Biology, images, analysis, design...
Use/Abuse Principles How To Related
"It has long been an axiom of mine that the little things are infinitely the most important" (Sherlock Holmes)




Berkson's Bias


Berkson's bias is a type of selection bias. It can arise when the sample is taken not from the general population, but from a subpopulation. It was first recognised in case control studies when both cases and controls are sampled from a hospital rather than from the community.

When we take the sample we have to assume that the chance of admission to hospital for the disease is not affected by the presence or absence of the risk factor for that disease. This may not be the case, especially if the risk factor is another disease. This is because people are more likely to be hospitalized if they have two diseases, rather than only one.

The best known example of this is given by Sackett (1979). He took a random sample of 2784 people from the community, and determined the presence or absence of respiratory disease and locomotor disease. He then looked at the same thing for those people within the sample who had been hospitalized in the previous six months. The results are shown below:

{Fig. 1}

If we only looked at the hospital sample, we would conclude that people with respiratory disease are much more likely to suffer from locomotor disease. In other words that there is an association between the two complaints. Moreover, any analysis of risk factors will (wrongly) suggest that the risk factors for locomotor disease are also risk factors for respiratory disease.

If, however, we look at the full community sample, we would conclude that having respiratory disease has no effect on whether or not one is likely to suffer from locomotor disease. The latter is of course the correct conclusion. The incorrect conclusion from the hospital sample arises because people who have both diseases are more likely to be hospitalized than people who only have one. The same bias is likely to arise if cases and controls are obtained from autopsy samples.