Selection Bias in Randomised Trials

There are a number of types of bias that can appear in randomised trials, with selection bias being just one of these.

What is Selection Bias?

In trials, individuals or groups are selected for analysis. If proper randomisation does not take place at the selection stage, selection bias can occur. This could be due to the individuals or groups chosen not being truly representative of the intended treatment population.

Selection bias can occur for a number of reasons. Recruiters to the study may not be fully blinded to the intended treatment or outcome – this could lead them to try and choose individuals based on what the recruiter thinks the next treatment tested will be. Similarly, when individuals or groups are not enrolled in one cohort, there can be more selection bias of those recruited later.

Selection bias is a huge problem for clinical trials as it distorts the true results from a trial. Ineffective treatments may be deemed to be more helpful than they truly are, which could be dangerous to patients who are prescribed the treatments.

How to Overcome Selection Bias

Blinded Trials – Blinding Recruiters

An obvious way to reduce selection bias is to ensure that recruiters do not have access to information about the treatments that a patient would be allocated to. By using allocation concealment for example, neither patients nor recruiters know which group the patient will be allocated into until they are allocated to it.

Blinded trials also ensure that recruiters are unable to see which treatment method a patient has been assigned to once they are entered into the trial. This can also be achieved by having recruiters who are not a part of clinical treatment or patient care assigning patients to groups.

Complete Randomisation

The simplest method of randomisation, complete randomisation is also the most effective way to avoid selection bias. Each patient has an equal chance of being allocated to a certain group. Groups do not need to be of equal size, and no stratification takes place through age or gender of the participants to make them similar.

As group sizes do not need to be of equal size, simple randomisation is rarely used in practice due to the unequal patient numbers. A lack of equal numbers in each group does not necessarily have a large effect on the trial if the sample sizes are sufficiently large enough however.

Restricted Randomisation

While not a perfect method for clinical trials due to the ability for recruiters to guess treatment allocations, restricted randomisation does have benefits. It is also most commonly used for clinical trials. Each group in a restricted randomised trial needs to have some similarities, whether that is how many people are in each group, age ranges of participants or some other defining characteristics.

If a trial is being run with multiple recruitment sites there can be two methods used to ensure that patients allocated to each different treatment group are the same.

Stratification by recruitment site

This is where the number of patients recruited to each different treatment group at each site is the same. This should be avoided where possible as it can be easy to work out which treatment a subsequent patient may be allocated to. With multiple recruitment sites, by not stratifying at the recruitment site, recruiters would need to also know how other sites have allocated their patients, which is unlikely. If there are imbalances between sites, this can be accounted for in analysis by adjusting for site-effects, generally assuring that stratification by recruitment site will not be needed.

If a study has used stratification by recruitment sites as part of its restricted randomisation method, unrestricted or complete randomisation can be used at each individual site. Indeed, when a study only has one recruitment site, any restriction of the number of patients able to be placed in each group acts as if it has been stratified by site.

Random block sizes

Random block sizes instead of permuted block sizes, should ideally be used. By varying how many patients are allocated into a group, and by having large group numbers, this reduces the probability of selection bias taking place.

Prognostic covariates

A patient in a trial comes with surrounding information; this can be information such as their age, gender, and severity of condition. By using this information as part of the stratification factors, recruiters who guess the next allocation of treatment will still not know how other patients with similar prognostic covariates were assigned. Using additional stratification factors can be useful when randomisation has been restricted, however it does not remove all bias from the study.

While avoiding bias can be challenging, ensuring that you have good experimental design is key to start off your trials in the best way.

Other Types of Bias

Our follow-up post on other types of bias in trials is also live. Further bias in trials looks into detection, attrition, and reporting biases, and ways to avoid them in your research.

Good Experimental Design

Fios can help at any stage of your research, helping you plan your studies to avoid common biases and ensuring you can get the most information out of the data generated.

For more information about how we can help, get in contact today.