Skip to content
Scintria
ToolsGrant writerFunding
HomeLiteratureClinical trialsEvidence LabProtocol packData analysisLearnProjectsLibraryReferencesApplications
Start free
ScintriaStart free

Learn / Clinical Research Management / Data management and trial statistics

Randomisation, blinding and the statistical analysis plan

13 minute read · 6-question quiz

Download this lesson as a colour Word or PDF study guide with the figures and answers. Included in paid plans; reading and quizzes stay free.Sign in

What you will learn

  • Compare simple, block and stratified randomisation

  • Distinguish intention-to-treat from per-protocol analysis

  • Describe the statistical analysis plan and why it is written before unblinding

Randomisation and a pre-specified analysis plan are what allow a trial to make fair comparisons.

Randomisation methods

  • Simple randomisation: each participant is allocated by chance, like a coin toss. Easy but can produce unequal group sizes in small trials.
  • Block randomisation: allocation in blocks (for example, of 4 or 6) so that group sizes stay balanced over time. Each block of 4 in a 1:1 trial contains two of each arm, in random order. Block sizes can be varied to make the sequence harder to predict.
  • Stratified randomisation: separate randomisation lists for important prognostic factors (for example, site or severity) so groups are balanced on those factors.
  • Cluster randomisation: groups (villages, clinics) are randomised rather than individuals.

Randomisation is carried out by an independent system (central web or telephone system, or sealed opaque numbered envelopes). Allocation concealment keeps the next assignment hidden from those enrolling participants, protecting against selection bias.

Blinding

  • Open label: everyone knows the allocation.
  • Single blind: participants do not know.
  • Double blind: participants and investigators (and assessors) do not know.

Blinding reduces bias in how participants report symptoms and how outcomes are assessed. It is not always possible (for example, in a surgical trial), and in such cases objective outcomes and blinded assessors help.

SimpleBlockStratifiedPure chanceCan be unbalanced in smalltrialsEasyBalanced after each blockVary block sizesCommon in trialsBalances key factorsSeparate lists per stratumNeeds few strata
Three common randomisation approaches.

Analysis populations

  • Intention-to-treat (ITT): all randomised participants are analysed in the group they were assigned to, whatever happened afterwards. This preserves the benefit of randomisation and is usually the primary analysis for superiority trials.
  • Per-protocol (PP): only participants who followed the protocol adequately are analysed. It estimates the effect of taking treatment as intended but may be biased.
  • Safety population: all who received at least one dose, analysed by treatment received.

The statistical analysis plan (SAP)

The SAP details every analysis: populations, handling of missing data, models, adjustments, subgroups, interim analyses and tables, figures and listings. It is finalised and signed before unblinding so the analysis cannot be steered by the results. Deviations from the SAP are documented and justified.

Worked example: block randomisation of 12 participants

A 1:1 trial uses blocks of 4, with arms A and B.

  • Block 1: A B B A
  • Block 2: B A A B
  • Block 3: A A B B

After 12 participants, each arm has exactly 6. If recruitment stops after 10, the arms differ by at most 2 (here 5 and 5, or 6 and 4). With simple randomisation, 12 participants could give 8 and 4 by chance. Blocking keeps the groups comparable during recruitment. Because a fixed block size of 4 makes the final allocation in each block predictable, many trials use random block sizes of 4 and 6.

0246Participants2Block 1: A2Block 1: B6After 3 blocks: A6After 3 blocks: B
Block randomisation keeps the arms balanced after every block.

Analyse as randomised

ITT keeps the groups comparable. Removing participants who stopped treatment can make the groups different in ways the randomisation was meant to prevent.

Do not peek and then change the analysis

Choosing outcomes, subgroups or tests after seeing the data increases false positive findings. If an unplanned analysis is done, label it as exploratory.

Check yourself

Which analysis population would you use to estimate the real-world effect of offering a treatment, and which would estimate the effect of taking it as intended?

Summary

  • Block and stratified randomisation keep groups balanced; allocation concealment protects against selection bias.

  • Blinding reduces bias from participants and assessors.

  • ITT analyses participants as randomised; per-protocol analyses only adherent ones.

  • The SAP is signed before unblinding.

Check your understanding

Question 1
1

Allocation concealment hides:

Question 2
2

Intention-to-treat analyses participants:

Question 3
3

In block randomisation with blocks of 4 and 1:1, each block has:

Question 4
4

The statistical analysis plan should be finalised:

Question 5
5

Double blind means:

Question 6
6

Per-protocol analysis may be biased because:

Draft study guide. Spotted an error or something unclear? Tell us at hello@scintria.com. This is educational material, not clinical or regulatory advice.

Clinical data managementSample size, interim analysis and reporting results