External and historical control studies
External control studies compare treated participants with controls from another source. Historical controls come from an earlier period, when care and measurement may differ.
Why this level?
Specify the research question: intervention, diagnostic accuracy, prognosis, aetiology, or screening intervention. Other question types are outside this hierarchy. NHMRC 2009 sourceExternal controls can serve different clinical questions and borrowing schemes. Their source, comparability and design facts determine the evidence; a large control pool does not raise it.
Fit
When to use it, and when not to
Use it when
- A concurrent randomized comparison is infeasible, and a credible external data source addresses the same clinical question.
- You can align eligibility, treatment strategies, follow-up and outcome definitions across sources.
Not when
- You can randomize participants to concurrent treatment strategies. Use a randomized trial to reduce baseline confounding. Read: Randomized controlled trial
- You only want to describe outcomes in a treated group. Use a case series and avoid a comparative treatment claim. Read: Case series
Before you start
What you need
- The trial you would ideally runDefine eligibility, treatment strategies, assignment time, outcome horizon and effect of interest before selecting external controls.
- Where and when controls were treatedAssess access, care standards, measurement schedules and why records entered the source. Earlier care may differ from current care.
- A common time zero and outcome processApply eligibility and start follow-up at comparable times. Align outcome definitions, censoring and surveillance intensity.
- Confounders recorded before treatment in both sourcesCheck overlap in prognostic factors. Weighting or matching cannot recover an absent comparison or adjust unrecorded causes.
- Measured factors suffice to compare treatment groups fairly.How to check: Justify why important prognostic causes are recorded in both sources. Examine residual confounding and sensitivity to unmeasured differences.
- Comparable controls exist for the treated population.How to check: Inspect overlap and extreme weights. Report who is excluded, and how that changes the target population.
- Outcomes and follow-up are comparable across sources.How to check: Check calendar changes, missing visits and assessment schedules. A trial scan schedule can detect events sooner than routine care.
- Missing records are handled under a named assumption.How to check: Missing completely at random (MCAR) means gaps are unrelated to recorded and unrecorded values; this is rarely plausible. Missing at random (MAR) means gaps depend only on recorded factors. Under MAR, consider multiple imputation; it is not automatically better than a justified complete-case analysis. Missing not at random (MNAR) is plausible if unobserved values themselves influence whether records are absent. Under MNAR, test departures with a delta or tipping-point analysis that shifts the unobserved values.
Plain language
How it works
Design both arms around the same target trial, including a shared time zero. Select controls without using their later outcomes. Compare eligibility, baseline factors, care era and outcome capture. Estimate a contrast with a prespecified model, matching or weighting when justified. A single-arm response rate against a fixed benchmark answers a narrower question than an adjusted external-arm comparison.
Larger external datasets reduce some random error but cannot erase systematic differences. If borrowing earlier controls into a new trial, state how compatibility is assessed and borrowing is limited. Consider reduced or no borrowing when sources conflict. Report uncertainty and sensitivity to source choice, confounding and missing data.
Worked example
A worked example
- Setting
- A treatment cohort is compared with eligible historical records at a common follow-up horizon.
- Question
- What is the crude difference in remission proportions?
- Approach
- Apply the same remission definition and eligibility rules. Describe crude results before prespecified adjustment and sensitivity analyses.
- Result
- Remission occurs in 60 of 100 treated patients (60 in 100) and 40 of 100 historical patients (40 in 100). The difference is 20 per 100; the ratio is 1.5.
- What it means
- The crude contrast does not show that treatment caused the difference. Calendar changes and patient selection may explain it.
In practice
Recent studies PubMed search · not vetted
Watch out
Common mistakes
The one to remember
External controls do not inherit randomization from a treated trial. Differences in eligibility, time zero, calendar time, treatment and outcome measurement can mimic benefit. Adjustment cannot remove unmeasured differences.
- Choosing historical controls after seeing the treated outcomes.Instead: Fix data-source criteria and the analysis before reviewing the treatment contrast.
- Starting treated follow-up only after a patient survives to receive treatment.Instead: Align eligibility and treatment assignment with follow-up. Otherwise immortal time can favor the treated group.
- Assuming a propensity score makes the sources equivalent.Instead: Check balance, overlap, missingness and outcome measurement. Explain what unmeasured confounding could remain.
- Dropping people whose exposure was not measured.Instead: This can create selection bias. Under MAR, consider multiple imputation that retains eligible people, and assess departures from MAR.
Inside Mallard
How Mallard uses this
Plans do not link to this design yet. Mallard has no dedicated sizing form for external or historical control studies.
design.external_control, the id a plan cue records.How to report it
No reporting guideline dedicated to external or historical control studies was found in a PubMed search on 2026-11-06. The search used design names and reporting-guideline or checklist terms in titles. Use STROBE for the nonrandomized comparison. Report source selection, eligibility, time zero, calendar periods, outcome capture, overlap, adjustment, missingness and residual confounding. If historical data are borrowed into a new trial, also report the borrowing model and its sensitivity. A reporting guideline is not a design manual: use it to make the study transparent, not as a substitute for careful design. Check the target journal's own requirements as well, which this page cannot know.
Related Learn pages
Related designs and methods
Evidence
Sources & further reading
- Blinatumomab vs historical standard therapy of adult relapsed/refractory acute lymphoblastic leukemia (opens in a new tab) · Blood Cancer J · doi 10.1038/bcj.2016.84
- Hierarchical commensurate and power prior models for adaptive incorporation of historical information in clinical trials (opens in a new tab) · Biometrics · doi 10.1111/j.1541-0420.2011.01564.x
- Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies (opens in a new tab) · BMJ · doi 10.1136/bmj.39335.541782.AD
- Multiple imputation for missing data in epidemiological and clinical research: potential and pitfalls (opens in a new tab) · BMJ · doi 10.1136/bmj.b2393