Spectrum bias is introduced when you apply an assay in different settings. In the examples in the previous page, the PPV changed dramatically (1.54% -> 46%) based on the pretest probability in the populations we were testing. Studies designed to identify cutoffs for various diagnostic tests can run into issues where the selection of patients skews what is considered “normal” based on the number of diseased patients included and whether a case-control design was utilized (especially for rare conditions). Here is what the prevalence of a disease can do to the characteristics of a test:
In a population with a larger proportion of disease | In a population with a smaller proportion of disease |
Sensitivity = static! Specificity = static! PPV = increased NPV = decreased | Sensitivity = static! Specificity = static! PPV = decreased NPV = increased |