External Control Arms: When a Trial Has No Randomized Comparator, and Why That Is Hard

An external control arm compares people who received an investigational drug against patients drawn from outside the study, such as a historical cohort, a disease registry, or a natural history dataset, rather than against a randomized comparator built into the same trial. Regulators sometimes accept this design, but only in narrow circumstances, and they scrutinize it hard. The reason is simple to state and difficult to fix: when the two groups are not randomized, they can differ in ways that either manufacture the appearance of benefit or bury a real one, and no statistical adjustment fully rescues a comparison that biology and time have already skewed.

This is educational content, not medical advice, and it is written to help you read a study, not to evaluate any specific drug.

What a single-arm trial and an external control actually are

In a conventional randomized controlled trial, a coin toss assigns participants to the new treatment or to a comparator, and randomization tends to balance both the factors we can measure and the ones we cannot. A single-arm trial has no such internal comparator. Everyone enrolled receives the investigational drug, and the outcome is then compared against some external reference for how those patients would otherwise have fared.

That reference is the external control arm. The International Council for Harmonisation guideline ICH E10 describes two broad flavors. A historical control is a group of patients treated at an earlier time, before the new drug existed. A concurrent external control is a group treated during the same period but in a different setting, such as another hospital or registry. Both sit outside the trial, which is exactly what makes them external, and both carry the same core vulnerability.

Why regulators treat this as a last resort

ICH E10 is blunt: externally controlled designs raise serious concerns about the comparability of the treated and control groups and about minimizing bias, so the guideline considers them usable only in unusual circumstances. That framing has held for decades, and it explains why single-arm submissions cluster in a specific corner of medicine.

The corner is defined by three conditions that, together, make an external comparison more believable. First, a serious or life-threatening disease with genuine unmet need, where a randomized trial may be difficult or ethically fraught. Second, a disease whose natural history is well understood and highly predictable, so the untreated trajectory is known rather than guessed. Third, an expected drug effect large enough to be self-evident and closely tied in time to the treatment, so the signal dwarfs the noise that bias introduces. A published 2024 framework in Therapeutic Innovation and Regulatory Science, which examined non-oncology approvals decided by the FDA and the European Medicines Agency between 2019 and 2022, found that the successful single-arm cases overwhelmingly involved rare diseases with established natural history and limited or no standard of care, and that a large effect size was a recurring feature of the applications that succeeded.

The three problems that make it hard

Comparability

Because there is no randomization, the treated patients and the external patients can differ systematically at baseline. Trial participants are often healthier, more closely monitored, and treated in more capable centers than the historical patients they are measured against. In the 2024 analysis, a failure to achieve covariate balance at baseline between the treated and external groups was a recurring regulatory critique, and reviewers noted that differences could persist even after statistical adjustment. Matching and propensity methods can align the characteristics you thought to record, but they cannot balance the unmeasured ones the way a coin toss does. That residual gap is where a false signal hides.

Time zero

Every fair comparison needs a shared starting line, an index date or time zero from which outcomes are counted in both groups. Aligning that moment across a live trial and an external dataset is deceptively hard, because the datasets rarely mark the same clinical event the same way. Get it wrong and you can introduce immortal time bias, where the treated group appears to survive longer only because a patient had to live long enough to receive and be recorded as receiving the drug. The FDA draft guidance on externally controlled trials, issued February 1, 2023 and still a draft, devotes sustained attention to defining and aligning time zero, and the 2024 framework found that non-contemporaneous external controls drew explicit criticism in a large share of the FDA applications it reviewed.

Endpoint definition

The treated and external groups also have to measure the same thing the same way. When outcomes are subjective, or depend on an imaging read or a biomarker assessed differently across eras and institutions, apparent differences can reflect measurement rather than medicine. There is usually no blinding in these comparisons, so knowledge of who received the drug can color how outcomes are assessed. The 2024 framework reported that a meaningful fraction of FDA applications were criticized for subjective, imaging, or biomarker endpoints that lacked good reliability studies, which is one reason hard, unambiguous endpoints such as death are far easier to defend than clinician-judged ones.

How to read a single-arm claim

When you encounter a single-arm result, ask what the control actually was and how recent it is, whether the two groups looked alike at the start, how time zero was defined on each side, and whether the endpoint was measured identically and objectively. Ask, too, whether the effect is large enough to survive those doubts, because a modest difference from an external comparator is the situation in which bias is most likely to be the whole story. The FDA draft guidance is a proposal open to these very questions, not a settled endorsement of the design, and reading it that way is the honest posture.