Overview of selection bias
Selection bias occurs when there are systematic differences between subsets of participants included and/or analyzed in a study such that the subset is not representative of the target population investigated in the trial.
Consequently, an accurate representation of the results may not be discovered due to a lack of including and/or analyzing the true target population.
How to detect selection bias
In clinical research, selection bias occurs if eligible participants are excluded from a sample population but should not have been, if the initial follow-up time of some subjects is excluded, or if some outcome events related to both the intervention and outcome are excluded.
Consequently, there may be systematic differences between participants in the study groups that prejudice the results in a particular direction and deviate their estimate away from the true value, leading to biased and misleading conclusions in clinical research.
Selection biases can affect the internal or external validity of a study. Internal validity relates to whether the study design and conduct was appropriate and free from bias. Selection bias that affects the internal validity of a trial is the most serious. It means that the participants included in a trial were not drawn from the same, or representative, populations.
As a result, the study is confounded since now the concern is that the study is no longer just comparing different types of exposure/risk but also comparing different groups of people.
Thus, there is no way to know if the results are due to the differences in the populations or exposure/risk factors. Selection bias that affects external validity concerns how generalizable the results of the study are to participants outside of the study. For example, individuals with comorbidities may be excluded from studies regardless of their exposure/risk factor, but those comorbidities are still likely to exist in the target population.
Types of selection bias
The types of selection bias and their significant differences are as follows.
1. Sampling bias
Sampling bias is when participants from a specified population are more likely to be selected to participate in a trial than others. Similarly, some subjects that were not included in a study - but should have been based on the inclusion criteria - may have been excluded.
2. Allocation bias
In a randomized controlled trial (RCT), allocation bias happens when the method used to assign subjects to the study groups is inadequate and produces systematic differences between the participants in the study groups. As a result, this type of bias can make the baseline characteristics between subjects in one group significantly different than those in the comparator group(s) in randomized trials.
3. Loss to follow-up and attrition bias
Loss to follow-up occurs in observational trials when a proportion of participants did not begin the trial at the same time as others and thus have not been followed from the start of the trial. Bias is introduced because it is not appropriate to assume that the effect of the intervention will remain constant overtime.
So, if some subjects begin the trial after others, the loss of follow time may result in an under- or overestimate of the effect of the exposure or intervention on the outcome. Similarly, attrition bias can occur in RCTs if participants systematically drop out of the study for reasons related to the intervention compared to those that remain in the trial.
4. Recall bias
Recall bias is when study subjects cannot accurately remember vital information. They either provide incomplete information or information that is just flat wrong. This type of selection bias can lead to missing data.
5. Volunteer bias (or self selection bias)
Sometimes called self selection bias, volunteer bias occurs when willing participants in your study do not represent your research population.
6. Exclusion bias
A type of selection bias that resembles non response sampling bias, exclusion bias occurs when researchers remove a specific subgroup from the research population.
How to avoid selection bias
In all forms of selection bias, the systematic differences that exist between participants limit the ability to equally compare the groups and arrive at unbiased conclusions.
Sampling bias can be prevented by carefully defining the target population and randomly sampling subjects from the population to ensure that all eligible participants have an equal chance of being included in the study as a sample population.
You can minimize selection bias - or any potential bias - by ensuring that the design and conduct of the study is appropriate. In observational studies, subjects in each group should be as comparable as possible and selected from the same location so that the groups are representative of the same population. Additionally, participants should begin follow-up at the same time and efforts should be made to minimize loss to follow-up.
In RCTs, a can be minimized by ensuring that an adequate randomization process is used. Ensuring that subjects are truly randomized maximizes the chances that baseline characteristics between participants in each group are as similar as possible.
The systematic error of attrition bias can be reduced by including all participants that drop out of the trial in the results using an intention-to-treat analysis method in which data are imputed for the subjects that drop out of the trial to help maintain comparability between groups.