Democritus said that everything that exists in this world is the result of chance and necessity. And Monod, who thought the same, made use of the way chance interweaves with our destiny to explain that we are no more than genetic machines. But today we are not going to talk about chance and its need to understand our mechanistic evolution, but about something quite different, although it may seems a crosswords puzzle: the need to use chance when designing scientific studies to control what is beyond our control.
And, indeed, randomization is one of the key elements of experimental studies. Whenever we’re planning a clinical trial to test the effectiveness of an intervention we need that the two groups, the intervention and the control group, are fully comparable, as it’s the only way to be reasonably sure that the differences we observe are the result of the intervention. Well, this allocation of participants to one of the two groups must be done randomly, without the intervention of the participant’s or the researcher’s will.
The great advantage of randomization is that it evenly distributes the variable that can influence the outcome, whether they are known or unknown to the researcher. Thus, we can state our null and alternative hypothesis and calculate the probability that observed differences are due to chance or to the effect of the intervention under study.
However, all of its advantages may be lost if we don’t randomize correctly. It is very important that randomization sequence is concealed and unpredictable, so it is impossible to know which group the next participant is going to be allocated to, even before deciding his inclusion in the study (to avoid that this knowledge can influence the decision to participate in the study).
It’s often performed by using sealed envelopes with hidden codes that are assigned to participants. Another possibility is to use hardware random number generators or random number tables. For the sake of security, it’s also desirable that randomization is made by people other than the study’s, in a centralized way or by telephone. In any case, we must avoid techniques that can be predictable, as the use of the days of the week, the name’s initials, birth dates, etc.
Techniques for randomization
There are several techniques for properly randomize, all having in common the fact that participants have a certain probability of being allocated to any of the test groups.
A very simple method is to allocate them alternately and systematically to one group or the other, by this method is only random for the first participant allocated. This is why it is often preferred to use other techniques of randomization.
The simplest way of randomized is called (no surprise) simple random allocation. It’s equivalent to toss a coin, having all the participants the same probability to be allocated to any of the two groups. But this is not always the case, because we can change probabilities and assign a different one to control and intervention groups. The problem with this method is that it creates groups of different size, so it may appear imbalances between groups, especially with small samples.
To avoid this problem we can do a block randomization, performing allocation to blocks of predetermined size (multiple of two) and assigning half of the participants to one group and the rest to the other. This ensures a similar number of participants in each group.
We can also divide the sample in groups based on a prognostic variable and make a random allocation within each group. This technique is called stratified randomization. It is important that strata are mutually exclusionary, as much different as possible from each other and as homogeneous as possible inside the strata. Some people recommend using block randomization within each stratum, but this may depend on the type of study.
Participants can also be allocated based on different functional or geographical groups to avoid contamination of some participants with the intervention of the opposite arm of the study. Let’s think that we want to test a cancer screening technique. It may be better to screen in some centers and not to screen in the others. If we do both arms at the same center, the control group participants can modify their lifestyle or require the benefit of the screening for them too.
Finally, there’re also a number of adaptive randomization techniques that changes throughout the study to adapt to emerging imbalances in the distribution of variables or in the number of subjects in each group. These techniques can also be used when we are interested in minimizing the number of those receiving the less effective intervention, once we know some of the results of the study.
And we’re concluding this topic. Before ending I only want to warn you not to mistake concealed randomization sequence with masking. Randomization prevent selection bias and ensure (although not always) a balanced distribution of confounder and effect modifiers. Masking is done after allocation has taken place and prevents information bias. But that’s another story…