5.1.2 Information bias

Information bias is the other type of systematic bias which can be present in AMR/AMU studies and surveillance. Information bias occurs when there is systematic error in the techniques used to measure or record data – this could be for exposures/risk factors, outcome of interest (usually disease state) or extraneous variables (potential confounders). There are two broad categories of information bias: measurement bias and misclassification bias.

Measurement bias describes errors in measurement of a continuous variable such as weight. For example, if the scales used are not working correctly, recorded weights will be wrong.

Misclassification bias results from error in assigning the correct category to a categorical variable. It can result from imperfect diagnostic tests (you may have heard of ‘false positive’ or ‘false negative’ test results), from errors in memory, or from errors in observing and recording data. Misclassification bias can be further categorised as differential or non- differential.

• Differential misclassification is where the probability of an exposure being misclassified depends on outcome status. For example, a patient is more likely to remember whether they previously stopped taking antibiotics halfway through the prescribed course, if they are diagnosed with a resistant infection. This is known as ‘recall bias’.
• Non-differential misclassification is when the probability of misclassification does not depend on the outcome status. For example, the chance of the species of a fish being wrongly identified is the same regardless of whether they carry resistant pathogens or not.