Prospective Randomized Clinical Trial

A prospective RCT randomizes participants at enrollment and then collects outcome data going forward in time — this is the standard design. A so-called retrospective RCT is a contradiction in terms; retrospective studies use existing data collected in the past and cannot randomize participants. Items labeled 'retrospective RCT' in some databases typically refer to secondary analyses of prospectively collected RCT data, not a distinct design.

Allocation concealment prevents anyone involved in enrolling or treating participants from knowing the next assignment before a participant is entered into the trial. Blinding prevents participants and/or investigators from knowing the assignment after enrollment. Both are necessary: without concealment, selection bias can enter during recruitment; without blinding, performance and detection bias can enter during treatment and outcome assessment.

Intention-to-treat (ITT) analysis includes all randomized participants in the group to which they were assigned, regardless of adherence, and is the primary analysis protecting the causal interpretation from confounding by adherence. Per-protocol analysis restricts to participants who adhered to the protocol and is subject to selection bias but may better approximate the biological effect of the treatment when adherence is high. Both should typically be reported.

A sample size calculation requires four inputs: the expected outcome rate or mean in the control group, the minimum clinically important difference you want to detect (effect size), the desired statistical power (commonly 80% or 90%), and the significance level (commonly alpha = 0.05 two-sided). Standard formulas or software (e.g., G*Power, R's pwr package) produce the required n per arm. Allow for anticipated dropout by inflating the calculated n.

Prospective registration of the protocol in a public registry (ClinicalTrials.gov, WHO ICTRP, or equivalent) before first participant enrollment is required by the Declaration of Helsinki, by most major journals as a condition of publication, and by most regulatory agencies. Registration timestamps the primary outcome and analysis plan, preventing outcome switching and publication bias.