Referral bias (admission rate bias) refers to a situation where the chance of exposed cases being admitted to the study is different to exposed controls. This happens frequently when cases are selected in a hospital whose activity is linked to the studied exposure. The admission rate bias may be due to a number of factors e.g. access to care, popularity of certain hospitals/ doctors etc [2].

In hospital-based studies, if the admission rates to hospital differ for different disease / exposure groups (e.g. admission rates of exposed and unexposed cases and controls differ), the association between exposure and disease will be distorted, and the relative odds of exposure to the putative cause may be spuriously increased or reduced. This bias is also known as Berkson's bias [2].

Example: in a study of risk factors for lung cancer, cases were compared to controls with regard to history of exposure to asbestos. Cases were recruited in the respiratory department of a hospital which is the National Reference centre for asbestosis. Controls were selected in the surgical wards of the same hospital. In that situation, it is likely that lung cancer cases of this respiratory department do not represent other cases with regard to history of asbestos exposure. Here, the selection of cases is linked to exposure. Selected cases are more likely to have been exposed to asbestos (than other lung cancer cases in the population), with an overestimation of 'a', resulting in an overestimation of the odds ratio.