denied treatment. But those benefits only happen if you go in with a deep understanding of the disease, how patients are actually treated in the real world, and whether the data you’re using truly mirror your trial population. It means investing in up-front planning and planning a sustained dialogue with regulators. It’s not an easier path — in some ways it’s harder — but for the right trial, it can be the difference between a promising therapy reaching patients and a missed opportunity.

Synthetic control arms could be one of the most powerful tools in your ClinOps toolbox, but only if you treat them like the precision instrument they are, not a buzzword. They’re not faster by default. They’re not easier. But they can enable you to do more with less — if — and this is another big if — you have the discipline to pull it off.