Topic 5: Randomisation Flashcards
What does random allocation mean exactly
That every patient has a known chance of being randomised to each treatment, but the treatment given cannot be predicted. So you control the probabilities of receiving each treatment, but have no opportunity to predict which treatment is allocated next
What are the two components of random allocation
Sequence allocation and allocation concealment.
What kind of bias does random allocation avoid
Selection
What does random allocation aim to achieve
Baseline comparability with respect to known and unknown factors
Which part of analysis does random allocation justify
Using statistical tests to identify differences in outcome between arms, as random samples is the basis of statistical inferences.
Why do we need baseline comparibility
To differentiate which treatment effects have occurred by chance and which have occurred from the treatment itself.
Give an example of a non-random allocation
Using birth days or date of enrolment.
What is allocation concealment
Concealing treatment allocation until the patient has entered the trial, from patients, care-providers and research staff
What is the difference between blinding and allocation concealment
Blinding is concealment after a patient has been entered into the trial. Treatment allocation is concealing it before and until they’re entered into the trial.
How is allocation concealment obtained
Allocation is generated and used in a way that those conducting randomisations and recruiting patients cannot predict what the next allocation will be.
Should allocation concealment be used in every single clinical trial
Yes
What are two main methods of allocation concealment
Person generates allocation sequence independently of the people recruiting the patients - Separating sequence allocation from recruitment. Treatment allocation carried out externally using a central telephone randomisation service.
What are the four types of sequence allocation
Simple randomisation, block randomisation, stratified block randomisation, minimisation.
What is simple randomisation
Each patient is allocated to a treatment with a fixed probability, independent to all previous allocation. Unrestricted randomisation
Give an example of simple randomisation
Flipping a coin, random number table
What do we tend to use for simple randomisation
Computer generated random number algorithm and setting a seed so that it is replicable.
What is the advantage of simple randomisation and when is it used
Simple. Only really used when simplicity is needed, for example in a complex trial that needs to be adapted quickly, and needs a simple form of randomisation to do this.
What are the disadvantages of simple randomisation
At any point there could be imbalance in the number of participants allocated to each treatment. It also doesn’t allow for balance across treatment arms in terms of baseline characteristics.
What is block randomisation
within each block, you space out the treatment arms in the treatment allocation ratio desired. A form of restricted randomisation
How is block randomisation carried out
Write down possible permutations you could make with the treatments, within the block size. Then use simple randomisation to select a sequence, and use this to allocate patients.
What are the possible permutations for treatments A and B and a block size of 4, 1:1 ratio?
AABB, ABAB, ABBA, BBAA, BABA, BAAB
How does block randomisation ensure balance, eg block size of 4
After every 4 patients have been allocated, you have perfect balance. At any stage, the group size differs by no more than half the block size.
What is the issue with a block size too small
Places too much of a restriction on randomisation, if the trial is not blinded, recruiters may be able to predict the next treatment.
What is the issue with a block size too large
The allocation sequence becomes more similar to simple randomisation. Its less predictable but there’s more chance of imbalance.
True/False: The block size is revealed to research team but not the wider trial
False. The block size is never revealed to any of the them.
What are advantages of block randomisation
Easily extendable to more than two arms and to unequal allocation.
When is block randomisation particularly useful
In interim analysis when you want balance midway through the trial.
What are the disadvantages of block randomisation
Can’t balance prognostic factors across arms, potential for person allocating treatments to predict next treatment.
What is random permuted block randomisation
We randomly vary the block size to avoid the potential for selection bias.
What is an advantages of using random permuted block randomisation
Makes allocation far less predictable, regarded as pretty much impossible to crack.
How is random permuted block randomisation done
Done in a loop. An example is 1) generating a random number from a set of possible block sizes 2) then allocating patients in block randomisation for that block size. Then repeat steps 1 and 2 until every participant has been randomised.
What is stratified block randomisation
A form of block randomisation that allows you to balance across prognostic factors
How does stratified block randomisation work
Divide the population into characteristics you want to stratify for and generate a completely separate, independent randomisation list for each of those strata. Within each stratum, some form of block randomisation is used.
How many randomisation lists does stratifying by male, female, 18-30, 30-55, 55+, UK, US and Canada make?
3 strata, sex is at 2 levels, age is 3 levels and region is 3 levels. That makes 2x3x3 = 18 randomisation lists.
What are the disadvantages of stratified block randomisation
You are not likely to finish recruitment and have every single list be completed. It is likely that the population will not be evenly distributed and you can still have the overall imbalance of participants per arm.
What is the potential difference between the number of participants allocated to each arm in stratified block randomisation
Half the block size x the number of lists you have.
How do you deal with the issues in stratified block randomisation
Choose only strata that are important prognostic factors, and can always use post-randomisation stratification on anything you don’t include, like in sensitivity analysis or in main analysis as a stratified variable.
When is stratified block randomisation not recommended
When the number of strata is large in comparison to the sample size, since you might not end up with a reasonable amount of people in each arm.
What is stratification in general used for
Providing balance to enhance the credibility of the results. Improve efficiency of the treatment effect estimates and help interpretation when subgroup effects are expected.
What is minimisation used as an alternative for
Stratified block randomisation
What is minimisation done by
A computer
What is minimisation as a form of sequence allocation
Keep track of the characteristics of patients entered into the trial, and allocation aims to minimise the difference between arms in terms of this characteristic.
Why does minimisation need a random element
Because it is deterministic in nature. If you know the characteristics of the last 10, you know the treatment for the 11th. So random element makes allocation less predictable.
Why is being able to predict the next treatment in the allocation sequence a problem
Possible selection bias
What is a usual random element for minimisation
10-20%
How is minimisation different to stratified block randomisation
We ensure balance for each factor independently. So instead of ensuring a balance in females under 30 and females over 30, we just ensure balance for sex and for age, independently.
How is the first allocation done in minimsation
Simple randomisation
How does minimisation work
When we know the treatment group with the lowest total for that characteristic, they are allocated to that treatment with high probability - determined the random element
Why might unequal allocation be used
To gain information about one treatment over the other, or if one treatment is very expensive, can contain costs
What is the main disadvantages of using unequal allocation
Slight loss of power, or higher sample size needs for equivalent power.