Bias, confounding and chance in epidemiological studies
Evidence-based practice involves the critical appraisal of studies to assess their possible contribution to clinical or public health practice. When appraising studies, there are three main issues that need to be considered for their impact on the results:
The statistical analysis only reports how likely it is that the results have occurred by chance; it is important to recognise that a study can only be used to inform practice if it is well designed, so that the results are credible. If a study is not credible, then the results are of no value.
Bias is a systematic error or flaw in the methodology of the study, which affects the results. There are two main types that we are likely to come across – selection bias and information bias.
- Selection bias is a flaw in the way subjects are selected for the study. Selection bias can occur when subjects selected are not representative of the population about which conclusions need to be drawn.
- Information (or measurement) bias can arise from errors in measuring exposure or outcome appropriately and can take several forms that may be relevant in different studies and settings.
In this chapter, we consider mainly observer and recall bias; publication bias is considered in Chapter 12.
- Observer bias occurs when a researcher measuring the outcome has knowledge of the subject’s exposure to a risk factor or intervention, and this knowledge affects how they assess outcomes. In borderline cases, he/she might be more likely to classify exposed subjects as having the outcome and unexposed subjects as not having the outcome (Figure 6a).
- Recall bias occurs when a subject with the outcome is more likely to remember an exposure or other events than a subject without the outcome of interest.