A randomized controlled trial (RCT) is a prospective experiment in which participants, or groups of participants, are assigned by chance to two or more interventions and their outcomes are compared. Random assignment distinguishes it from studies in which treatment is chosen by patients, clinicians, or investigators. In medicine, RCTs evaluate treatments and other health interventions by making comparisons less dependent on the characteristics that influence treatment choices. (cancer.gov)
Randomization and causal inference
The central purpose of randomization is to support causal inference. In nonrandomized comparisons, confounding can arise when factors affecting treatment selection also affect outcomes. For example, patients receiving one treatment may initially have more severe illness. Proper randomization prevents assignment from systematically depending on such prognostic factors, including factors that researchers have not measured. It does not guarantee identical groups in any particular trial: chance imbalances can remain. (cochrane.org)
Randomization concerns assignment, not recruitment. A randomly assigned study population is therefore not necessarily representative of the population that may eventually receive the intervention. Broad applicability depends additionally on eligibility criteria, participant characteristics, and clinical settings. (fda.gov)
The allocation sequence must also be protected. Allocation concealment prevents recruiters from knowing forthcoming assignments before enrollment is confirmed. Without this safeguard, enrollment decisions can undermine an otherwise random sequence. Concealment differs from blinding, which operates after assignment. (cochrane.org)
Controls and blinding
The control group supplies the comparison needed to interpret outcomes. Depending on the question, it may receive an established treatment, usual care, no intervention, or a placebo. A placebo resembles the intervention but lacks its specific active component; placebo-controlled comparisons help distinguish intervention effects from responses associated with receiving treatment. Placebos are not required for every RCT. (nccih.nih.gov)
Blinding means withholding knowledge of assignment from specified people, such as participants, care providers, or outcome assessors. It reduces opportunities for expectations to influence care or measurement. Reports are more informative when they identify who was blinded rather than merely use labels such as “double-blind.” Blinding may be impractical in comparisons involving visibly different procedures, including some surgical interventions, even though allocation concealment remains feasible. (cochrane.org)
Trial designs
RCTs use several forms of experimental design:
- Parallel-group trials: participants remain in their assigned intervention groups during the comparison.
- Crossover trials: participants receive interventions in a randomized sequence and serve as their own controls. Analysis must consider within-person correlation, period effects, and effects carried over from earlier treatment.
- Cluster-randomized trials: schools, practices, communities, or other groups are randomized. Outcomes within a cluster may be correlated, requiring corresponding design and analytical adjustments.
- Factorial trials: participants are assigned to combinations of interventions, allowing more than one intervention question to be investigated and interactions to be examined. (cochrane.org)
Trials also differ in their intended comparison. Superiority trials test whether an intervention performs better than its comparator. Noninferiority trials assess whether any disadvantage remains within a prespecified acceptable margin. Equivalence trials assess whether differences lie within prespecified limits in both directions; simply failing to detect a difference does not establish equivalence. (fda.gov)
Outcomes and statistical analysis
A trial protocol specifies the population, interventions, follow-up, outcomes, and planned analysis. The primary endpoint provides the principal basis for answering the research question. It may measure survival, symptoms, function, or another prespecified outcome; secondary endpoints address additional questions. Sample-size planning uses assumptions about a meaningful effect, outcome variability or event frequency, and statistical power. (fda.gov)
Results describe effect magnitude and uncertainty, commonly through estimates and confidence intervals. Hypothesis tests and p-values address statistical compatibility with a specified hypothesis, not clinical importance by themselves. Multiple outcomes and subgroup comparisons require attention because repeated testing can increase false-positive findings. (fda.gov)
An intention-to-treat analysis compares participants according to their original assignment, regardless of subsequent adherence or treatment switching. Its target is generally the effect of assignment. Analyses targeting the effect of actually following treatment answer a different question and require additional assumptions. Missing outcomes still pose difficulties: invoking intention-to-treat does not automatically resolve them. The intended treatment effect and handling of post-assignment events must be clearly specified. (fda.gov)
Ethics and transparent reporting
RCTs involving people are subject to research ethics, including independent review, proportionate risk assessment, and informed consent, with specific provisions for participants unable to consent. The Declaration of Helsinki states that new interventions generally should be compared with the best proven intervention. Placebo or no-intervention controls are permissible under defined conditions, including when no proven intervention exists, or when compelling methodological reasons apply without exposing participants to additional serious or irreversible harm from withholding proven care. (wma.net)
The declaration also calls for public registration before recruitment begins and disclosure of results, including negative and inconclusive findings. Reporting standards complement these ethical obligations. The 2025 CONSORT statement provides a 30-item checklist and participant-flow diagram for trial results; SPIRIT 2025 provides recommendations for protocols. These standards improve transparency but do not themselves guarantee a well-conducted trial. (wma.net)
Interpretation and limitations
Randomization does not eliminate every source of bias. Departures from assigned interventions, missing outcomes, biased measurement, and selective reporting can compromise results. Cochrane’s risk-of-bias framework therefore evaluates these problems separately rather than treating the randomized label as sufficient evidence of reliability. (cochrane.org)
External validity concerns applicability beyond the studied participants and conditions. Restrictive eligibility criteria or unusually optimized care can limit this applicability. Pragmatic trials investigate interventions under conditions closer to routine practice, whereas efficacy trials may emphasize carefully controlled delivery. The distinction concerns study purpose and implementation, not whether random assignment is used. (nccih.nih.gov)