7.3 Information bias

Errors in measuring a binary exposure or disease were studied in Section 2.4. Nondifferential misclassification of a binary exposure biases the odds ratio towards \(1\), while differential misclassification can bias it in either direction. Its common sources are:

• recall bias, when cases remember or report past exposure differently from controls;
• interviewer bias, when an interviewer who knows the subject’s disease status probes harder for exposure in cases;
• detection bias, when exposure leads to closer medical surveillance and so to earlier or more complete diagnosis of the disease (women taking hormones see doctors more often, so more of their slow-growing cancers are found).

The defences are in the data collection: exposure measured from records made before the disease, interviewers kept unaware of case status and of the hypothesis, the same procedures for every subject, and objective measures where they exist.

Questions on this section

Stuck on something here? Ask below and it stays attached to this topic.