When Is an External Control Group Acceptable in Research?

Can Participants Outside Your Study Provide a Credible Control Group?

Suppose researchers are conducting a single-arm clinical trial of a new intervention for a rare disease. Everyone enrolled receives the experimental treatment. Because recruitment is difficult, the investigators consider using patients from an existing disease registry as a comparison group.

The registry contains hundreds of patient records, including diagnoses, treatments, and clinical outcomes. At first glance, it seems to offer a practical alternative to recruiting and randomizing additional participants.

But the registry patients were not enrolled under the same protocol. Their eligibility criteria may differ, their outcomes may have been measured differently, and their treatment decisions may reflect characteristics that also influence prognosis.

The central question is whether people observed outside the trial can represent what would have happened to trial participants under the comparator condition. External controls can be useful, but their credibility depends on how convincingly researchers address that question.

External Controls Are Acceptable When They Support a Credible Treatment Comparison

How External Controls Work and What Determines Their Validity

What Is an External Control?

An external control is a comparison group drawn from outside the focal intervention study rather than formed through the study's own concurrent randomized allocation process.

In clinical research, externally controlled trials commonly compare participants receiving an investigational intervention under a defined protocol with individuals whose outcomes are obtained from another study, registry, healthcare database, or documented population.

The International Council for Harmonisation (ICH) E10 guideline distinguishes external controls from internal concurrent controls. External participants may have been observed before the intervention trial or during approximately the same period in another setting.

The defining issue is the external origin of the comparator, not whether its data were collected prospectively or retrospectively.

How Are External Controls Different From Historical Controls?

Historical controls are defined by when the comparison observations occurred. External controls are defined by their relationship to the focal study.

Many historical controls are external controls, but not all external controls are historical.

A concurrent external comparator may reduce some calendar-time differences, but it does not eliminate selection bias or differences in treatment settings. Conversely, historical-control evidence requires particular attention to secular trends and changing practices.

Where Can External-Control Data Come From?

Several sources may be suitable, depending on the research question and available information.

• Previous clinical trials: Individual participant data from earlier studies may provide well-defined eligibility criteria and outcomes, although trial protocols and populations may differ.
• Disease registries: Registries may provide longitudinal information on diagnosis, treatment, and outcomes for patients with a particular condition.
• Electronic health records: Clinical records may contain relevant treatment histories and outcomes, but their completeness and measurement consistency require assessment.
• Administrative and claims databases: These can provide broad coverage of healthcare utilization and some outcomes, although clinical detail may be limited.
• Natural history studies: Observational studies describing disease progression may be useful when the untreated or conventionally treated course is sufficiently characterized.

In educational research, external comparisons might use students from another institution, a previously evaluated program, or an independently maintained administrative dataset.

Regardless of the source, researchers must establish that the data contain the variables necessary to construct a meaningful comparison. Dataset size alone does not establish suitability.

When Might an External Control Be Preferable to an Internal Control?

External controls are sometimes considered when a randomized controlled trial would be difficult to conduct or when recruiting additional control participants would substantially delay research.

Potential applications include rare diseases, conditions with limited eligible populations, and certain studies of interventions where randomization is impractical.

They may also supplement randomized evidence by providing information about outcomes in routine practice or populations not adequately represented in the original trial.

However, these advantages do not mean external controls are methodologically equivalent to randomized controls. Researchers should first establish what uncertainty the external comparison is intended to resolve and whether the available data can support that inference.

In regulatory settings, acceptability is particularly context-dependent. The U.S. Food and Drug Administration's February 2023 draft guidance on externally controlled trials discusses design and analytical considerations but does not establish blanket acceptance of external controls for demonstrating effectiveness.

What Does It Mean for External Data to Be Fit for Purpose?

Fit-for-purpose data are sufficiently relevant and reliable for the particular research question and analysis.

Relevance concerns whether the data represent an appropriate population, intervention or comparator, outcomes, and follow-up period.

Reliability concerns whether the information was collected and recorded with sufficient accuracy, consistency, completeness, and traceability.

A dataset may be excellent for describing disease prevalence but inadequate for estimating a treatment effect if it lacks important prognostic variables or consistently measured outcomes.

Similarly, a student information system may reliably record final grades while providing little information about prior academic preparation, access to educational resources, or participation in competing interventions.

Researchers should assess the dataset against the intended analysis rather than assuming that an established registry or institutional database is inherently research-ready.

Why Must Eligibility Criteria Be Aligned?

Participants in the intervention and external-control groups should be selected using comparable eligibility rules whenever possible.

Suppose the experimental trial enrolls patients with advanced disease who have already failed two previous treatments. If the external registry includes newly diagnosed patients with milder disease, the resulting outcome difference may reflect prognosis rather than treatment effectiveness.

Researchers should reconstruct the trial's eligibility criteria within the external data, including relevant diagnoses, disease stages, previous treatment, age restrictions, and other important factors.

This requires information that was actually recorded. If eligibility cannot be determined from the external dataset, the resulting uncertainty should be treated as a substantive design limitation.

Researchers must also consider whether the eligible external population sufficiently overlaps with the intervention population. A comparator may be too different from the intervention group to support a credible treatment-effect estimate.

Why Is Time Zero So Important?

Time zero is the point at which eligibility, treatment strategy, and the beginning of outcome follow-up are aligned for the intended comparison.

In a randomized trial, this commonly corresponds to randomization. In an externally controlled study, researchers must construct a comparable starting point for participants in the external dataset.

For example, the intervention group might begin follow-up when treatment starts, while external patients begin follow-up at diagnosis. If these points represent different stages of disease, outcome comparisons may be distorted.

Time-related biases can also arise if patients must survive or remain eligible for a period before being classified into a treatment group.

One particularly consequential problem is immortal time bias. This occurs when a period during which the outcome could not have occurred under the study's classification rules is incorrectly assigned or handled, potentially favoring one treatment group.

Aligning eligibility, treatment assignment, and follow-up is therefore central to a credible external comparison.

How Can Target Trial Emulation Improve the Design?

Target trial emulation is an approach to observational causal inference in which researchers explicitly describe the hypothetical randomized trial they would ideally conduct and then attempt to reproduce its key design features using available data.

The target trial typically specifies eligibility criteria, treatment strategies, assignment procedures, follow-up, outcomes, causal contrasts, and the analysis plan.

For external controls, this approach encourages researchers to define the comparison before examining treatment effects and to align participants at a common starting point.

It also helps reveal situations where the available data cannot adequately reproduce the desired trial. For example, a registry might lack the information needed to determine eligibility at treatment initiation or distinguish between relevant treatment strategies.

Target trial emulation improves design transparency but does not make observational data randomized. Its validity still depends on assumptions such as sufficiently controlled confounding, consistency, positivity, and accurate measurement.

What Is the Main Threat From Confounding?

Confounding occurs when characteristics associated with both treatment assignment and the outcome distort the estimated treatment effect.

External controls are particularly vulnerable because participants were not randomized between the experimental and comparison treatments within the focal study.

Patients receiving a new treatment may differ from external patients in disease severity, access to specialist care, previous treatment, socioeconomic circumstances, or other prognostic factors.

Researchers can attempt to address measured confounding through regression adjustment, propensity score matching, weighting, standardization, or other suitable methods.

However, these methods depend on the availability and quality of relevant covariates. They cannot guarantee removal of unmeasured confounding.

Furthermore, propensity score methods do not establish comparability merely because a balance statistic meets a conventional threshold. Researchers must consider which variables were measured, whether they were measured before treatment, and whether the resulting comparison remains clinically meaningful.

Why Can Outcome Measurement Differ Across Data Sources?

Experimental trials often use standardized outcome definitions, scheduled assessments, and protocol-specific procedures. External databases may rely on routine clinical documentation or administrative reporting.

For example, trial participants might undergo scheduled imaging at fixed intervals, while registry patients receive imaging only when clinically indicated. Apparent differences in disease progression could partly reflect different opportunities to detect progression.

In educational research, one institution might use standardized assessment rubrics while another relies on course-specific grading practices.

Researchers should evaluate whether outcomes can be defined consistently and whether ascertainment procedures introduce systematic differences.

When measurement procedures cannot be harmonized, statistical adjustment may not resolve the resulting bias.

Should the External Comparator Be Selected Before Seeing the Results?

Preferably, yes. Researchers should prespecify the external data source, eligibility rules, outcome definitions, and analysis plan before examining comparative outcomes.

Selecting a comparator after observing the intervention results creates opportunities to choose datasets or analytical decisions that favor a particular conclusion.

The FDA's draft guidance on externally controlled trials emphasizes the importance of addressing design issues before the externally controlled trial is conducted, including data-source selection and the analytical approach.

Researchers should document feasibility assessments, justify the inclusion or exclusion of candidate datasets, and maintain a transparent record of analytical decisions.

Where exploratory analyses are unavoidable, they should be distinguished from prespecified confirmatory analyses.

Can External Controls Be Combined With Randomized Controls?

Yes. Some designs investigate whether external data can supplement an internal randomized control group.

Potential approaches include borrowing information from external controls through prespecified statistical models or using external evidence to improve estimation under suitable assumptions.

Such methods require particular caution because external participants may not be exchangeable with randomized controls.

Borrowing strategies can reduce uncertainty when external information is compatible with internal data, but they can also introduce bias when the external population differs in important ways.

Researchers should justify the borrowing model, evaluate compatibility, and examine how conclusions change under different assumptions about external-data contributions.

The presence of some randomized participants does not automatically validate the external evidence.

What Makes External-Control Evidence Too Weak for a Causal Conclusion?

External-control evidence becomes difficult to interpret when important prognostic variables are unavailable, treatment and outcome definitions are incompatible, eligibility cannot be reconstructed, or there is little overlap between the groups.

Additional concerns arise when the intervention and comparison groups differ substantially in calendar time, healthcare setting, follow-up intensity, or access to supportive treatment.

These problems are especially consequential when the expected treatment effect is small relative to plausible sources of bias.

Researchers should not assume that a sophisticated statistical model can compensate for an unsuitable comparison. Sometimes the defensible conclusion is that the available external data cannot answer the intended causal question.

Using Registry Patients as an External Control for a Single-Arm Trial

The investigators then examine whether the groups were genuinely comparable.

They discover that the trial excluded patients with severe comorbidities, while the registry included them. The trial also required confirmation of disease stage using a newer diagnostic procedure that was not routinely available when some registry patients were enrolled.

Furthermore, trial follow-up began at treatment initiation, whereas registry follow-up was originally recorded from diagnosis.

These differences affect eligibility, prognosis, and the interpretation of the survival endpoint.

To improve the design, the researchers attempt to reconstruct equivalent eligibility criteria, align the start of follow-up, harmonize outcome definitions, and adjust for measured prognostic factors.

If important information remains unavailable, the resulting estimate may still be substantially biased. The investigators should report those limitations and avoid presenting the observed 15-percentage-point difference as an established treatment effect.

Where a defensible comparison cannot be constructed, the registry may remain valuable for describing disease progression without supporting a reliable comparative effectiveness claim.

Common Errors in Externally Controlled Research

How to Decide Whether an External Comparator Is Worth Using

Begin by defining the hypothetical comparison you want to make. Specify who would be eligible, which treatment strategies would be compared, when follow-up would begin, and what outcomes would be measured.

Then assess candidate external datasets against those requirements. The data should determine whether the comparison is feasible, but they should not be allowed to redefine the research question merely to produce a favorable result.

Where multiple external datasets are available, selection should prioritize relevance, reliability, and compatibility rather than the apparent size of the treatment difference.

Researchers should also consider whether an additional comparison group would provide useful evidence about the robustness of their conclusions.

External controls may be particularly valuable when they make otherwise unavailable comparative evidence possible. Their usefulness, however, should be judged by the credibility of the causal contrast rather than by the convenience of accessing existing records.

Before Using an External Control, Verify the Essential Design Requirements

Questions About External-Control Groups and Real-World Data

External Controls Are Credible Only When the Comparison Can Be Justified

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.