Clinical trials
Start with the study objective, then choose the assignment unit and design. The order moves from feasibility and conventional comparisons toward specialized designs with more demanding conduct and analysis. Complexity depends on the question and setting; it is not measured by the number of flows.
Which study should I use?
| Study type | When to use it | How it differs |
|---|---|---|
| Pilot or feasibility trial | Resolve uncertainties about recruitment, delivery, retention, measurement or trial procedures before a larger evaluation. Prespecify criteria for proceeding, amending or stopping. | Feasibility is the primary objective. A small efficacy trial is not automatically a pilot, and an imprecise treatment effect does not determine whether a definitive trial should proceed. |
| Single-arm interventional study | Estimate response, safety or another prespecified outcome after a common intervention, often in early development or a setting where the objective justifies a single arm. | Without a concurrent randomized comparison, outcome changes can reflect prognosis, selection or background care. A historical benchmark adds assumptions and does not create randomization. |
| Parallel-group randomized trial | Compare interventions by randomly assigning individual participants to parallel groups and following them over a common period. Define the intended treatment effect before analysis. | Each participant follows one assigned strategy, unlike crossover. Allocation is at participant level, unlike cluster trials; randomization differs from the stated nonrandom assignment used in a nonrandomized study. |
| Nonrandomized comparative interventional study | Compare prospectively assigned interventions using a stated nonrandom rule, with an explicit plan for baseline differences, time trends and selection. | Investigators assign the intervention, distinguishing this from routine-care observational comparisons. Assignment is not randomized, so measured adjustment alone cannot establish exchangeability. |
| Randomized crossover trial | Compare reversible intervention effects within participants when the condition is sufficiently stable and period or carryover effects can be addressed. | Participants contribute observations under more than one treatment, unlike parallel groups. Randomization assigns sequences. Persistent effects, cure or rapidly changing disease can make crossover unsuitable. |
| Factorial randomized trial | Randomize combinations of intervention factors when the joint design can answer the planned factor-specific or interaction questions efficiently. | A factorial trial crosses factors, rather than merely adding unrelated parallel arms. Effects depend on the contrast and scale; absence of a significant interaction test does not establish no interaction. |
| Cluster-randomized parallel trial | Randomize hospitals, practices, schools or other clusters when the intervention is delivered collectively or individual assignment would cause substantial contamination. | Randomization occurs at cluster level while outcomes may be measured in individuals. Observations within a cluster are dependent; many individuals cannot compensate automatically for too few clusters. |
| Dose-escalation and dose-finding study | Study early dose or regimen selection using a reviewed rule-based, model-assisted or model-based design with safety constraints and defined decision authority. | The objective is dose selection and characterization, not necessarily a definitive treatment-effect claim. A maximum tolerated dose and an optimal biologic or recommended dose need not coincide. |
| Stepped-wedge cluster trial | Randomize clusters to intervention transition sequences, with repeated observations over calendar periods, when a staged rollout is scientifically and operationally justified. | Clusters transition from control to intervention at different times. Unlike a parallel cluster trial, intervention exposure is tied to calendar time, requiring explicit separation of time trends and intervention effects. |
| Adaptive randomized trial | Use planned adaptations such as sample-size reassessment, arm dropping or allocation changes when their operating characteristics and operational safeguards are established in advance. | An adaptive design prespecifies decisions based on accumulating information. It is not an unrestricted protocol amendment; it can overlap with a platform or dose-finding design. |
| Platform trial | Evaluate multiple interventions with shared infrastructure and prespecified rules for adding, retaining or closing arms. Define each comparison and its eligible population separately. | A platform continues across changing arms and calendar epochs. A fixed multi-arm trial need not be a platform, and a platform need not use response-adaptive randomization. |
Design features can overlap
A pilot can be randomized or nonrandomized. A factorial design can randomize clusters. A platform may use adaptive decisions, but not all platforms use response-adaptive allocation. These starters emphasize different design features; combine the relevant specifications when a protocol uses more than one.
Terms used in these guides
- Randomization, concealment and blinding
- Randomization generates assignment by chance. Concealment prevents knowledge of upcoming assignments. Blinding limits knowledge of assigned intervention during conduct or assessment. They are distinct procedures.
- Estimand
- The treatment-effect question defined by population, treatment conditions, outcome variable, strategies for intercurrent events and population-level summary.
- Intercurrent event
- An event after treatment initiation that changes the interpretation or existence of the outcome, such as treatment discontinuation or rescue treatment. It differs from simply failing to measure an outcome.
- Analysis population
- A prespecified set of participants and assignment or receipt rules used for an analysis. Assignment-based, safety and per-protocol populations can overlap and should not be counted as disjoint groups.
- Multiplicity
- The presence of multiple claims, outcomes, comparisons or interim decisions requiring a planned approach to false-positive conclusions or other decision-error criteria.
What registration means
A clinical trial registry publicly records the trial plan and identifying information. Examples include ClinicalTrials.gov and WHO primary registries such as ISRCTN. Register before first participant enrollment, record the registry name and identifier, and maintain amendments. WHO ICTRP searches registry records; it is not itself a registry. A disease registry containing patient data serves a different purpose. Nonrandom assignment does not by itself remove an interventional study from clinical-trial registration considerations.
Using reporting guidance
Use CONSORT 2025 for randomized-trial reports and SPIRIT 2025 for randomized-trial protocols, with the relevant design extensions linked in each guide. Extensions retain their actual publication versions; several predate the 2025 core statements. Single-arm and nonrandomized studies need guidance suited to their design and intervention. Reporting checklists do not replace design justification or validate statistical calculations.