Points to consider for externally controlled trials — PMDA Early Consideration Summary

In Plain Language

An externally controlled trial compares people receiving a test treatment with a separate group of people who did not receive it, rather than with a control group randomized within the same trial. PMDA explains that this approach may be considered when a randomized controlled trial is difficult, the disease course is predictable, and enough is known about factors that influence that course. The separate groups can differ in patient characteristics, care, timing, and how outcomes are assessed, so an apparent treatment effect may partly reflect bias rather than the treatment. The document therefore asks developers to plan the comparison carefully, describe the external control selection and analysis in advance, and make the groups as comparable as possible. It also emphasizes that analysis cannot remove every source of bias, and that another trial design should be considered if bias cannot be sufficiently minimized.

What This Guidance Covers

The document concerns clinical trials designed with the intention of comparing a test treatment group with an external control group. It assumes that the results are intended to help evaluate a drug’s efficacy and safety and may be used as part of documentation for marketing approval applications and related purposes. An external control can consist of patients treated before the new trial (historical controls) or patients treated under different conditions during the same period. Possible data sources include groups from other clinical trials and real-world data (RWD), such as registry data.

The document assumes individual participant- or patient-level data are available for the external control, regardless of the data source. It does not apply to comparisons of an open-label single-arm trial’s results against a specific threshold. It also does not address so-called hybrid controls, in which external control data are combined with the control arm of a randomized controlled trial. However, it says the document can serve as a reference for issues in such designs, including whether the external control and randomized control are sufficiently similar to pool, and whether the test treatment group is comparable with the combined control group.

The document does not address the reliability of RWD used as an external control. It points readers to relevant notifications for that subject. It also notes that different considerations apply depending on the data source, and supplements the points about registry use in the named registry-related documents.

Why It Matters

Randomization helps make groups comparable, and blinding can reduce the potential for bias in comparisons. In an externally controlled trial, the treatment group and the external control are not randomized concurrently. Their patient populations may therefore differ, including in known or unknown factors that affect disease progression or outcomes. The trial is also generally an open-label, single-arm trial, meaning participants and investigators know which treatment is being given. That knowledge, as well as retrospective selection of external data, can affect participant selection and outcome assessment.

These issues limit how confidently a difference between groups can be attributed to treatment. Even if developers try to improve comparability, strict control of bias remains difficult. If the observed difference is small, the document says it may be difficult to determine whether a treatment effect has been demonstrated, because potential bias could account for the difference. PMDA therefore frames an external control as a design to consider in selected circumstances—not as a straightforward substitute for a randomized comparison.

Key Recommendations

1. Consider the design selectively. Randomized controlled trials should be conducted when feasible as the trials providing the main evidence of efficacy and safety for applications. An externally controlled trial should be considered when a randomized controlled trial is difficult, the disease or condition has a predictable course based on available knowledge, and sufficient information is available about factors that may influence disease progression. The document also says use should be considered where, based on information already available about the test treatment, a certain degree of treatment effect is reasonably expected—even allowing for bias—so that a conclusion about treatment effect may still be drawn.
2. Plan to minimize bias before the trial begins. The protocol and related documents should specify the external control population or the method for selecting it, the statistical analysis plan (including confounder adjustment), and the criteria for trial success based on statistical analysis. This should be done before the first participant is enrolled in the test treatment group. Changes during the trial should be avoided; if unavoidable, their timing, rationale, and content must be clearly documented and appropriately justified.
3. Choose and justify the data source. For control data from another clinical trial, assess comparability using available protocols and information about eligibility criteria, treatment, and endpoint definitions. Consider differences in objectives, evaluations, interpretation, and the timing of the trials, as well as changes in disease assessment or management over time. Explain why the selected trial is appropriate and was not chosen arbitrarily, particularly when its results are already available or published. For RWD, consider the data’s suitability and limitations, including how it was collected and whether important confounding factors are recorded. The document strongly recommends early consultation with PMDA about RWD use for purposes such as applications.
4. Examine comparability across the trial’s key elements. This includes patient populations and confounding factors; treatment details and concomitant therapies; when data were collected; index dates and observation periods; endpoints and how they are evaluated; intercurrent events; and the number of participants or patients available. For each, consider whether the groups and measurements are sufficiently similar to support an interpretable comparison.
5. Select appropriate statistical methods and explain them. The document does not recommend a specific statistical method. Methods should be selected in light of the design, data source, and data characteristics, with their appropriateness, assumptions, and implementation described. Prespecify the analysis and, where appropriate, sensitivity and supplementary analyses to examine the influence of assumptions and potential biases.
6. Reconsider the design if bias cannot be adequately reduced. If planning indicates that evaluation bias cannot be sufficiently minimized, the document says alternative trial designs should be considered. It also strongly recommends seeking PMDA clinical trial consultation in advance about whether an externally controlled trial can support an application and when principles in the document may be difficult to address.

Practical Implications

Interpretation: Teams will need to treat selection and analysis of an external control as central design decisions, not as steps that can be settled after treatment-group results are known. The document’s emphasis on advance specification is intended to reduce opportunities for choices to be influenced by outcomes. Teams should therefore identify who can access external-control outcome data while the population and analysis approach are being planned.

Interpretation: Feasibility work can help determine whether a data source contains enough eligible patients and whether relevant data have been collected. But the document cautions against detailed investigation of endpoint results or closely related variables during that work, to avoid bias from prior knowledge of external-control outcomes. If a feasibility study is conducted, its timing and content should be documented and available for explanation in regulatory submissions.

For clinical operations and data teams, the implications include checking whether external data capture treatment, follow-up, endpoints, and important confounders in enough detail. For statisticians, the work includes defining the estimand—the precise description of the treatment effect the trial intends to estimate—then prespecifying the comparison, adjustment methods, and plans for missing data and sensitivity analyses. Regulatory teams should ensure that the rationale for the data source and any differences between groups can be clearly described. These are practical interpretations of the considerations in the document, not additional PMDA requirements.

Technical Details

The document organizes trial planning around the challenge of comparability. Important technical points include:

• Population and confounding. Identify potential confounding factors—factors associated with group differences that could influence the estimated treatment effect—and check whether the external source records them, how they are measured, and how they are assessed. Where possible, apply the trial’s inclusion and exclusion criteria to the external population. Describe the selection method, including any matching techniques, so that the rationale for choosing a similar population is clear and not arbitrary. For multiregional trials involving participants from outside Japan, the document says to take into account the named ICH guidelines on ethnic factors and multiregional trial planning.
• Treatment. Examine dosage, treatment duration, adherence, and concomitant therapies in each group. Clinical trial protocols may provide details for trial-derived controls, while RWD may not record them or may contain incomplete information. Differences between clinical-trial settings and routine medical practice may also affect outcomes.
• Timing, index date, and follow-up. Standards of care, diagnostic criteria, concomitant therapies, and disease assessments can change over time. Consider whether differences in the data-collection periods affect comparability. If prospective collection of external-control data is feasible, the document recommends considering data from the same period as the test treatment group, while also accounting for differences in patient background. Define the index date—the start of the observation period—so that groups are not followed from meaningfully different points. Ensure the external data support the required observation period.
• Immortal time bias. The document highlights this concern for time-to-event endpoints. Its example is a comparison of a treated group with an untreated RWD control when death is the event of interest. If the treated group’s index date is the failure of prior therapy, time between that failure and starting the test treatment can be counted as immortal time for that group. The untreated control’s time from prior-therapy failure to death is observed directly. The index-date definition should avoid evaluating one group in a way that includes such immortal time.
• Endpoints and assessment. Select endpoints that address the trial objective and clinical question, and make them objective and clearly defined. Blinded evaluation or an independent review facility may be useful in some cases. Ensure the endpoint and evaluation are consistent across groups. For RWD, examine whether endpoint definitions, evaluation criteria, methods, timing, and frequency are sufficiently clear and standardized to support a valid comparison, and justify the endpoint’s appropriateness both in planning and when interpreting results.
• Intercurrent events. These are events, such as stopping assigned treatment or using an alternative treatment, that affect interpretation or the existence of measurements. Anticipate them and consider strategies when defining the treatment effect. The document says it is desirable for these events to be observed with similar rigor in both groups, but recognizes this may not be possible, particularly with RWD. Consider what can be captured and how differences or incomplete records may affect evaluation.
• Sample size and feasibility. The sample size for both the test treatment group and the external control group must be sufficient to achieve the primary objective of the trial. Because applying eligibility criteria or matching may leave few suitable external controls, conduct a feasibility study of available patient numbers and data collection. Avoid detailed examination of endpoint results or closely related variables in that study.

For analysis, prespecify the statistical analysis plan before trial initiation and describe it in the protocol and/or analysis plan, depending on its content. When historical control outcomes are available, the document gives as an example of a safeguard ensuring that people planning the analysis are not exposed to those outcomes when defining the plan. In principle, results obtained using statistical methods prespecified before trial initiation are considered to have the highest level of scientific credibility for interpretation. If changes to the analysis plan are made for unavoidable reasons, the timing, reasons, and details of the changes must be clearly documented and justified.

A propensity score may be used to compare groups while adjusting for confounding factors. If used, prespecify the covariates and estimation methods. If the plan allows scores to be re-estimated when the original approach does not adequately balance patient characteristics, define the imbalance criteria and re-estimation procedures in advance. This is important because knowing external-control outcomes when estimating scores may raise concerns that scores were derived after the fact to produce favorable results.

Identify missing data in light of the estimand and plan how to handle them. Missing information may be more common in historical data or RWD, including missing information about intercurrent events. Assess the impact of missing data, including whether their amount makes results less robust or harder to interpret. Plan sensitivity analyses to examine deviations from statistical assumptions, including assumptions about the missing-data mechanism, and supplementary analyses under varying conditions. The document treats advance planning and conduct of these analyses as important to interpreting results that rely on assumptions to address potential bias.

Who Should Read It

This document is relevant to sponsors considering a single-arm trial with an external comparison, especially when a randomized controlled trial is difficult. It will also be useful to statisticians designing population-selection and adjustment strategies; clinical operations and data teams evaluating the completeness and timing of trial or registry data; investigators and endpoint reviewers assessing outcome measurement; and regulatory professionals preparing or reviewing evidence for an application. It is particularly pertinent when using historical trial data or RWD, and when deciding whether the available information can support a credible comparison. PMDA strongly recommends advance clinical trial consultation meetings when considering use of these results as evidence in an application or when the document’s principles appear difficult to address.

Official PMDA Source

• Official title: Points to consider for externally controlled trials (Early Consideration)
• Issuing body: Pharmaceuticals and Medical Devices Agency, Center for Product Evaluation
• Date on cover: March 24, 2025
• PMDA PDF: Official PMDA document
• PMDA list page: Early Consideration documents

Source: Pharmaceuticals and Medical Devices Agency (PMDA) website, PMDA list page. This page was created by editing and summarising the PMDA document.

Consult the official document; this summary does not replace it. PMDA provides the English version for reference purposes, and the Japanese original prevails if the two differ. Early Consideration documents are reference information reflecting PMDA’s views at the time of publication; those views may change as scientific knowledge develops.

Related Clinical Biostats Resources

• Bayesian Sample Size Using Historical Borrowing (Power Prior) — relevant to a statistical approach involving historical information, although the PMDA document does not recommend a specific external-control method.
• Prior Distributions: Choosing and Justifying Priors — provides background on specifying and justifying prior distributions in Bayesian analyses.
• Understanding ICH E9 Statistical Principles — offers broader context on statistical principles for clinical trials.
• All Clinical Biostats tutorials — browse the wider tutorial collection.
• Statistical Calculators — access the site's statistical calculation tools.