Randomization, concealment and blinding
Why randomization works
It makes the groups comparable on everything, including things nobody measured or thought of, up to chance. That is the property no amount of adjustment can reproduce in observational data, and it is the reason a trial's confounding problem is solved rather than managed.
Concealment is what protects it
A perfect random sequence is worthless if the person deciding whether to enrol a patient can see what they would get. Selection then re-enters through eligibility judgements, and the empirical evidence is that inadequately concealed trials report larger effects. Central randomization, or sequentially numbered opaque sealed envelopes done properly, is the mechanism; unsealed lists and predictable schemes are not.
What concealment and blinding each protect
Choosing the scheme
- Simple randomization: unbiased and can leave the groups unequal in a small trial.
- Block randomization: keeps group sizes close, and with a fixed known block size the last allocation in each block becomes predictable, so vary the block size and keep it concealed.
- Stratified randomization: balances a few strong prognostic factors within centres or categories; stratification factors should then appear in the analysis.
- Minimisation: allocates to balance several factors, deterministic unless a random element is retained.
- Cluster randomization: allocates groups, which changes both the analysis and the sample size and needs consent handled at the right level.
Blinding, said precisely
Say who was blinded: participants, clinicians delivering the intervention, outcome assessors, data analysts. Outcome assessor blinding matters most for subjective endpoints and is often achievable when participant blinding is not. Where blinding is impossible, use objective or independently adjudicated outcomes and say what was done instead.
Baseline tables are not for testing
A significance test on a baseline characteristic in a randomized trial answers the wrong question. With random allocation an observed difference is compatible with chance, and a test cannot say whether randomization worked or whether the imbalance matters for the result. Report the distributions so the reader can judge whether an imbalance matters prognostically, and adjust for prespecified prognostic covariates whatever the table shows.
What to watch out for
Allocation concealment and blinding are different things and protect against different biases, and reporting one as the other is common. Concealment stops the person enrolling a participant from knowing what the next allocation will be, and it is possible in every trial. Blinding stops people knowing the allocation afterwards, and is sometimes impossible. A surgical trial that cannot be blinded can still be perfectly concealed, and the phrase double blind means so many different things that it should be replaced by saying who was blinded.
Clinical example
A surgical trial cannot blind the surgeon or the patient. It can still use central telephone randomization, blind the outcome assessor and the analyst, and adjudicate complications by a panel unaware of allocation, which is where most of the available protection lives.
- CONSORT 2010 statement: updated guidelines for reporting parallel group randomised trials (opens in a new tab) · BMJ · doi 10.1136/bmj.c332
- Assessing allocation concealment and blinding in randomised controlled trials: why bother? (opens in a new tab) · Evid Based Nurs · doi 10.1136/ebn.4.1.4