What is bias in epidemiological studies?
"People who agree to take part in epidemiological studies may differ in all sorts of ways from those who don’t. If these differences correlate with either the exposures or the outcomes being studied, the findings are likely to be biased."
Ouch! Time for a quick explanation of bias.
If people in epidemiological studies differ from the general population this tends to affect generalisability (external validity) and rather than cause bias or affect internal validity. For there to be bias the likelihood of exposure must correlate with the likelihood of the outcome. Think of a simple example. Doctors get less lung cancer than the general population. In the British Doctors Study, doctors who smoked were more likely to get lung cancer. The finding is not biased because there is no link between the tendency of doctors to smoke and the tendency of doctors to get lung cancer.
What might bias look like? Imagine we found a link between smoking and liver diseasee in the British Doctors Study. We can quickly see that there might be a link between smoking (exposure) and alcohol consumption which would lead to an association with lung cancer (outcome). If the tendency for alcohol consumption and smoking to correlate in individuals were unique to doctors we could see how studying doctors might lead to bias. But we can also see that this type of situation is unusual.
In fact this is exactly what the authors of the cited paper reported "Participating women were generally healthier, of higher social status, and older than the baseline cohort. However, selection bias in the chosen scenarios was limited..."
Ref 1. Bliddal M, Liew Z, Pottegård A, Kirkegaard H, Olsen J, Nohr EA. Examining Nonparticipation in the Maternal Follow-up Within the Danish National Birth Cohort. Am J Epidemiol. 2018 Jul 1;187(7):1511-1519. doi: 10.1093/aje/kwy002.
Competing interests: No competing interests