Question 3

College answer

Not available.

Discussion

This is a hugely positive move into the direction of fair SAQ stem wording and explicit instructions for the candidates. With a stem like this, it is unlikely that anybody would ever score poorly just because they failed to guess the exact structure the examiners had expected. 

The best possible reference for this ended up being Suresh (2011) from the Journal of human reproductive sciences, as it contains a simple explanation of these exact types of randomisation, and in the exact same order. It was much harder to find solid literature on the disadvantages of adaptive design, perhaps because most authors are more preoccupied with its (admittedly, impressive) positive features; but Scott et al (2002) have some excellent critique.

Contrary to the examiners' wishes, the following answer is a simple unordered list, because the 800-pixel width of Deranged Physiology does not lend itself well to tables with many columns. Thus:

Simple randomisation

• Randomization based on a single sequence of random assignments
• Advantages: simple and easy to implement; can be expected to produce even numbers of participants in the groups of a large trial
• Disadvantages: with small sample sizes, can result in unequal groups

Block randomisation

• Randomisation based on small balanced blocks of predetermined group assignments (but the blocks are randomly assigned). 
• Advantages: protected against unequal group sizes in small trials; achieves balance in sample size
• Disadvantages: groups may be generated that are not comparable in important covariates, which can introduce bias and decrease the power

Stratified randomisation

• Block randomisation based on separate blocks for each combination of covariates, with simple randomisation performed within each block
• Advantages: controls for the possible influence of covariates, eg. age or comorbidities, ensuring that groups have balance
• Disadvantages: becomes complicated to implement if many covariates must be controlled; usually limited to 23 variables; works only when all subjects have known baseline characteristics at the time of randomisation

Covariate adaptive randomisation

• Randomisation of each new trial participant according to specific covariates and the assignments of previously randomised participants
• Uses the method of minimisation of an imbalance function, allocating patients more and more deterministically the more the groups become unbalanced
• Advantages: ensures good balance of covariates between groups, flexible, efficient (a smaller sample size may be possible), ethical advantage (assigning fewer patients to treatment arms with inferior outcomes)
• Disadvantages:
  • increases complexity which can harm recruitment;
  • treatment assignments may be predicted with certainty in some situations (i.e it's not really "randomisation"), which could lead to selection bias
  • but then statistical analysis methods usually treat the data as if there was randomisation, which may not be a valid strategy
  • participants' characteristics still need to be well identified before they are enrolled, just as in stratified randomisation;
  • efficiency effect may be negligible in large trials;