Trial design and analysis
The trial of antibiotic prophylaxis and pelvic infection discussed at the outset of this section featured two important aspects of trial design, randomization and blinding (also known as masking).13,16 Patients were randomized either to the antibiotic treatments or placebos, which helped ensure that groups being compared in the study were comparable for measured characteristics and hopefully for unmeasured characteristics as well. The trial also involved double-blinding, which meant that both trial personnel and participants would not know whether the participant received the test treatments or just the placebos.
Double-blinding helps ensure that neither participants nor trial personnel change their behaviour according to whether a participant has been assigned the active treatment, which might influence the outcomes studied.10 If a drug has a known effect such as changing the colour of urine or affecting LDL levels, then a trial may become unblinded. If someone assessing the outcome of a trial becomes aware of a participant’s treatment assignment, this might bias assessment of outcomes that involve subjective judgement.16
As participants lost to follow-up can decrease statistical power and introduce bias, trial design may involve planning to recruit a larger sample than necessary and making efforts to minimize losses to follow-up.16 If possible, participants who withdraw from an intervention should continue to be followed to ascertain the treatment outcomes.16
An intention-to-treat analysis includes all patients according to the treatment arm they were originally randomized to, regardless of whether the patient received or completed the treatment regimen.2,14 This contrasts with a per-protocol analysis (also known as an on-treatment or as-treated analysis), which has been defined as including only participants “who completed the study without any major protocol violations.”17 Intention-to-treat analyses should be the first analysis performed, because deviation from an intention-to-treat approach may undermine the comparability of groups studied.16
When researchers are interested in more than one intervention, a factorial design may be used.16 When a 2×2 factorial design is used, this involved randomizing participants to A and not-A, and then randomizing patients to B and non-B, where not-A and not-B may be no-intervention, a comparison intervention, or a placebo. The effect of A may be analyzed by comparing all patients in the A group to the not-A group, while the effect of B can be analyzed by comparing all patients in the B group to the not-B group. Factorial designs with more than two interventions are also possible.