Spectrum bias²  exists when the population under investigation does not reflect the general population or the clinically relevant population. It is also termed case-mix bias.  Spectrum bias may account for differences in reported results between investigations.   Be particularly careful using data generated from tertiary referral centers, early in the development of the test, where Referral Bias may have effected the case-mix.  The spectrum of disease prevalence and severity seen at the tertiary hospital  to be different from that at a community hospital or clinic, and this will effect test performance
 

If the prevalence of disease is different from your population, use caution when extrapolating predictive values from the study population. A high prevalence of disease increases the positive predictive values relative to application of the same test on a low prevalence population. The opposite occurs if disease prevalence is low.
 

If the severity of disease differs,  it may influence results for test sensitivity. In general severe disease or abnormalities are easier to detect.  If at all possible examine studies of the test which have been applied to multiple populations.
Consider the following example
 

For more on bias see articles in Bandolier 27 and Bandolier 26