RANDOMIZATION aims to avoid systematic error (BIAS) due to the imbalance in CONFOUNDING factors between INTERVENTION and COMPARATOR groups in a clinical trial. While it would be possible to record multiple patient characteristics (such as age, sex, diabetes status, etc) and then divide patients into approximately equal groups, this method could only be applied once all participants have been recruited and, more importantly, it does not take into account unmeasured or unknown factors which might still affect the outcome of the trial. Moreover, if the treating physician or researcher is able to decide allocation, selection bias might compromise the results of a study. Randomization solves these problems and has become the cornerstone of modern clinical research.
Proper randomization relies on computer-generated random numbers. The use of naturally occurring patterns – such as allocating patients dialyzing on the morning shift to intervention and those dialyzing on the afternoon shift to comparator – is not randomization. Although it may appear that patients fall into these natural groups ‘at random’, it cannot be guaranteed that confounding factors will be evenly spread between such groups.
Randomization methods can be divided into two categories: fixed allocation randomization and adaptive randomization. Most trials use fixed randomization, in which participants are allocated to intervention or comparator with a ‘fixed’ probability that does not change through the study. Fixed randomization can be further divided into simple, blocked and stratified randomization. Adaptive randomization is more complex and permits the allocation probability to change as the study progresses. Below you will find a brief description of these methods.
A) Simple Randomization
Simple randomization uses a random digit table (generated by a computer) or can even be done by flipping a coin. Because the allocation of each participant is completely independent, it is impossible to guess which group a participant will be allocated to before performing the randomization – the essential feature of ALLOCATION CONCEALMENT. However, for trials with a small sample size this method can often result in imbalanced groups – simply by chance.
B) Blocked Randomization
In this method, the principal investigator defines block sizes so that randomization will occur in blocks of a fixed number. This method has the potential advantage of decreasing the likelihood that at the end of the study there will be differences in the number of subjects across groups of treatment (imbalances). However, a potential disadvantage to this method is that investigators might guess the treatment allocation at the end of the block. This could result in selection bias since a researcher might consider not randomizing the next patient if certain the next patient will be allocated to a specific arm of the study.
C) Stratified Randomization
In this method, patients are randomized in strata of covariates considered to play a role in the outcome of study (e.g age, CKD stage). This method reduces the risk of chance imbalance in important covariates since only after a patient is assigned to each stratum will they be randomized into the active or the placebo groups. As it is more complex, it can lead to bias if used with simple randomization techniques. Since the participant cohort is further divided into strata, the issue of imbalances for small sample sizes can be hard to overcome.
A randomization system that varies depending on the characteristics of participants already randomized is known as adaptive.
A) Minimization Technique: Randomizing Patients According to an Algorithm
The minimization technique applies an algorithm to change the likelihood that a given patient will receive active or placebo treatment according to prespecified baseline characteristics. It is essentially a more complex form of stratified randomization that continuously adjusts probabilities to maintain balance throughout the recruitment period.
B) Adaptive Randomization According to the Response
In this method, every new patient is randomized according to the response of the previous patients. As an example, if the first patient is randomized to the active treatment and responds the subsequent patients will be also randomized to the active treatment until no response is achieved. When this happens, the next patient will then be switched to the control treatment.
Be aware that while minimization and response adaptive randomization are interesting methods, they are complex to implement and require sophisticated statistical support
To prevent selection bias in patient recruitment, it is important that the researcher or clinician choosing to enroll the patient does not know (or cannot guess) what the patient’s likely treatment allocation will be – a process known as ALLOCATION CONCEALMENT. There are many ways to achieve this, but common techniques include assigning the randomization schedule to specific individual(s) (separate to the researchers enrolling patients) and the use of opaque envelopes or secure online systems.
- Rethinking Clinical Trials [see Design/Experimental Designs and Randomization Schemes]
- PennState STAT509: Design and Analysis of Clinical Trials
- EMA Statistical Principles for Clinical Trials (see Section V)
- GraphPad [includes random assignment generator]
- Vickers AJ. How to randomize. J Soc Integr Oncol. 2006; 4(4): 194–198 [PMID 17022927]
- Field Trials of Health Interventions [see Chapter 11: Randomization, blinding and coding]