Pilot study vs feasibility study is not an either-or choice: in the framework behind CONSORT's pilot extension, every pilot is a feasibility study. A feasibility study asks whether a future trial can be done, should be done and, if so, how. A pilot runs that trial, or part of it, on a smaller scale. Neither is built to test efficacy.
The label tells reviewers what to check. A pilot that leads with a p-value for the clinical outcome claims more than its design can carry. Our guide to sample size for small clinical studies already gives a short wording template for that case. This post covers what to measure, how to decide whether to proceed, and how to report it.
Pilot study vs feasibility study: where the line falls
These definitions come from the 2016 conceptual framework for pilot and feasibility studies that the CONSORT pilot and feasibility extension uses. Earlier definitions, such as those originally used by the UK National Institute for Health Research, treated the two as mutually exclusive, so name the framework you follow.
- Feasibility study — asks whether a future study can be done, should be done and, if so, how. It need not run any part of that study.
- Pilot study — a feasibility study that also runs the future study, or part of it, on a smaller scale. It can be randomised or not.
Counting eligible patients in clinic records, or asking clinicians whether they would randomise, is feasibility work but not a pilot, because nothing from the future trial is run. Recruiting, randomising and following participants through the planned procedures is a pilot. Choose the label by what the study did, not by which sounds more modest.
Feasibility outcomes measure the process, not the treatment effect
A feasibility objective asks whether the definitive trial can run as planned. Give each one a measure, a denominator and a time frame before the first participant consents. Objectives can cover:
- Recruitment — eligible patients per site per month, and the share of those approached who consent.
- Randomisation and blinding — whether allocation by chance is accepted, and whether blinding holds.
- Retention — the share with primary outcome data at the planned time point.
- Adherence and delivery — whether the intervention is received and delivered as the protocol describes.
- Outcome measurement — whether instruments are completed, and how variable the outcome is.
- Acceptability — what participants and staff think of the procedures, which may be gathered by interview.
Report each as an estimate, not a yes or no. Item 17a of the CONSORT pilot extension asks, for each objective, for results with an expression of uncertainty, such as a 95% confidence interval (CI), by randomised group where relevant.
Why a pilot should not test whether the treatment works
The CONSORT pilot and feasibility extension does not recommend making a test of effectiveness an objective of a pilot trial. The aim is to judge feasibility, and a pilot is usually too small to test effectiveness. Any hypothesis test should be adequately powered, and an underpowered one should be treated as secondary, with a caveat in the Discussion.
The effect estimate is the deeper problem. It comes from a small sample, so its confidence interval is wide. A chance high estimate gives a definitive trial too small to detect a smaller but worthwhile effect. A chance low one can make a worthwhile trial look pointless. Size the definitive trial on the smallest difference that would matter clinically; the pilot can supply an estimate of variability or of the control-group event rate.
The opposite error is reporting no numbers at all. Give every feasibility estimate with its interval, and present any clinical outcome by group as information, not as a test. The same boundary stops an uncontrolled series from claiming effectiveness, as our post on choosing between a case report, case series and cohort study explains. "Feasible" is a word a pilot can support; the table of Discussion wording that fits each study design draws the same line for a case series.
Set stop, amend or go progression criteria before the pilot starts
Progression criteria are prespecified thresholds for deciding whether, or how, to proceed to the definitive trial. The CONSORT pilot and feasibility extension asks for them in item 6c, and for the implications for progression, including any proposed amendments, in item 22a. They concern feasibility, not whether the treatment worked.
The extension's authors note that in many pilot studies these criteria may be best viewed as guidelines rather than strict thresholds, and that a traffic-light system is increasingly used. A 2023 set of recommendations on progression criteria adds practical advice. Map each criterion to a feasibility objective, justify it, write it as a guideline rather than a rule, avoid judging criteria in isolation, and report them in a table. Those recommendations are a commentary, not part of the CONSORT checklist. Fix the criteria in the protocol; our guide to writing a research protocol for a clinical study lists the sections a protocol should cover.
A progression-criteria table you can copy
The study below is invented to illustrate the structure; its numbers are not real data. A two-arm randomised pilot compares [intervention] with usual care at two sites over six months, randomising 36 participants, 18 per arm. Intervals are 95% CIs: Wilson score intervals for proportions, an exact Poisson interval for the rate.
| Feasibility objective | Measure | Green: go | Amber: amend | Red: stop | Observed (95% CI) |
|---|---|---|---|---|---|
| Recruitment | Participants randomised per site per month | 4 or more | 2 to 3.9 | Below 2 | 36 in 12 site-months: 3.0 (2.1 to 4.2) — amber |
| Consent | Consented, of eligible patients approached | 50% or more | 30% to 49% | Below 30% | 36 of 80: 45.0% (34.6 to 55.9) — amber |
| Retention | Primary outcome collected at [time point], of those randomised | 80% or more | 65% to 79% | Below 65% | 31 of 36: 86.1% (71.3 to 93.9) — green |
| Adherence | Intervention arm attending at least three-quarters of planned sessions | 75% or more | 50% to 74% | Below 50% | 14 of 18: 77.8% (54.8 to 91.0) — green |
| Outcome data | Primary outcome measure fully completed, of those followed up | 90% or more | 75% to 89% | Below 75% | 29 of 31: 93.5% (79.3 to 98.2) — green |
Read the rows together. Recruitment and consent are amber while the other three are green, so the verdict is amend, not stop. Every one of the five intervals crosses two bands, which is expected with 36 participants and is why the criteria work better as guidelines than as rules. State the decision in a paragraph like this:
Progression. Prespecified progression criteria (protocol version [number], [date]) covered [objectives]. [n] criteria met the go threshold and [n] fell in the amend range: [criterion, observed value and 95% CI]. No criterion met the stop threshold. We propose to proceed to a definitive trial with these amendments: [amendment, and the criterion it addresses]. Clinical outcomes are presented descriptively by group, with 95% CIs, to inform that trial.
Report a randomised pilot with the CONSORT pilot and feasibility extension
Randomised pilot and feasibility trials have their own reporting guideline, the CONSORT extension to randomised pilot and feasibility trials, published in 2016 and listed by the EQUATOR Network. It has a 26-item checklist, a separate abstract checklist and a flow-chart template. It kept some CONSORT items, adapted most, removed some and added new ones, including how participants were identified and consent obtained. The table shows five of the adapted or new items that change a standard trial report.
| Item | What it asks for | What to write |
|---|---|---|
| 2b | Specific objectives or research questions for the pilot trial | Feasibility objectives, not an effectiveness hypothesis |
| 6c | If applicable, prespecified criteria for judging whether, or how, to proceed with a future definitive trial | The progression-criteria table and its protocol version |
| 7a | Rationale for numbers in the pilot trial | Why the number suits the feasibility objectives, not a power calculation |
| 17a | Results for each objective, with expressions of uncertainty, by randomised group where relevant | Each feasibility estimate with its 95% CI |
| 22a | Implications for progression, including any proposed amendments | The go, amend or stop decision and the changes it requires |
Which version to cite, and which studies it covers
The extension was built on the 2010 version of CONSORT and, as of September 2026, has not been reissued for CONSORT 2025. Its developers say it can be used together with CONSORT 2025, and the CONSORT 2025 authors recommend using existing extensions until they are aligned. No 2025 version of the pilot extension exists yet, so do not cite one.
The extension covers any randomised study that runs a future definitive trial, or part of it, on a smaller scale, whatever the design and whatever the authors call it. It does not directly apply to internal pilots built into the design of a main trial, to non-randomised pilot and feasibility studies, or to phase II studies. Its developers say many of its principles may still apply to them. For non-randomised work, a 2019 editorial on reporting non-randomised pilot and feasibility studies recommends the extension as the main reference, adapting or dropping items that do not fit. It is guidance, not a separate formal checklist.
A small pilot still faces registration and ethics checks
Check your pilot against the International Committee of Medical Journal Editors (ICMJE) definition of a clinical trial: any study that prospectively assigns people to an intervention to evaluate a health outcome. The definition turns on what the study does, not its size or label. If your pilot meets it, registration is due at or before the time of first patient consent for enrolment; our post on clinical trial registration and its timing lists acceptable registries. The CONSORT pilot extension also asks you to report ethical approval or approval by a research review committee, with a reference number.
If you find an error in this guide or anything else we have published, report the problem to us.