Randomized controlled trial

"Randomized" means each participant has a known chance of getting each option, but nobody can predict which one they will get [1]. Done properly, randomization has two parts: an unpredictable allocation sequence, and keeping that sequence hidden from the people enrolling participants, called allocation concealment [1]. When it works, it balances both known and unknown factors that could affect the result, so that on average each group has the same outlook before the intervention starts [1][3].

"Controlled" refers to the comparison group. A trial can compare the intervention with a placebo or with an active comparator, meaning another real option [2]. Random assignment also lets researchers use probability to say how likely it is that a difference between groups reflects chance alone [1].

The word "random" on its own tells a reader little. The CONSORT reporting guideline asks authors to say how the sequence was made, such as with a random-number table or a computer random number generator [1]. When a trial report describes each group at the start, clear baseline differences between the groups can point to a problem with the randomization process [3].

The CONSORT authors note that RCTs are considered the gold standard for evaluating interventions because they can minimize or avoid bias [1]. They also note that poor design and poor reporting are linked to biased estimates of effects, so the label alone does not settle how far a trial can be trusted [1]. The Cochrane risk-of-bias tool has a separate domain for bias arising from the randomization process [3].

How randomized trials count toward a grade on this site is set out in how we grade evidence.

• Placebo: the dummy comparison many RCTs use.
• Double-blind: who in a trial is kept unaware of group assignment.
• Risk of bias: how reviewers judge whether a trial's result can be trusted.
• Pre-registration: recording a trial's plan before its first participant is enrolled.