Prevention & care
Randomized trials: assignment, analysis targets and application to care
On this page
A randomized trial assigns participants or other units to comparison strategies through a random process. Its value depends on preserving that process, measuring relevant outcomes and analyzing the intended comparison. CONSORT 2025 provides current reporting guidance; completing its checklist does not itself establish an unbiased result.[1]
Separate allocation, masking, purpose and analysis
| Term | Meaning | Distinction |
|---|---|---|
| Random sequence | An unpredictable process assigns comparison strategies. | Alternation or a predictable calendar rule is different.[3] |
| Allocation concealment | Upcoming assignments are hidden during enrollment. | Protects who enters which group before allocation.[3] |
| Blinding | Assigned treatment identity is masked after allocation. | Different from concealing the next assignment.[3] |
| Random sampling | Randomly selects people from a population. | Random treatment allocation does not imply representative sampling.[8] |
| Trial structure | Parallel groups, randomized clusters, or successive crossover periods. | The unit and timing of comparison affect interpretation.[1] |
| Explanatory and pragmatic | Different emphasis on ideal conditions or usual-care decisions. | Efficacy and effectiveness form a design continuum.[6] |
| Intention-to-treat and per-protocol | Targets assignment versus following the specified strategy. | Analyzing only adherent people can introduce bias.[7] |
| Estimand | The treatment effect the study intends to estimate. | Specify how switching, rescue treatment or death affects the question.[4] |
A random list needs protection during recruitment
Schulz and Grimes explain how a genuinely random sequence can be undermined if recruiters discover the next assignment and alter enrollment. Concealment protects the allocation process; blinding addresses knowledge of treatment afterward. Participants cannot always be blinded, for example to a visible care programme. Reports should state who was masked and how, rather than relying on the ambiguous label double-blind.[3]
Treatment discontinuation and missing outcomes are different
ICH E9(R1) defines a framework connecting the clinical question, design and analysis. Stopping assigned treatment, taking rescue medicine or switching treatment can change the effect of interest even when an outcome is observed. Withdrawal from study follow-up creates a missing-data problem. A treatment-policy question can use outcomes regardless of switching; a hypothetical question asks about a specified alternative scenario. The choice should be explicit and clinically meaningful.[4]
A cluster trial randomizes groups of people
In a cluster-randomized trial, groups such as clinics, schools or communities are assigned to strategies. This can fit an intervention delivered to a whole service or reduce spillover between comparison groups. People in the same group share circumstances, so their outcomes are not fully independent. Reports need the number of clusters and people, and an analysis accounting for that structure.[10]
A crossover trial randomizes the order of treatments
In a planned crossover trial, each participant receives treatments in a randomized sequence and acts as their own comparison. A period is the time receiving one treatment; carryover means an earlier treatment's effects persist into a later period. A period effect is change with time unrelated to treatment. The design generally suits stable conditions with reversible, short-lived effects; it does not suit every illness.[11]
Evidence & uncertainty
In COLOFOL, more frequent follow-up after colorectal-cancer surgery did not significantly reduce five-year mortality compared with less frequent testing. Both groups received follow-up. This particular null comparison does not establish that every surveillance approach is ineffective or equivalent.[9]
Selected-source overview · Independent review pendingPublished questions
Are allocation concealment and blinding the same?
Selected-source overview · Independent review pending. Read the evidence context.
No. Concealment prevents foreknowledge of upcoming assignments during recruitment. Blinding masks treatment identity after allocation. An unblinded trial can still conceal allocation; the report should describe both processes clearly and identify who was masked.[3]
Evidence behind this answer
- Allocation concealment in randomised trials: defending against deciphering
original randomization methods paper · 2002
Selected original published five-page PDF passages inspected through a hosted reprint; original indexed PubMed bibliographic record inspected · Sequence versus concealment versus blinding and recruitment manipulation
This answer summarizes selected sources. An independent appraisal of the complete evidence has not been completed.
Does a CONSORT-compliant report guarantee a high-quality trial?
Selected-source overview · Independent review pending. Read the evidence context.
No. CONSORT 2025 supports clear reporting, including participant flow, analysis denominators, delivery and harms. Those details make strengths and deficiencies assessable. The statement explicitly does not function as a quality-assessment instrument.[1]
Evidence behind this answer
- CONSORT 2025 statement: updated guideline for reporting randomised trials
original current trial reporting guideline · 2025
Selected original publisher HTML substantive statement passages inspected; official current published-statements page inspected · Current 30-item statement, flow/denominators, design extensions and not a quality instrument
This answer summarizes selected sources. An independent appraisal of the complete evidence has not been completed.
Does prospective registration prevent every later change?
Selected-source overview · Independent review pending. Read the evidence context.
No. Protocols can change, and reports should explain important amendments and their timing. Registration and accessible plans make comparison possible; they do not ensure successful conduct, complete outcomes or a benefit.[2]
Evidence behind this answer
- SPIRIT 2025 statement: updated guideline for protocols of randomised trials
original current trial protocol reporting guideline · 2025
Selected original publisher HTML substantive statement passages inspected; official current published-statements page inspected · Protocol plans, outcomes/analysis/oversight, transparency and limitations
This answer summarizes selected sources. An independent appraisal of the complete evidence has not been completed.
Is switching treatment simply missing data?
Selected-source overview · Independent review pending. Read the evidence context.
No. An observed outcome after switching may answer one clinical question but not another. ICH E9(R1) distinguishes these intercurrent events from missing observations and asks researchers to define their intended effect explicitly.[4]
Evidence behind this answer
- ICH E9(R1): Addendum on estimands and sensitivity analysis in clinical trials
official final statistical guideline · 2019
Selected original final 22-page PDF passages inspected; current EMA guideline listing checked · Final 2019 framework, intercurrent events/missing data, effect attributes and sensitivity
This answer summarizes selected sources. An independent appraisal of the complete evidence has not been completed.
Is analyzing only adherent people automatically less biased?
Selected-source overview · Independent review pending. Read the evidence context.
No. Adherence can relate to prognosis and events after allocation. A per-protocol effect may be relevant, but valid estimation requires measured information, appropriate methods and assumptions; simply excluding nonadherent participants sacrifices the original randomized comparison.[7]
Evidence behind this answer
- Per-Protocol Analyses of Pragmatic Trials
original trial analysis methods paper · 2017
Original publisher indexed abstract and investigator-hosted published PDF indexed substantive passages inspected; direct author PDF retrieval failed; no full-text appraisal · Assignment/adherence targets, postrandomization confounding and necessary information/assumptions
This answer summarizes selected sources. An independent appraisal of the complete evidence has not been completed.
Can imputation reveal the actual missing outcomes?
Selected-source overview · Independent review pending. Read the evidence context.
No. It uses assumptions and available information. Prevention of missingness, transparent methods and sensitivity analyses remain important. A precise estimate under one assumption does not establish that the assumption matches the unseen data.[5]
Evidence behind this answer
- The Prevention and Treatment of Missing Data in Clinical Trials
original consensus statistical methods report · 2010
Original publication record and selected original book/chapter indexed recommendations inspected; dynamic publisher HTML and NCBI direct retrieval did not expose full book text; no full-book appraisal · Design prevention, missingness assumptions, uncertainty and sensitivity recommendations
This answer summarizes selected sources. An independent appraisal of the complete evidence has not been completed.
Does pragmatic mean every aspect resembles ordinary care?
Selected-source overview · Independent review pending. Read the evidence context.
No. PRECIS-2 considers separate design domains, each on a continuum. Eligibility, delivery, follow-up or analysis can differ in pragmatism. The relevant question is whether those choices fit the intended care decision and setting.[6]
Evidence behind this answer
- The PRECIS-2 tool: designing trials that are fit for purpose
original trial design tool · 2015
Original PubMed abstract and original publisher indexed substantive domain/design passages inspected; no independent full-text appraisal · Nine explanatory/pragmatic domains and alignment to intended decisions
This answer summarizes selected sources. An independent appraisal of the complete evidence has not been completed.
Did COLOFOL prove that no cancer follow-up is useful?
Selected-source overview · Independent review pending. Read the evidence context.
No. It compared two active testing schedules in selected postoperative patients. Its nonsignificant mortality difference and confidence interval do not establish exact equivalence, compare follow-up with none, or prescribe care for other cancers.[9]
Evidence behind this answer
- COLOFOL: more versus less frequent follow-up testing after stage II/III colorectal cancer
original randomized active comparator trial · 2018
Selected original publisher full HTML methods/results/flow counts and funding passages inspected; original PubMed abstract/indexed results inspected · 2,555 randomized/2,509 ITT; 5-year mortality 13.0 vs 14.1, CI −1.6 to 3.8; two active testing schedules
This answer summarizes selected sources. An independent appraisal of the complete evidence has not been completed.
Do thousands of patients always mean thousands of independent comparisons?
Selected-source overview · Independent review pending. Read the evidence context.
No. Patients grouped within randomized clinics or schools can have related outcomes. The cluster count and analysis matter alongside the total patient count.[10]
Evidence behind this answer
- CONSORT 2010 statement: extension to cluster randomised trials
Original reporting/methodological framework · 2012
Full publisher body retrieved after native403; selected substantive scope/design/recruitment/analysis passages read. · Scope; Box 2 design, recruitment and analysis; minimum reporting versus quality distinction
This answer summarizes selected sources. An independent appraisal of the complete evidence has not been completed.
Does crossover just mean someone changed treatment during a trial?
Selected-source overview · Independent review pending. Read the evidence context.
No. A randomized crossover design assigns a planned treatment sequence. Unplanned switching in a parallel-group trial is a different situation.[11]
Evidence behind this answer
- CONSORT 2010 statement: extension to randomised crossover trials
Original reporting/methodological framework · 2019
Full publisher body retrieved after native403 and PMC cookie screen; abstract and selected methodological features/design passages read. · Abstract; methodological features and glossary; design applicability; reporting/flow
This answer summarizes selected sources. An independent appraisal of the complete evidence has not been completed.
Compare plans with what was actually done
SPIRIT 2025 concerns trial protocols, while the final report describes conduct and results. A protocol should make the interventions, outcomes, analysis and oversight assessable before results are known. Registration and accessible analysis plans help readers recognize later changes. Legitimate amendments can occur, but their timing and reasons matter. A registered or detailed plan does not guarantee complete follow-up, successful implementation or favorable results.[2]
Follow the participants and the denominator
CONSORT 2025 requests participant flow, intervention delivery, harms and the numbers contributing to each outcome at each time point. Enrollment, allocation, receipt of treatment, follow-up and analysis are different counts. An overall abstract number may not be the denominator for every endpoint. Readers also need effect estimates and uncertainty for prespecified outcomes, with important changes and unplanned analyses identified. Reporting transparency enables appraisal; it is not a quality score.[1]
Removing nonadherent participants does not recreate randomization
Hernán and Robins distinguish effects of assignment from effects of following a strategy. Intention-to-treat estimates reflect the adherence patterns and care in that trial. A per-protocol question can be useful, but adherent and nonadherent people may differ in prognosis. Simply excluding deviations can create confounding or selection bias. Appropriate causal methods require measured adherence and relevant prognostic information, including changes after randomization, together with additional assumptions.[7]
Missing-data methods require assumptions
The National Research Council’s 2010 report prioritizes reducing missing outcomes through trial design and follow-up. Analyses should account for missingness and its uncertainty rather than treat an inserted value as an observed fact. Different reasons for missing data can support different assumptions. Sensitivity analyses examine whether plausible departures change the conclusion. They cannot prove which unseen outcomes actually occurred, and a sophisticated analysis cannot compensate reliably for avoidable loss of crucial information.[5]
Pragmatic design is a set of choices
PRECIS-2 evaluates nine domains, including eligibility, recruitment, setting, intervention delivery, adherence flexibility, follow-up, outcomes and analysis. A trial can be relatively pragmatic in one domain and explanatory in another. The purpose is to align design with the decision the results should inform. More intensive support may clarify efficacy under selected conditions but differ from usual care. Calling a trial real-world is insufficient without describing those choices and the actual comparator.[6]
A large trial retained uncertainty around a null comparison
COLOFOL randomized 2,555 people after curative surgery for stage II or III colorectal cancer; 2,509 entered its intention-to-treat analysis. The unblinded study compared five with two planned CT/CEA testing occasions in 24 centers across three countries. Five-year overall mortality was 13.0% versus 14.1%; the reported difference’s confidence interval ranged from −1.6 to 3.8 percentage points. Cancer-specific mortality also showed no significant reduction. The endpoint, exclusions, active comparator and interval all matter: nonsignificance alone is not proof of exact equivalence or a personal follow-up schedule.[9]
A valid comparison still needs a relevant setting
Rothwell describes how recruitment pathways, eligibility, care systems, clinician expertise, treatment timing, comparator care and follow-up affect applicability. Randomization balances assignment in the enrolled comparison; it does not automatically make participants a random sample of everyone receiving care. Even broad pragmatic recruitment can yield an average effect that differs from a particular setting’s effect. Applying results requires clinical context and the wider evidence, including outcomes and burdens important to the person.[8]
Enrollment and later periods can change the comparison
In cluster trials, recruiting individuals after the service's allocation is known can introduce selection bias. A large patient count does not erase that issue or the correlation within groups.[10]
A crossover report should explain the treatment order, separation between periods and losses to follow-up. Repeated observations from one person remain related. Changing treatment unexpectedly in an ordinary parallel trial is a different event from a prespecified randomized crossover design.[11]
References
- CONSORT 2025 statement: updated guideline for reporting randomised trials
Original current trial reporting guideline
Hopewell and colleagues; PLOS Medicine/SPIRIT-CONSORT group · Content/publication date: 2025
Accessed: 2026-10-14 UTC · Selected original publisher HTML substantive statement passages inspected; official current published-statements page inspected
Cited sections: Current 30-item statement, flow/denominators, design extensions and not a quality instrumentIdentical simultaneous BMJ e081123/JAMA/Lancet/Nature Medicine/PLOS versions are ONE publication family and allowance. Official group confirms identical contents. Current 2025 core replaces 2010; appropriate specialized extensions have additional scope, not independently appraised here.
- SPIRIT 2025 statement: updated guideline for protocols of randomised trials
Original current trial protocol reporting guideline
Chan and colleagues; PLOS Medicine/SPIRIT-CONSORT group · Content/publication date: 2025
Accessed: 2026-10-14 UTC · Selected original publisher HTML substantive statement passages inspected; official current published-statements page inspected
Cited sections: Protocol plans, outcomes/analysis/oversight, transparency and limitationsIdentical BMJ e081477/JAMA/Lancet/Nature Medicine/PLOS versions count as one work. Protocol statement has genuinely different scope from CONSORT results statement, within joint reporting-guideline family; neither proves bias elimination.
- Allocation concealment in randomised trials: defending against deciphering
Original randomization methods paper
Schulz and Grimes; The Lancet · Content/publication date: 2002
Accessed: 2026-10-14 UTC · Selected original published five-page PDF passages inspected through a hosted reprint; original indexed PubMed bibliographic record inspected
Cited sections: Sequence versus concealment versus blinding and recruitment manipulationDOI 10.1016/S0140-6736(02)07750-4. Historical methodological explanation; old trial-reporting percentages not used as current 2026 rates. USAID/hosted publisher reprints same underlying work; no extra allowance.
- ICH E9(R1): Addendum on estimands and sensitivity analysis in clinical trials
Official final statistical guideline
International Council for Harmonisation · Content/publication date: 2019
Accessed: 2026-10-14 UTC · Selected original final 22-page PDF passages inspected; current EMA guideline listing checked
Cited sections: Final 2019 framework, intercurrent events/missing data, effect attributes and sensitivityFinal adopted November 20, 2019; document-history draft/date November 17 and filename December 3 are not adoption date. Pharmaceutical regulatory framework, not an asserted universal legal requirement or one mandatory estimand for every question.
- The Prevention and Treatment of Missing Data in Clinical Trials
Original consensus statistical methods report
National Research Council; National Academies · Content/publication date: 2010
Accessed: 2026-10-14 UTC · Original publication record and selected original book/chapter indexed recommendations inspected; dynamic publisher HTML and NCBI direct retrieval did not expose full book text; no full-book appraisal
Cited sections: Design prevention, missingness assumptions, uncertainty and sensitivity recommendationsDOI 10.17226/12955; all NAP chapters/NCBI book/PDF represent ONE book budget. Historical foundational consensus used with final ICH 2019 estimands framework; not falsely described as a new 2026 guideline.
- The PRECIS-2 tool: designing trials that are fit for purpose
Original trial design tool
Loudon and colleagues; The BMJ · Content/publication date: 2015
Accessed: 2026-10-14 UTC · Original PubMed abstract and original publisher indexed substantive domain/design passages inspected; no independent full-text appraisal
Cited sections: Nine explanatory/pragmatic domains and alignment to intended decisionsDOI 10.1136/bmj.h2147; PMID 25956159. Design tool, not a risk-of-bias grade or clinical intervention result. Later reliability study not claimed separately appraised.
- Per-Protocol Analyses of Pragmatic Trials
Original trial analysis methods paper
Hernán and Robins; New England Journal of Medicine · Content/publication date: 2017
Accessed: 2026-10-14 UTC · Original publisher indexed abstract and investigator-hosted published PDF indexed substantive passages inspected; direct author PDF retrieval failed; no full-text appraisal
Cited sections: Assignment/adherence targets, postrandomization confounding and necessary information/assumptionsDOI 10.1056/NEJMsm1605385; October 5, 2017 issue and article-update note distinguished from check date. Same authors’ 2016 target-trial article and 2020 evolving book are distinct works, not independent effectiveness replications.
- Factors That Can Affect the External Validity of Randomised Controlled Trials
Original external validity methods essay
Peter Rothwell; PLOS Clinical Trials · Content/publication date: 2006
Accessed: 2026-10-14 UTC · Selected original publisher HTML substantive methods passages inspected
Cited sections: Selection pathways, context, clinicians, treatment timing, comparators and pragmatic caveatsStable methods discussion; dated historical condition-specific percentages not used as current prevalence. DOI 10.1371/journal.pctr.0010009; PMID 16871331; PMC 1488890 same paper. 2005 Lancet external-validity paper not separately cited or budgeted.
- COLOFOL: more versus less frequent follow-up testing after stage II/III colorectal cancer
Original randomized active comparator trial
Wille-Jørgensen and colleagues; JAMA · Content/publication date: 2018
Accessed: 2026-10-14 UTC · Selected original publisher full HTML methods/results/flow counts and funding passages inspected; original PubMed abstract/indexed results inspected
Cited sections: 2,555 randomized/2,509 ITT; 5-year mortality 13.0 vs 14.1, CI −1.6 to 3.8; two active testing schedulesDOI 10.1001/jama.2018.5623; PMC 6583244 same report. 2509 ITT differs from 2555 randomized; cancer-specific denominators 2,498. Nordic/cancer/foundation grants disclosed. Original five-year report used as methods example; later 2024 ten-year follow-up not independent trial or current personal schedule recommendation.
- CONSORT 2010 statement: extension to cluster randomised trials
Original reporting/methodological framework
Campbell, Piaggio, Elbourne and Altman · Content/publication date: 2012
Accessed: 2026-10-20 UTC · Full publisher body retrieved after native403; selected substantive scope/design/recruitment/analysis passages read.
Cited sections: Scope; Box 2 design, recruitment and analysis; minimum reporting versus quality distinctionSpecialized 2012 framework alongside current CONSORT 2025; reporting guidance does not certify study quality.
- CONSORT 2010 statement: extension to randomised crossover trials
Original reporting/methodological framework
Dwan, Li, Altman and Elbourne · Content/publication date: 2019
Accessed: 2026-10-20 UTC · Full publisher body retrieved after native403 and PMC cookie screen; abstract and selected methodological features/design passages read.
Cited sections: Abstract; methodological features and glossary; design applicability; reporting/flowSpecialized 2019 framework; no personal washout interval or analysis formula supplied.
Scope & review
- Population
- Readers interpreting randomized comparative health research
- Exposure or intervention
- Random assignment, concealment/masking, designs, effect targets, plans, missingness and applicability
- Comparison
- Randomized alternative strategies including active comparisons; no assumption control means placebo or no care
- Outcomes
- Assignment/per-protocol estimands, clinical outcomes, harms and confidence intervals; one selected postoperative trial null
- Timeframe
- Current 2025 protocol/reporting guidance, final 2019 estimands; historical methods and original 2018 five-year trial report
- Jurisdiction
- International educational methods; ICH regional regulatory context and three-country postoperative example; no personal treatment or testing timetable
Selected-source educational explanation. Official guidance, original study findings and editorial interpretation are distinguished. No exhaustive search, independent systematic synthesis or qualified medical review has been completed.
Sources checked on their individual access dates. Selected sources were checked; ongoing surveillance and qualified review have separate statuses on the coverage page.
Revision history
- 2026-10-14 UTC: Added a bounded study and causal methods foundation guide; qualified clinical and research-methods review remains pending.
- 2026-10-20 UTC: Added supported foundational, practical and evidence explanations after reading full target bodies and relevant current/pending adjacent coverage. Original sections, questions and sources preserved as exact prefixes; qualified review pending.
What this article still leaves open
- Which clinical effect and analysis population were prespecified?
- What follow-up or adherence assumptions affect interpretation?
- How do the trial’s care setting, comparator and outcomes relate to the decision?
Suggest a correction
Submit a content correction to the editorial queue, or download a local note. Leave out personal medical details. Submitted text is stored privately with a browser receipt session and is visible to this browser and the editor. Clearing cookies removes access to your submission history. This is not an account or a medical service.
Your submitted corrections for this article
Explore the topic map
Browse the health domains connected to this article. The map also identifies topics whose articles remain unwritten.