BRS - Avoiding Bias

BRS - How Do I Avoid Bias?

Selection Bias

It's important to remember that in most studies, you are comparing two groups to see if the intervention or exposure results in different outcomes between the groups. Ideally, then, you want to two groups to look as SIMILAR as possible with the exception of the exposure or intervention of interest. Selection bias happens when the groups also differ in some characteristic that would confound the results, and is a threat to the validity of the study results.

So, in case-control studies, you'll pick people with the outcome, and MATCH them to people without the outcome. You may match them according to sex, age, co-morbid diseases and risk factors, etc.

For cohort studies (retropective and prospective) you'll start with a group of people with the exposure of interest and a group without - you'll want to make sure you selet the cohorts so that they are similar, except for the exposure.

For randomized controlled trials, the act of randomization is the best way to ensure that the groups are similar, or that if they differ, that difference happens by chance. If there is a difference between the groups that occurs by chance, it is possible to correct for that difference using multivariable statistics (see RESULTS).

Recall Bias

In retrospective studies, especially case-control studies, researchers frequently must ask the subjects about possible exposures that happened in the past. Recall bias is the tendency for those with the outcome to remember an exposure (especially if it's a generally known or suspected association) and those without the outcome to fail to remember one. Having an existing set of data to examine for exposure would be ideal, but often such studies are left with the possibility of recall bias.

Blinding/Masking

Only for CONTROLLED TRIALS - Once the subjects are in the two arms of the study, it's important to ensure that they are treated similarly during the course of the study. For retrospective studies, there's nothing you can do to guard against unequal treatment other than to look for it. In prospective studies, safeguards against unequal treatment should be in place. Most notably, in intervention trials (especially of drug therapy), placebos are used (that are matched to the active drug for taste, shape, color, etc.) to ensure that the subjects have the same experience of getting the intervention. Double blinding is when both the subject and the investigators caring for the subjects are unaware of the intervention assignment. Single blinding is (usually) when just the patient is blinded to the intervention. For a comparison of dissimilar interventions (a pill and an injection, for instance) a "double dummy" approach is used - the subjects get both a pill and a shot, but the groups differ in which of the interventions is the placebo.

Outcome Assessment Bias

The final major issue to consider is to ensure that the investigators who are adjudicating outcomes that might be subjective - wound healing, degree of improvement, final diagnosis of a syndrome, etc. - remain unaware of the intervention status of the subjects. Otherwise, these outcome assessors may decide differently on the presence of the outcome based on the knowledge of the intervention status.

There are lots of other biases out there, but these are the most important for general use. All of the above affect the INTERNAL VALIDITY of the study - the degree to which you can tell the true effect of the exposure/intervention on the subjects in the study.